OM Composer's Book 3 PDF

THE OM COMPOSER’S
BOOK
Volume 3
Collection Musique/Sciences
directed by Jean-Michel Bardez & Moreno Andreatta
The Musique/Sciences series contributes to our understanding of the relationship between

two activities that have shared intimate links since ancient times: musical and scientific
thought. The often-cited Quadrivium (music, astronomy, geometry, and arithmetic)
reminds us that, in an age imbued with the spirit of the Gods, it was not uncommon to
think of these two modes of thought as twins. During the twentieth century, music and
science developed new links, establishing relationships with mathematics and opening
new lines of musical research using information technology. Modeling, in its theoretical,
analytical and compositional aspects, is more than ever at the center of a rich musicologi-
cal debate whose philosophical implications enrich both musical and scientific knowledge.
The pleasure of listening is not diminished when it is more active, more aware of certain
generating ideas—au contraire.
Published works
• Gérard Assayag, Frano̧is Nicolas, Guerino Mazzola (dir.)

Penser la musique avec les mathématiques ? (2006).
• André Riotte, Marcel Mesnage
Formalismes et modèles musicaux (I & II) (2006).
• Carlos Agon, Gérard Assayag, Jean Bresson (Eds.)
The OM Composer’s Book vol. I (2006).
• Franck Jedrzejewski, Mathematical Theory of Music (2006).
• Guerino Mazzola, La vérité du beau dans la musique (2007).
• Moreno Andreatta, Jean-Michel Bardez, John Rahn (dir.)
Autour de la Set Theory. Rencontre musicologique franco-américaine (2008).
• Moreno Andreatta, Jean-Michel Bardez, John Rahn (Eds.)
Around Set Theory. A French/American Musicological Meeting (2008).
• Jean Bresson, Carlos Agon, Gérard Assayag (Eds.)
The OM Composer’s Book vol. II (2008).
• Emmanuelle Rix, Marcel Formosa (dir.)
Vers un sémiotique générale du temps dans les arts (2008).
• Gérard Assayag, Andrew Gerzso (Eds.)
New Computational Paradigms for Computer Music (2009).
• Rozalie Hirs, Bob Gilmore (Eds.)
Contemporary compositional techniques and OpenMusic (2009).
• Moreno Andreatta, François Nicolas, Charles Alunni (dir.)
A la lumière des mathématiques et à l’ombre de la philosophie.
Dix ans de séminaires MaMuPhi (2012).
• Nicolas Viel, La musique et l’axiome.
Création musicale et néo-positivisme au 20e siècle (2014).
THE OM COMPOSER’S
BOOK
Volume 3
Edited by
Jean Bresson, Carlos Agon, Gérard Assayag
Preface by
Roger Dannenberg
Collection Musique/Sciences
Musique/Sciences Editorial Board
Carlos Agon, UMR STMS Ircam/CNRS/UPMC, Paris
Gérard Assayag, UMR STMS Ircam/CNRS/UPMC, Paris
Marc Chemillier, Ecole des Hautes Etudes en Sciences Sociales, Paris
Ian Cross, University of Cambridge
Philippe Depalle, McGill University, Montréal
Xavier Hascher, University of Strasbourg
Alain Poirier, Conservatoire National Supérieur de Musique et de Danse, Paris
Miller Puckette, University of California, San Diego
Hugues Vinet, UMR STMS Ircam/CNRS/UPMC, Paris
Communication
Marine Nicodeau
Editions
Claire Marquet
Cover design
BelleVille
Volume editor
Jean Bresson
Editorial coordination and support
Aaron Einbond
Tous droits de traduction, d’adaptation et de reproduction par tous procédés réservés pour tous pays.
Le code de la propriété intellectuelle du 1er juillet 1992 n’autorise, aux termes de l’article L. 122-5, 2e
et 3e a), d’une part, « que les copies ou reproductions strictement réservées à l’usage du copiste et non
destinées à une utilisation collective » et, d’autre part, « que les analyses et les courtes citations dans
un but d’exemple et d’illustration ». « Toute représentation ou reproduction intégrale ou partielle, faite
sans le consentement de l’auteur ou ayant cause, est illicite » (article L.122-4). Cette représentation
ou reproduction par quelque procédé que ce soit constituerait donc une contrefao̧n sanctionnée par les
articles L. 335-2 et suivants du Code de la propriété intellectuelle.
ISBN 978-2-7521-0283-6

c 2016 by Editions DELATOUR FRANCE/Ircam-Centre Pompidou
www.editions-delatour.com
www.ircam.fr
Contents
Preface
Roger Dannenberg vii
Introduction
Jean Bresson, Carlos Agon, Gérard Assayag 1
Variazioni su AlDo ClEmenti

Michele Zaccagnini 3
Folk material transformations

and elaborations in A Vida é Nossa
Gonçalo Gato 15
Tri Bhuwana for 12 voices and Balinese gamelan:

Three worlds/tripartite world
Philippe Boivin 37
Programming modular progressions in OpenMusic

Matthew Lane 55
Musicalising sonification: Image-to-music

conversions using OpenMusic
Luiz Castelões 77
On “slow” computer-aided composition

Julien Vincenot 93
OM-Darwin: Generative and descriptive

aspects of genetic algorithms
Geof Holbrook 111
Computer-aided composition in the creation

of As I ride the late night freeways
Matthew Schumaker 121
Materials and techniques in D’improvviso

da immobile s’illumina for bass clarinet,
two orchestras, piano, and percussion
Federico Bonacossa 135
Musique instrumentale concrète: Timbral transcription

in What the Blind See and Without Words
Aaron Einbond 155
Recomposing Beethoven with Music Neurotechnology
Eduardo Reck Miranda and Anders Vinjar 173
Composing for the resonance: Finding new relationships

between architecture and musical composition
Ambrose Field 193
Sketching, synthesis, spatialisation:

The integrated workflow of Cognitive Consonance
Christopher Trapani 205
Germination
Jean-Luc Hervé and Serge Lemouton 219
Dialogue with OpenMusic in the process of composing

Nothing that is not there and the Nothing that is
Takéshi Tsuchiya 229
Koch’s Space
Julián Ávila 245
Electronic sound creation in Balænoptera for

bass clarinet, electronic sounds, and live electronics
Fabio De Sanctis De Benedictis 259
Modelling a gesture: Tak-Sı̄m

for string quartet and live electronics
Alireza Farhang 275
Electronic dramaturgy and

computer-aided composition in Re Orso
Marco Stroppa and Jean Bresson 301
Rima Flow: Oral tradition and composition

Alessandro Ratoci 321
Ab-Tasten: Atomic sound modelling

with a computer-controlled grand piano
Marlon Schumacher 341
Appendix
OpenMusic 363
Computer music tools and technology 373
About the authors 377
vi
Preface
Computer-assisted composition and algorithmic composition are the modern realisations

of a very natural and ancient musical imperative. While computers are a recent phe-
nomenon, formal systems, symbolic manipulation, and even algorithms have been a
part of music composition and musical thinking since ancient times. Computers are
now impacting art, science, and culture in so many ways that it is tempting to refer
to the collective change as simply “the computing revolution”. However, various new
directions and practices can be parsed out and assessed independently. One of these
directions in music, and the main theme of this book, is the use of computers to create
and explore musical structures and formalisms, using the capabilities of computers—
precision, vast memory, and speed—to complement our human ability to pose questions,
invent structures, and evaluate musical artifacts.
Music, form, organisation, and structure are linked in many ways, some subtle and
some direct. Music is not representational; by and large, it does not denote objects or
actions, as do traditional painting and literature. Perhaps this lack of representational
meaning leads to a focus inward. If music does not look outward by describing and
referencing, we are invited to look inward toward music itself as the main object of
interest. Composers have always been as concerned about form as about sound, and
thinking about form leads naturally to abstract representations and concepts. Music,
of all the arts, is perhaps the most formalised and most closely connected to symbolic
representations. In “representations” we include systems of music notation found in
many cultures as well as bodies of music concepts and theory. Long before any notion
of computers, composers saw that their representations provided guidance to construct
music. Guido d’Arezzo’s famous mapping of vowels to pitches, canonical writings of
Palestrina, isorhythm in motets of Vitry and Machaut, eighteenth-century musical dice
games, and Cage’s elaborate instructions for “chance” music are all examples of musical
representations and theory being used in a generative manner to produce music.
Formal systems for music generation are an integral part of music as a discipline of
study and research. Taking for granted that music is important, we try to understand
music and we devise formalisms that capture what we observe. What better way to
test and refine these formalisms than to use them for music generation? The output
reveals the strength and limitations of the theory. For example, every researcher who
writes a computer program embodying rules of traditional harmony soon learns that
traditional harmony deals with chord-to-chord voicing but says nothing about tension,
development, variety, surprise, or when to break the rules. Naïve programs generate
trivial harmonisations (e.g. I-V-I-V-I) and soon challenge researchers to find new rules
and new models that give a more complete picture of harmony.
To the extent that music theory has always tried to distill music to formal descriptions,
constraints and procedures, we can say that computer-assisted composition (CAC) has
been a goal of musical study for centuries. It seems that musicians have been ready all this
vii
time with rich formal symbol systems, just waiting for the right tools to enable further
exploration. This is illustrated by the work of Hiller and Isaacson, who were already
creating the first music composition programs at a time when each new computer was
considered a historic development. In this sense, CAC might be considered a traditional
practice of composition and music theory that has simply made slow progress until
recently due to its dependence on symbolic computation and ultimately automation
enabled by modern computers.
Sixty years after Hiller and Isaacson, computers and CAC have not replaced the
composer, even in situations where one might think music is highly constrained and
subject to formalisation. Ironically, CAC seems to be flourishing in experimental music
where things are the least constrained and where there is no over-arching formalism, at
least none that is known and applicable to music composition and production. Perhaps
this is the real strength and contribution of CAC in today’s musical world: CAC offers
composers a new way to construct and explore formalisms that lead to musical structure,
form, and organisation.
It is not that we do not already have such formalisms. Rather, a trend beginning in
the 20th Century has been a sort of meta-composition in which composers first create
new formalisms, new rules, new objectives, new questions, and then work within this new
framework to create new music. Here is where CAC really shines. If new formalisms can
be expressed computationally, then the ramifications of the formalisms can be explored
quickly, and either the formalism can be quickly rejected or it can be refined according
to the musical tastes of the composer. This approach allows composers both to work
faster and to extend their production to forms and sounds that might be unimaginable
through a more direct path of putting pen to paper.
If composition with computers has become meta-composition, do we have a cor-
responding meta-theory? Can we formalise the practice of CAC? There are general
studies of creativity and a few studies of computer music composition practice, but no
specific theories akin to traditional harmony or serialism. We are left at a stage of
development where the best one can do to understand the field is to study individual
cases. While the literature describes many “toolkits” and software techniques, and many
composers mention using algorithmic techniques, it is rare to find in-depth descriptions
of the intentions, decisions, and implementation details of the compositional process.
The chapters of this book, however, offer a wealth of practical and detailed case
studies of CAC in practice. The variety of concepts, strategies, and techniques that
fill these pages illustrates that CAC is not a musical genre or a musical theory. It is
more of a meta-theory, an approach to the construction, exploration, and evaluation of
musical theories and their application to composition. The role of the computer and
formalisms ranges from the most abstract, where the computer generates material to be
freely incorporated into composition, to the most specific, where the computer might
generate the final composition as digital audio. By considering all of these descriptions
as case studies, we can begin to form an overall impression of CAC practice.
This book has the common theme of OpenMusic software, which is both a visual
programming language and a variety of libraries supporting different kinds of musical
computation including selection, audio analysis, constraint satisfaction, genetic algo-
rithms, interpolation, and synthesis techniques. Thus, the chapters serve in some way
as a commentary on OpenMusic itself and the roles that programming languages and
computer music systems can play in assisting composition. One feature of OpenMusic
viii
that seems to be critical to its success is the ability to translate between the worlds of
signals and symbols. Many composers use signal analysis to obtain pitch and rhythmic
material, but the resulting data can be visualised in OpenMusic using fairly conventional
notation. This ability to see intermediate results as “music” seems to play an important
role in the exploration of musical ideas. Another feature of OpenMusic is the availability
of ready-made representations for melodies, chord sequences, and time-varying functions.
These high-level abstractions seem to facilitate the rapid construction of programs to
generate and manipulate music data in meaningful ways. Finally, OpenMusic is able to
render music as audio as well as notation from these musical representations, allowing
composers conveniently to hear as well as see the results of their models and choices.
In conclusion, computer-assisted composition is a recent phenomenon with historical
roots that can be traced back to the earliest theories and representations of music. Now
that computers are ubiquitous and enable composers easily to explore generative theories
of music, we can expect CAC to take an increasingly important role in common musical
practice. The chapters in this book provide a wealth of material on how composers think
about CAC as well as details of their individualised pursuit of new musical structures.
For both composers and non-composers, this collection is a wonderful introduction to an
emerging practice that challenges and expands our understanding of composition.
Roger Dannenberg
ix
Introduction
Ten years after publication of the first OM Composer’s Book, we are happy to present
this new volume comprising 21 contributions of contemporary music composers that
describe original works and experiments carried out using the OpenMusic computer-
aided composition environment.
OpenMusic (OM) is a visual programming language dedicated to musical modelling
and composition. It allows composers to program musical processes combining the
expressive power of a programming language with symbolic and graphical representa-
tions.1 Amongst a variety of existing works on domain-specific programming languages
for music, computer-aided composition was formalised as a specific field of research in
the early nineties.2 Today this field represents an original approach to music composition
using computer processes and programming: in contrast with most computer music en-
vironments that focus on performance and real-time interaction, it promotes an “offline”
paradigm to aid composition of musical structures. A program in OpenMusic typically
generates musical sequences, scores, sounds or any kind of musical material involved
in a compositional process. For that purpose, it brings forward advanced temporal
modelling and integrates rich structures and formalisms in music programming. The
works presented in this book provide evidence of this and illustrate these characteristics
of computer-aided compositional approaches.
Contributions to The OM Composer’s Book often hint at a future where distinctions
between composing and programming using computer-aided composition software be-
come obsolete. Indeed, computer-aided composition environments are not meant to let
computers create music, but to provide musicians with means to have the computer do
what they want; that is, to become programmers themselves.
It is interesting to note, however, that most of the authors in this book underline
a certain emancipation from the strict computational approach which can emanate
from excessive use and trust in computer systems, emphasising the interplay between
algorithmic and “intuitive” processes involved in composition. OpenMusic in some way
facilitates this orientation in compositional processes by merging programming, musical
data editing, and visualisation.
Computer-aided composition evolved along with the developments in technology and
musical æsthetics of recent decades. As the reader will notice in several chapters of
this book, there is a clear tendency to connect compositional processes in OpenMusic
1A presentation of OpenMusic is given in the Appendix of this book.

2 “We conceive such an environment [of computer-aided composition] as a specialised computer language
that composers will use to build their own musical universe. [...] This leads us to reflect on the various
existing programming models, as well as on the interfaces [...] which make it possible to control
this programming, and on the representations [...] of the musical structures, which will be built and
transformed using this programming.” G. Assayag, “Computer-assisted composition today”, 1998.
1
The OM Composer’s Book. 3
with external tools and software—in particular in the field of digital sound processing.3
We have decided to organise the book from this perspective. The first chapters deal
with the traditional “symbolic” core of the computer-aided composition framework, and
later move into “mixed” approaches that utilise more and more sound generation and
processing. Of course many alternative paths could also be taken to read through the
chapters of the book.
We were glad to receive so many contribution for this volume, and are grateful to
all the authors for their participation. We would also like to thank the editorial team
of the Musique/Sciences collection for their support, and in particular Aaron Einbond
whose labours bring this book to its present form. Contributions to the OM Composer’s
Book 3 came from France, the United Kingdom, Italy, Spain, Germany, the United States,
Canada, Japan, Iran, Portugal, and Brazil. The cultural, geographic, and æsthetic
diversity of topics described in this book remind us that computer-aided composition
is a lively and continually widening domain, offering exciting perspectives for musicians
and computer music researchers around the world.
Jean Bresson, Carlos Agon, Gérard Assayag, Editors.
3A glossary of technology and software mentioned in different chapters is also provided in the Appendix
of the book.
2
Michele Zaccagnini
[...] our everyday experience in building things tends to give

us the intuition that creating complexity is somehow difficult,
and requires rules or plans that are themselves complex. But the
pivotal discovery that I made some eighteen years ago is that [...]
such intuition is not even close to correct. [...] what I found—to
my great surprise—was that despite the simplicity of their rules,
[...] even some of the simplest programs that I looked at had
behaviour that was as complex as anything I had ever seen.
Stephen Wolfram [2]
Variazioni su AlDo ClEmenti (2013) is a piece for chamber orchestra in four movements
(“Invenzione”, “Sinfonia”, “Preludio” and “Corale”). It calls for 11 instruments: flute
(doubling piccolo), clarinet (doubling bass clarinet), glockenspiel, piano, 3 violins, 2
violas, cello, and double bass. The piece is based on a four-note set {A, D, C, E} (taken
from Aldo Clementi’s “playable” letters) and is written entirely in OpenMusic. The
processes implemented in the OpenMusic patches are mainly concerned with rhythmic
treatments, while pitch manipulations are fairly simple. The four movements express a
unified æsthetic: the exploration of repetitiveness and of the psychoacoustic boundaries
between predictability and chaos. In this paper, I start by describing the general æsthetic
that informs my compositions; the second section aims at unveiling the algorithms that
produced the first two movements of Variazioni: “Invenzione” and “Sinfonia”; the third
and final section includes conclusions and a description of current implementations of
the algorithms.
A Clementian æsthetic
The misunderstandings are generated by those who think, if subconsciously, of
music as “discourse” inadvertently making it a caricature of an arch that describes
a useless orgasm. Exaltation and depression are closed chapters: no matter how
disguised they might be, they are modest symbols of an extinct dialectic. A forte
followed by a piano, a high note followed by a low one, a “sweet” timbre followed
by a “raw” one are in themselves sonata-like cells of the greater Sonata Form.1
Aldo Clementi (author’s translation)
1 “Gli equivoci nascono solo da quanti, anche inconsciamente, pensano la Musica come discorso e
quindi, non accorgendosene, come caricature di un arco che descrive un inutile orgasmo. Esaltazione
e depressione sono capitoli chiusi: comunque camuffati, sono modesti simboli di una dialettica già
estinta. Un forte seguito da un piano, una nota acuta seguita da una grave, un timbro dolce da uno
aspro: ciò è di per sé dialettica, cellula sonatistica di una più grande Forma-Sonata.” [1], p. 48.
3
Michele Zaccagnini
In 1971, Aldo Clementi composed B.A.C.H., a short piece for solo piano. The one-
page score is supposed to be repeated at least three times; it is a sort of “musical
mechanism” which neither develops nor varies, yet it seems to possess an inner force
that propels it forward in a seeming perpetual motion. B.A.C.H. was a turning point
from which Clementi began deliberately to erase any sort of contrast that could initiate
a dialectic of tension and release. Instead, his compositions became a research of musical
stasis. Radical as this might sound, Clementi realised how avoiding any form of contrasts
was the only antidote to a sort of “narrative bias” that affects our perception of music.2
Clementi’s æsthetic philosophy is accompanied by a rich eschatological explanation that
eventually declares music a doomed art form, not too differently but possibly more
negatively than his early mentor Theodor Adorno.
My encounter with the music and æsthetic thought of Clementi has been an important
moment in my compositional journey. As idiosyncratic as his æsthetic approach might
seem, I believe Clementi’s lesson carries tremendous positive implications. His determina-
tion to abandon compositional escamotage in order to privilege the exploration of frozen
textures has been a fundamental inspiration that has pushed me to pursue my current
compositional path. Like Clementi, I am deeply interested in studying and experimenting
with the construction of static textures—textures that derive their raison d’être solely
from their constructional principles. I explore the perceptual boundary within static
textures that divides regular and irregular, predicable and chaotic. Where does one
begin and the other end? Surely, the two are not mutually exclusive, but how much
“space” divides them? The attention to textural detail rather than to the larger musical
discourse is the very aspect that links Clementi, my work, and OpenMusic.
I can describe my approach to composition as a sort of experimental process articu-
lated in four steps:
1. æsthetic premise
2. procedural hypothesis
3. design of algorithm
4. observation/assessment
In what follows I will describe steps 2 and 3 in detail. Step 4 is left for the listener.
Computer-aided composition has been a crucial addition to my compositional process
since my modus operandi heavily relies on the ability to assess speedily the outcomes
of experiments given certain hypotheses. In general, computer-aided composition allows
a composer to parse a large number of results, thereby allowing for a greater number
of hypotheses to be tested and evaluated. Variazioni su AlDo ClEmenti was written
following this experimental approach: each of the movements is a “chosen” realisation of
a hypothesis aimed at exploring the simple-complex boundary.
2 Toillustrate this point, one might think how even a single contrasting element in an otherwise uniform
texture is enough to trigger an expectation.
4
Algorithmic procedures
Invenzione
“Invenzione”, the first movement of Variazioni, is a canon in which each instrument
plays a fixed set of four notes: the first four instruments enter by playing the set at its
original transposition A, D, C, E in four different octaves (A4, A6, A3, A5); successive
entrances introduce different transpositions that gradually undermine the original set’s
diatonic area. As instruments enter the canon, the sonic texture gradually builds up
to 11 voices; as the movement reaches its midpoint, instruments start dropping out of
the canon causing the texture to gradually thin out. Different from a traditional canon,
generally thought of as a polyphonic structure, “Invenzione” is anything but: as the
texture builds up, the different voices of the canon start blurring together producing an
increasingly amorphous texture. Voices blur with one another as the traditional elements
accounted for by a polyphonic construction do not appear and are deliberately avoided:
instruments play the same pitch-sets, with the same dynamic (always p) and with no
specific timbral treatment (e.g. ordinario); as a result, the simple diatonic fragments
played by different instruments gradually start blending together, ultimately producing
in a monophonic drone.
The minimal construction of the movement in terms of pitch, timbre, and dynamics
is counterbalanced by a fairly complex principle that informs the rhythmic placement
of each note. To determine the rhythmic placement of the four-note melodic fragments
I designed what I later named the “bouncing algorithm”. This algorithm “mimics” the
motion of an object in an enclosed space. The object moves at a given speed along
one (horizontal) dimension in a space of given width. The direction of the movement is
alternatively left-to-right and right-to-left: the object moves left-to-right until its right
edge touches the right boundary of the space, at which point it starts moving right-to-left;
conversely, the direction of its movement will revert to left-to-right whenever its left edge
touches the left boundary. Therefore, the bouncing algorithm calculates the positions
of the object at different times by tracing its movements.3 For example, in a space of
size 10, an object of size 2 and speed 6 will start moving left-to-right from position 0 at
time 0, to position 6 at time 1, then changing direction halfway through the next step
(when the right edge touches the outside wall) and landing on position 4 at time 2, then
changing direction again to land on position 2 at time 3, etc. The graphical realisation
of this example can be seen in Figure 1.
Figure 1. Bouncing object: space=10, speed=6, duration=2. First four moves.
3 The object’s position is always measured from its left edge.
5
Michele Zaccagnini
The bouncing algorithm can be used to determine the position of different notes
within a time interval. To move from a space domain to a time domain, we can simply
“translate” the dimension of the space into a time interval and the size of the object into
the duration of a sound event.4 If, in the example from Figure 1, I assign = 1 then the
space of size 10 can be thought of as a measure of 5/4 while the object size of 2 results
in a note duration of ♩ (one quarter note). Therefore the bouncing motion of Figure 1
will look something like the score of Figure 2.
Figure 2. Scored movement of the example in Figure 1.
The bouncing algorithm is, in other words, a way to generate patterns. These patterns
have some interesting characteristics:
• They are periodic.

• Their variability is highly sensitive to slight variations in their respective values:
two objects with similar, but not identical, combinations of speed and size can
produce very different patterns.5
While these patterns are not necessarily easy to predict, the algorithm designed to
generate them allowed me to gain control over several perceptual variables of the textures
they can produce. First, the algorithm creates a continuous rhythmic reshuffling of
notes. The relative placement of objects in a given space will be highly variable at
different times, given that the objects’ sizes and speeds are different from one another.
Second, by dynamically tweaking some parameters, the algorithm can be implemented to
produce varying degrees of complexity in the sound texture. It is fairly easy to illustrate
this point by visualising an enclosed space, such as the one of Figure 1, and imagine
the effect of introducing more bouncing objects in the same space: the system surely
gains in complexity. On the other hand, if I expand the dimension of the space while
keeping the number of objects constant, the resulting system becomes clearer and more
predictable, i.e. objects will “spread out”. In Invenzione I decided to use these features
of the bouncing algorithm to create a sort of “perceptual paradox”: the two parameters
most responsible for increasing and decreasing the system’s complexity, space dimension
and number of objects, play one against the other. The increase of the number of voices
is counterbalanced by the increase in the dimension of the time interval in which they
move iteration after iteration. Conversely, as voices begin to exit, the size of the space
decreases compressing the rhythmic distribution of the objects.
4 The speed has no direct musical correlate: it simply determines the “shifts” of the note within the time
interval at each iteration.
5 InA New Kind of Science [2], in a section titled “System of Limited Size and Class 2 Behavior” (p. 255),
Stephen Wolfram shows a case of pattern behaviour very similar to the “bouncing algorithm”. Wolfram
observes how a dot moving in a limited space will produce highly variable patterns depending on its
speed and the size of the space. These examples, which do not include the “bouncing” feature nor do
they contemplate objects of different sizes, present a remarkable variability in terms of their periods.
6
The formula to obtain the position of the object given the set of three variables is
obtained by first calculating the position Po of the object in “open space”:
Po = s × t
where s is the “speed” of the object and t is the iteration number. The position Pf of
the object moving forward is given by:
Pf = Po mod (L − d)
where L is the time-space interval and d is the duration-size of the object. The position
Pb while moving backwards is given by:
Pb = L − d − Pf .
θ is the variable that determines which direction the object is travelling (0 if moving
forward, 1 if moving backwards):
Po
θ= mod 2.
(L − D)
The position Px of the object within each successive time-unit is then:
Px = (1 − θ)Pf + θPb .
The patch in Figure 3 implements this bouncing formula, and reports the object
position given the same set of parameters: the speed (s), the time interval (L), the
duration (d), and the number of iterations (t). In this patch, the arithm-ser function
outputs a sequence of ascending numbers corresponding the iteration index (in the
example from Figure 1: t0 , t1 , t2 ...)
Since the algorithm outputs values that fall between 0 and L − d, each successive
value will need to be displaced in the score by a value equal to time interval times the
iteration number (L × t).6 For example, for a note at logical time t3 the value output by
the algorithm of Figure 3 needs to be added to L × 3.7 Figure 4 shows the top-level loop
generating the voices of the canon according to a set of input lists corresponding to the
speed, space dimension, duration and number of iterations. The internal bouncing loop
calls the main algorithm (Figure 3), and inner-offs iteratively adds L × t (the “inner-
offsets”) to each output value in order to shift it to its proper time interval.
If the inner offsets are added within each one of 11 the voices, there is another kind
of offset that needs to be added to the voices: the canonic, or outer, offsets. To create
the “perceptual paradox” mentioned earlier, a new voice enters whenever the size of the
time interval increases end exits whenever it decreases. To calculate the canonic offset
then, we need to observe first how L behaves over the course of the movement: L grows
arithmetically from 1 to 11 and then decreases in a palindromic fashion. Furthermore,
each interval is repeated a number of times equal to its size (L). It follows that after
6 Iterations take place as in stop-frame motion.

7 Inthe example in Figure 1, L = 10 so the first iteration starts at 0, the second starts at 10 with a
resulting position of 10 + 6 = 16, the third starts at 20 with a position of 24, etc.
7
Michele Zaccagnini
Figure 3. The implementation of the “bouncing” formula in OpenMusic.
the first voice begins the next one will be delayed by 1,8 the second by 4 (L2 = 2),
the third by 9. Interestingly, while the time interval grows arithmetically, the canon
entrances follow a geometric series. These secondary offset are calculated outside of
the loop from Figure 4 and added to the raw and inner-offset values inside by means
of a simple om+ function. The right outlet of the bouncing omloop outputs a list of
durations to be paired to each of the voice’s onset lists. Eventually, the onset lists and
their corresponding durations are used to build a list of chord-seq objects. After all
the rhythmic values are ready, they are matched with the pitch cells as described at
the beginning of this section, starting from octave transpositions of the pitch-set at its
original level and gradually, as each of the canonic voices enters, transposing the pitch-
set. Each chord-seq produced is then funnelled into a multi-seq object (see Figure 5) and
eventually exported as a MIDI file.
8 The L values, as well as the offsets, onsets, and durations, are later converted to millisecond values by
means of a multiplier (250ms). This multiplier makes the smallest possible subdivision a sixteenth note
at a tempo of ♩ = 60.
8
Figure 4. Patch generating the canonic voices. Note the inner and outer offsets added to the
main bouncing patch result.
9
Michele Zaccagnini
Figure 5. Multi-seq rendition of “Invenzione”.
Sinfonia
“Sinfonia” is also based on the diatonic cell {A, D, C, E}. This second movement is
similar to the one described in the preceding sections, in that it also aims to “muddy
the waters” of its transparent diatonic cell. But if “Invenzione” blurs the material by a
gradual process of layering, rhythmic reshuffling (“bouncing algorithm”), and “centrifu-
gal”, i.e. increasingly non-diatonic, transpositions, the second movement is even more
subtle and mysterious in its inner workings and perceptual result. Throughout the entire
movement the four notes of the set sound without interruption and always in the same
register; no other transposition interferes with the diatonic cell. Subtly, the timbre of
each note changes by means of re-articulation by different instruments and intervening
microtonal inflections.
The score explicitly states how between each note and its re-articulation, there should
be no gap. Figure 6 shows how each note-end corresponds to an attack in another
instrument with the same, if slightly inflected, pitch. This connection is made clear to the
conductor by a dotted line. As a result, the entire movement produces an uninterrupted
hue of sound. Some processes subtly work under the surface to produce slight variations
in the texture, though they are hardly clear to our perception. Far from being indicative
of any developing narrative—the lack of which constitutes the “negative” premise of the
piece—these varying traits of the texture only contribute to its general sense of fragility.
The slight variations of hue come from three distinct processes. First, the re-articula-
tion of the four notes follows a specific order. Second, each re-articulation spells one of
four different microtonal inflections. Third, the onset of each re-articulation is based on
10
Figure 6. Beginning of “Sinfonia” (dotted lines indicate the connections of the same pitch-set
elements if microtonally inflected).
a “hidden” polyrhythm. Interestingly, all three of these processes take advantage of the
same input: a matrix extracted by the OpenMusic function permutations with the set
(1 2 3 4). This function outputs all possible re-orderings of the set: in the case of a four-
element set, 4! = 24 permutations. The 24 × 4 matrix is transformed into a 6 × 16 matrix
by removing the output’s first level of parentheses and using the list-modulo function
with a modulus of 6.9 The result is a matrix of six rows:

1 2 3 4 2 1 3 4 3 1 2 4 4 1 2 3

1 2 4 3 2 1 4 3 3 1 4 2 4 1 3 2

1 3 2 4 2 3 1 4 3 2 1 4 4 2 1 3

1 3 4 2 2 3 4 1 3 2 4 1 4 2 3 1

1 4 2 3 2 4 1 3 3 4 1 2 4 3 1 2

1 4 3 2 2 4 3 1 3 4 2 1 4 3 2 1
9 list-modulo groups the elements of a list spaced at a regular interval and returns these groups as a list
of lists.
11
Michele Zaccagnini
Each of the six rows contains four instances of each element (i.e. four ones, four twos
etc.). In Figure 7 we can see how the patterns-›pitches omloop parses these lists. Four
global variables (a, c, d, e) are initiated with a list of four values, each corresponding to
a microtonal inflection of the variable’s original pitch: natural (no inflection), a quarter
tone up, a quarter tone down, and an eighth tone up or down (chosen at random). A
permut-random function contained in the micro-shift sub-patch shuffles the order of the
inflections within each variable at instantiation. Each time a variable is called a different
inflection is selected, creating microtonal oscillations around the centre pitch. Therefore,
this loop is responsible for both determining the order of notes being played, by selecting
one of the four variables depending on the order of the input list’s four elements, and the
rotation of microtonal inflections, by parsing the variable’s content.10
Figure 7. Picking the right note and inflection based on permutation lists.
10 Whenever an index “finds” its match, it parses the list stored as a global variable by means of a Lisp
nth function (in combination with loop counters). This escamotage is necessary as OpenMusic does
not provide “destructive” list operators such as the pop function. However, a simple line of code within
a Lisp object would do the job as well.
12
So far these processes are relatively straightforward: simple permutations operating

on pitches and timbre. The rhythmic organisation, however, presents more problems
both conceptually and in terms of programming. Conceptually the problem can be
summarised by questioning the meaning of operating permutations on rhythms. Since
the set of permutations illustrated earlier was to be the unifying element of the entire
movement, I created a set of four rhythmic values (in absolute time) to be associated with
the permutations: 1000, 1500, 2500, and 3500 ms.11 These values are used to determine
the onset of each note by subdividing the timeline and allowing the corresponding note to
fall only on one of those subdivisions. A note assigned to value 3500 ms, for instance, can
only have an onset at 0, 3500, 7000 ms, etc. To determine on which of the subdivisions
the note will actually fall, we need to look at the ordering of notes of the 6 × 16 matrix
illustrated above. The first note, which has a default onset at 0 ms, will determine the
onset of the following note since it is allowed to play “undisturbed” for at least its assigned
value. For instance, if the first note has an assigned value of 3500 ms and the second
1000 ms, the first note will have an onset at 0 ms while the second will have to “miss”
the first 3 subdivisions by 1000 and wait for the next available subdivision: 4000 ms.
The assigned values work then as “hidden polyrhythms” as well as setting the smallest
possible duration for the note. But actual durations are calculated by finding the next
occurrence of the same note (independent of its timbral inflection) and prolonging it up
until the following attack. In other words: duration(ai − 1) = onset(ai ) − onset(ai−1 ).
Figure 8 shows how all these calculations take place within an omloop. In particular,
the patch time-slots (visible at the right of the figure) finds the next available onset given
the previous assigned value and onset. The operation is relatively simple, performing a
Euclidean division of the last onset and multiplying the result for the next assigned value.
Conclusions
This chapter describes one step in a path on which I embarked several years ago and that I
am still pursuing. From an æsthetic perspective, my music aims to explore different kinds
of repetitive textures propelled by perceptually ambiguous inner workings—to induce a
semi-hypnotic state of consciousness in the listener.
As I have shown in this chapter, my attempts at inducing a specific perceptual
response in the listener are not based on abstract æsthetic intuitions but rather on build-
ing tools that could, if tested and implemented properly, correctly create a connection
between the composition and its perceptual outcome. In other words, the central task of
my work is to find and harness those compositional variables responsible for effectively
creating a link between the composition and the way music is perceived.
If this goal seems too abstract and/or ambitious to be successfully carried out yet, I
have come across a few expedients that have proven effective at easing the burden. First
of all, I have looked at patterns and phenomena that exist outside of the musical context.
The bouncing algorithm was designed starting from the simple intuition that to create a
musical environment that gradually oscillates between more or less chaotic states, I could
simply try to emulate an enclosed space with bouncing objects and “extract” some of its
11 Proportional to prime numbers 2, 3, 5, and 7.
13
Michele Zaccagnini
Figure 8. Left: Calculating onsets based on previous onsets and rhythmic values. Right:
Computation of the next available onset using Euclidean division. If the remainder of the
division is zero then the left value of the division is simply multiplied by the next assigned
value; if the division has a nonzero remainder the “ceiling” of the division is taken.
salient features. The sort of dynamic equilibriums that environments of the sort possess
can, as the movement “Invenzione” proves, be translated into music.
Second, relying on computer-aided composition to develop my algorithm has been a
crucial addition that has sped up calculations that would otherwise be tedious and time-
consuming. A computer-aided composition environment such as OpenMusic is an ideal
space to implement these experiments since it integrates powerful tools for iteration such
as omloop and the mapcar function with score and graphic tools. It is helpful to have
different way of displaying the results of my endeavours not only by using proportional
rhythmic notation objects (chord-seq and multi-seq) but also graphical ones (bpf, etc.).
References
[1] Michela Mollia. Autobiografia della musica contemporanea. Cosenza: Lerici, 1979.
[2] Stephen Wolfram. A New Kind of Science. Champaign, IL: Wolfram Media, 2002.
Acknowledgements: I would like to thank William C. Hoston, Yu-Hui Chang, Joshua Fineberg, and
Jean Bresson for their invaluable help in developing my ideas.
14
Folk material transformations and
elaborations in A Vida é Nossa
Gonçalo Gato
In this chapter I will discuss the compositional procedures associated with the use of
OpenMusic to compose the piece A Vida é Nossa (2013) for symphonic wind band.1
The main compositional work was based on folk material, which I transcribed and/or
analysed. After proposing a terminology for some archetypical procedures of computer-
aided composition, various configurations of “manual” and “machine” composition are
explored along with æsthetic reflection. Active human decision making plays a key
role, and as a result the computer-aided compositional process features much the same
æsthetic issues as purely manual composition. This opens up the scope of algorithmic
redesign and/or manual procedures, as the composer remains focused on his primary
musical goal.
Terminology
Computer-aided composition (CAC) places human-machine interaction at its core and
beautifully show-cases the interface between art and technology. After composing this
way for some years, I found it useful to define certain terms so as to clarify the general
procedures composers carry out while composing with algorithms:
Manual complementation: manual composition performed so as to coexist with al-

gorithmically generated materials that are conceived differently, thus completing a
given musical texture. It typically creates an additional musical layer with varying
degrees of relation to the preexisting algorithmic materials.
Manual elaboration: manual composition performed so as to develop preexisting al-
gorithmically generated materials. This presupposes a shared conception.
Algorithmic elaboration: an automated procedure designed and performed so as to
develop certain manually composed initial materials.
Manual intervention: manual modification/alteration of algorithmically generated ma-
terials. This can encompass simple actions such as assigning or changing the
instrumentation or dynamics, or more “destructive” actions such as changing or
deleting some of the pitches or durations.
1 Composed
as part of doctoral practice-based research carried out at the Guildhall School of Music and
Drama, London.
15
Gonçalo Gato
A Vida é Nossa
Gonçalo Gato
Austero, q = 120
° 4 3 2 4 2 4
∑ ∑ ∑ ∑ ∑ ∑ ∑ ∑ ∑ ∑ ∑
Manual implementation: using the output ofmanual an algorithm as a reference so that it is
Piccolo &4 4 4 4 4 4
complementation
manually incorporated into the∑ score. This4 ˙means J that4 it is up
- œ- œ ‰ œ. 2 ‰ œ. Œ
to theœ ‰ composer
4 œ. ‰ Œ
to
4 ∑ ∑ ∑ 3 2 4 œ™ . Œ
&4 Flauta 1 4 ∑ ∑ 4 4 ∑
f
write
& 4 the∑ notes in
4 Flauta 2 ∑ the score,
∑ generally
∑ 3
4 ∑ implying
∑ 2
4 b˙
4 œmanual
4 Jbœ ™ œ œ ‰ 4 bœ completion.
- - 2 . ‰ ‰ œ. 4 Œ ‰
4 bœ. Ó ∑
f
Automatic
4
&4 implementation:
∑Oboé 1∑ ∑ copying
∑ 4 an
3 ∑ algorithm
∑ 2˙
4
4 œ™ output
4 J ‰ 42 ‰ into
- œ- œ- œ. œ.
Œ 4 the
4 ‰ Œ score
œ. œ.
‰ Œ using ∑ auto-
mated
4
procedures.
∑ ∑
This
∑
usually
∑
implies
3 ∑
a
∑
MIDI
2 b˙
or
- - MusicXML
. .
4 œbœ ™ œ œ ‰ 2 bœ ‰ ‰ œ 4 Œ ‰
file export.
.
bœ Ó ∑
f
Oboé 2 &4 4 4 4 J 4 4
f
In Figure 1 the first page of the score of A Vida é Nossa 44is shown
Corne Inglês
4
&4 ∑ ∑ ∑
along∑ with
∑ ∑ 3
4 ∑ ∑ 2
4 ∑ 4
4 ∑ 2
4 ∑
annotations,
4 providing
∑
Requinta em Mib
&4 ∑ an example
∑ of
∑ how ∑ these
A43Vida ∑ 4procedures
2 ∑ 4
é Nossa 4 ∑ can
2
4 ∑ shape
4
4 various
∑ materials.
∑
4 3 2 b˙ 4 ˙ b-œ ™ œ 2 b>œbœ œ >œœœ œ 4 b>œbœœ œ œœ Gonçalo Gato
Clarinete 1 em Sib & 4

Austero,
∑
q = 120
∑ ∑ ∑ 4 ∑ ∑ 4 4 J 4 ≈ 4Œ bœ ®Ó ∑
° 4 ∑ ∑ ∑ ∑ 3 ∑ ∑ 2f ∑ 4 2 4
Piccolo & 4 4 4 4 - -∑ 4 > bœ∑ > œnœ 4 > ∑ ∑
4 3 2 4 œ œ œ œ œ œ 2 œbœ ≈œœ 4 Œ bœ œœbœ œœœ®Ó
Clarinete 2 em Sib & 4 ∑ ∑ ∑ ∑ 4 ∑ ∑ 4˙
manual 4complementation
J J 4 4 ∑
Flauta 1 & 4
4 ∑ ∑ ∑ ∑ 3 ∑ ∑ 2 f˙ 4 -œ™ -œ œ ‰ œ. 2 ‰ œ. Œ 4 œ. ‰ Œ œ. ‰ Œ ∑
4 4 4 J 4 4
4 ∑ ∑ ∑ ∑ 3 ∑ ∑ 2 4 2 4
Clarinete 3 em Sib & 4 4 4f ∑ 4 ∑ 4 ∑ 4 ∑ ∑
4 3 2 4 - - 2 . . 4 bœ. Ó
Flauta 2 & 4 ∑ ∑ ∑ ∑ 4 ∑ ∑ 4 b˙ 4 œJbœ ™ œ œ ‰ 4 bœ ‰ ‰ œ 4 Œ ‰ ∑
Clarinete Baixo ? 4 ∑ ∑ ∑ ∑ 3 ∑ ∑ 2 4 2 4
em Sib 4 4 4 f ∑ 4 - -∑- . 4 .
∑ 4 .
∑ ∑
4 3 2˙ 4 œ™ œJ œ ‰ œ 2 ‰ œ Œ 4 œ‰ Œ œ.
Oboé 1 & 4 ∑ ∑ ∑ ∑ 4 ∑ ∑ 4 4 4 4 ‰Œ ∑
Saxofone Alto 1 & 4
4 ∑ ∑ ∑ ∑ 3 ∑ ∑ 2f ∑ 4 ∑ 2 ∑ 4 ∑ ∑
4 4 4 4 4
4 3 2 b ˙ 4 œb -
œ ™ -
œ œ b .
œ .
œ b .
œ
Oboé 2 & 4 ∑ ∑ ∑ ∑ 4 ∑ ∑ 4 4 J ‰ 42 ‰ ‰ 4
4Œ ‰ Ó ∑
4 3 2 4 2 4
Saxofone Alto 2 & 4 ∑ ∑ ∑ ∑ 4 ∑ ∑ 4f ∑ 4 ∑ 4 ∑ 4 ∑ ∑
4 ∑ ∑ ∑ ∑ 3 ∑ ∑ 2 4 2 4
Corne Inglês & 4 4 4 ∑ 4 ∑ 4 ∑ 4 ∑ ∑
4 ∑ ∑ ∑ ∑ 3 ∑ ∑ 2 4 2 4
Saxofone Tenor & 4 4 4 ∑ 4 ∑ 4 ∑ 4 ∑ ∑
4 ∑ ∑ ∑ ∑ 3 ∑ ∑ 2 4 2 4
Requinta em Mib & 4 4 4 ∑ 4 ∑ 4 ∑ 4 ∑ ∑
?4 ∑ ∑ ∑ ∑ 3 ∑ ∑ 2 4 2 4
Saxofone Barítono 4 4 4 ∑ 4 ∑ 4 ∑ 4 ∑ ∑
4 3 2 b˙ 4 ˙ bœ- ™ œ 2 b>œbœ œ >œœœ œ 4 b>œbœœ œ œœ
Clarinete 1 em Sib & 4 ∑ ∑ ∑ ∑ 4 ∑ ∑ 4 4 J 4 ≈ 4Œ bœ ®Ó ∑
?4 3 2 4 2 4
Fagote 1 4 ∑ ∑ ∑ ∑ 4 ∑ ∑ 4f ∑ 4 ∑ 4 > ∑> 4 ∑ ∑
>
4 3 2 4 œ -œ -œ œ œ œ 2 œbœbœ≈œœœnœ 4 Œ bœ œœbœ œœœ®Ó
Clarinete 2 em Sib & 4 ∑ ∑ ∑ ∑ 4 ∑ ∑ 4˙ 4 J J 4 4 ∑
?4 ∑ ∑ ∑ ∑ 3 ∑ ∑ 2 4 2 4
Fagote 2
¢ 4 4 4f ∑ 4 ∑ 4 ∑ 4 ∑ ∑
Clarinete 3 em Sib & 4
4 ∑ ∑ ∑ ∑ 3
4 ∑ ∑ manual
4 ∑ 4 intervention
2 4 ∑ 4 + elaboration
2 ∑ 4
4 ∑ ∑
2.
+ +
° 4 ∑ ∑ ∑ ∑ 3 ∑ ∑ 2 4 2 4œ Œ
Trompas 1,2 em Fá & 4 4 4 ∑ 4 Ó ‰ 4˙ 4 Ó ∑
Clarinete Baixo ? 4 ∑ ∑ ∑ ∑ 3 ∑ ∑ 2 4 ∑ œ™ 42 b ˙ ∑ 4œ
em Sib 4 4 4 ∑ 4 p mf 4 ∑ ∑
3.
+ +
4 ∑ ∑ ∑ ∑ 3 ∑ ∑ 2 4 2 4
Trompas 3,4 em Fá &4 ∑ ∑ ∑ ∑ 43 ∑ ∑ 24 ∑ 4 Ó ‰
∑ bœ ™ 424 ˙ ∑ 4œ Œ Ó ∑
Saxofone Alto 1 &4 4 4 ∑ 4 b˙ 4œ ∑ ∑
p mf
4 33 22 4 22 44
Trompete 1 em Sib
Saxofone Alto 2 & 44
& ∑∑ ∑∑ ∑∑ ∑∑ 44 ∑∑ ∑∑ 44 ∑∑ 44 ∑∑ 44 ∑∑ 4 ∑∑ ∑∑
4 33 22 4 22 44
Trompete 2 em Sib
Saxofone Tenor & 44
& ∑∑ ∑∑ ∑∑ ∑∑ 44 ∑∑ ∑∑ 44 ∑∑ 44 ∑∑ 44 ∑∑ 4 ∑∑ ∑∑
4 33 22 4 22 44
Trompete 3 em Sib
Saxofone Barítono & 44
? ∑∑ ∑∑ ∑∑ ∑∑ 44 ∑∑ ∑∑ 44 ∑∑ 44 ∑∑ 44 ∑∑ 4 ∑∑ ∑∑
con sord.
bœ œ œ 2 ˙ œ
Trombone Tenor 1
? 44
? ∑∑ ∑∑ ∑∑ ∑∑ 33 ∑∑ ∑∑ 22 bœ œbœ 44 œbœ œ ∑J J 24 44 Œ Ó ∑∑
Fagote 1 4 44 44 ∑ 4 4 ∑ 4 ∑
mp austero f
con sord.
bœ œ œ 2 ˙
? 44
? ∑∑ ∑∑ ∑∑ ∑∑ 33 ∑∑ ∑∑ 22 44 œbœ ∑J J 424 44 œ Œ Ó ∑∑
Trombone Tenor 2
Fagote 2
¢ 4 44 44 #œ œ∑ œ 4 œ ∑ 4 ∑
manual intervention
œ œ œ b+
˙ elaboration
mp austero f
?4 ∑ ∑ ∑ ∑ 3 ∑ 2 œ œ œ 4 œ œ œ J 2.+ J 2 +
∑ œ 4 Œ ‰bœ ™ œ™ ‰ Ó
Trombone Baixo ° 44 43
∑ ∑ ∑ ∑ ∑ 24 ∑ 44 Ó 24 44 œ
Trompas 1,2 em Fá &4 4 4 mp ∑austero
4 ‰ 4˙ 4 Œ
f
Ó mf ∑
œ™ b˙ œ
Tuba
?4 ∑ ∑ ∑ ∑ 3 ∑ ∑ 2 4 jp j 2 mf 4 Œ ‰ ‰ Ó
¢ 4 4 4 4 œ 3.œ+ œ 4 b˙+ 4œ bœ™ œ™
4 3 ∑ 2 œ œ∑ œ 4 Óœ œ œ ‰ 2 4
algorithm output 4 mp
Trompas 3,4 em Fá &4 ∑ ∑ ∑ ∑ 4 ∑ 4 Œ Óf ∑
austero bœ ™ 4 ˙ 4œ f
l.v.
°? 4
∑ ∑ ∑ ∑ 3 2 4 ∑ p 2b ˙ ∑ 4œ ∑ ∑
Tímpanos
¢ 44 4 bœ ™ œ ™ bœ ™ œ ™ 4 4 4 mf 4
3 ∑ 2 ˙ 4 2 4
∑ ∑ ∑ ∑ ∑ 4 f —∑ 4 ∑ ∑ ∑ ∑
manual complementation
Trompete 1 em Sib &4 4p fff 4 4
Vibrafone
l.v. sempre
° 4¡
A
∑bw 3 2 ∑ 44
3
2 4
Trompete 2 em Sib
Percussão 1 &4 w ‘∑ ‘∑ ˙ b˙ Œ∑ b œ œ œ 4 Œ ∑Œ ˙ ˙ Œ b∑ œ œœ 4 ∑ j 4 ˙ ∑ 4˙ ∑ ∑
œœ œb œ‰ ‰ œ œ œ œ
ppp bœ bœ bb ww
ww œ ˙˙˙ ˙˙ ˙˙ bb œœœœ œœœœ œœœœ mp
mf mp f
Marimba
4 ∑ 3 ∑ ∑ ∑ 3 2 4 2 4
Trompete 3 em Sib &4 ¡ 4 ∑ ∑ 4 ∑ 4 ∑ 4 ∑ 4 ∑ ∑
4 A 3
Œ 3 2 4 2 4
Percussão 2 &4 j 4 ˙™ b˙ ™ ˙™ b˙ ™ 4 ∑ 4 ∑ 4 ∑ 4w bw ‘
bbb œœœœ bbb œœœœ ˙˙˙˙ œœœœ ˙˙˙˙ b œ b œœœ ™™™ ˙˙˙ œ b œbb œœ œœ ˙˙ ™™ ppp
con sord.
b œ œ œ
Trombone Tenor 1
?4 p
4 ∑ ∑ ∑ ∑ 3
4 ∑ ∑ 2 4 b œ
4 bœ œbœ 4 œ œ J J 42 ˙ 4œ Œ
4 Ó ∑
Wood Blocks mp austero f
4 Ó Œ Œ‰ j 3 ‰ Ó™ 2con sord. 4 bœ œ Œ œ 242 ˙ ™ j 4 j
Percussão 3 / 44 œ
? œ ∑œ œ œœœ œ œ œœ ∑ œœ ∑œ œ œ œ œ∑ œ 443 œ œ ∑œ œ 24 Œ 44 œ 444 œ œœœŒ Óœ ™
∑ 4 #œ œœ œ 4 œœ œbœ Jœ œ J 4 œ œ œ œ œœœ ∑œœœœœœœ
automatic implementation
Trombone Tenor 2 4 p
mp austero Sinos Tubulares f
Prato Suspenso
Para o Tantã Tantã Para o Glock.
arco
Percussão
Trombone Baixo4
?4 w ∑ w— ∑ ∑∑ ∑∑ 43 ∑∑ ∑∑ 24 œ˙—œ œ 44 œ œ œ ∑œJ œ œJ& 24 b˙œ 4œ Œ ∑ ‰bœ ™ œ™ ∑‰ Ó
¢ / 4 mp p
— #œ — 4
mp austero f f mf
Tuba
?4 ∑ ∑ ∑ ∑ 3 ∑ ∑ 2 4 j j2 4 Œ ‰ ‰ Ó
¢ 4 4 4
œœœ
4
œœœ
œ œ œ 4 b˙ 4œ bœ™ œ™
Figure 1. First page of the score (in C). output
algorithm Dashed boxes indicate manual procedures. mp austero
l.v.
f f
°? 4 ∑ ∑ ∑ ∑ 3 2 4 ∑ 2 ∑ 4 ∑ ∑
Tímpanos
¢ 4 4 bœ ™ œ ™ bœ ™ œ ™ 4 4 4 4
˙—
manual complementation
p fff f
Vibrafone
l.v. sempre
° 4¡
A
3 2 4 3
2 4
Percussão 1 &4 w bw ‘ ‘ ˙ b˙ Œ b œ œ œ 4 Œ Œ ˙ ˙ Œ b œ œœ 4 ∑ 4 ww j 4˙ 4˙ œœ œb œ‰ ‰ œ œ œ œ
˙˙˙ bb œœœœ œœœœ œœœœ mp
16
bœ bœ bb ww œ ˙˙ ˙˙
ppp
mf mp f
Marimba
4 A¡ 3 3
Œ 3 2 ∑ 4 ∑ 2 ∑ 4
Percussão 2 &4 œ œ ˙ œ ˙ j œ ™ ˙ œ 4 ˙™ b˙ ™ ˙™ b˙ ™ 4 4 4 4w bw ‘
bbb œœœ bbb œœœ ˙˙˙ œœœ ˙˙˙ b œ b œœ ™™ ˙˙ b œbb œœ œœ ˙˙ ™™ ppp
p
Wood Blocks
4 j 3 ™ 2 4 2 j 4 j
œ œ œ œœœ œ œ œœ Ó
Œ œœ œ
Percussão 3 /4 œ œ œ œ Œ ‰œ œ 4 œ œ œ œ‰ Ó 4 Œ œ 4 œ œ œ œ œŒ 4 œ™ œ 4 œ œœœ œ œ™ œ œ œœœ œœœœœœœ
p automatic implementation
Prato Suspenso Sinos Tubulares
Para o Tantã Tantã Para o Glock.
arco
4 w w— 3 2 4 2 œ 4
Percussão 4
¢ / 4 mp ∑ ∑ 4 ∑ ∑ 4 p˙— 4 ∑ &4 — #œ — 4 ∑ ∑
f
Folk material transformations and elaborations in A Vida é Nossa
Folk material
In their account of Claude Debussy’s (1862-1918) approach to folk material, François
Lesure and Roy Howat write:
Debussy was always consistent on the point that a folk or national music should not
be used for its themes but rather in the manner of Albéniz: “Without using actual
popular tunes he is the kind of person who has them in his blood. They have become
so natural a part of his music that one barely distinguishes a demarcation line”. To
a Hungarian friend, he wrote: “Your young musicians could usefully take inspiration
from them, not by copying them but by trying to transpose their freedom, their
gifts of evocation, colour, rhythm. [...] One should only ever use the folk music of
one’s country as a basis, never as a technique”. [7]
We can say Debussy prescribed a progressive approach for dealing with folk materials.
Béla Bartók (1881-1945) is a another composer particularly relevant to consider in this
context. Malcolm Gillies writes:
Indeed, after this early Violin Concerto none of his works escapes a strong folk
influence. In his later lecture The Relation between Contemporary Hungarian Art
Music and Folk Music (1941, in Béla Bartók Essays, 348-53), Bartók exemplified
three types of arrangement: where the folk melody is mounted like a jewel, where
melody and accompaniment are almost equal in importance, and where the folk
melody is a kind of inspirational motto to be creatively developed. [6]
Bartók was important in my own music education, and while composing A Vida é Nossa
the same æsthetic issues brought me closer to his techniques.
It is interesting to recall some of Bartók’s harmonic methods. Referring to the
Fourteen Bagatelles, Op. 6 (1908, two of which directly quote folksongs), Gillies explains
that “any sense of functional harmony is persistently undermined by the use of ostinato
figures (nos. 2, 3, 5, 10, 13), quasi-bitonal writing (nos. 1, 13), streams of parallel 5ths and
7ths (no. 4), of 4ths (no. 11), of tritones (no. 8), or of piled-up 3rds (nos. 7, 9, 10)” [6]. As
I read these accounts, I began to see ways of manipulating the folk music I was selecting,
and the techniques I developed resonate with Bartók’s approaches.
Frequently the strategy I follow is to have folk tunes as part of some texture that is
based on them, but also have other features that turn them into a new and autonomous
sonic entity (thus “transposing their freedom, their gifts of evocation, colour, rhythm [...]
as a basis, never as a technique”, as Debussy prescribed). For instance, the simple idea
of constructing heterophony could be enough to achieve this effect. This gave rise to
the “rhythm-focusing heterophony” and “proliferating canonical heterophony” textures,
on which I will go into detail further below. Other times, the folk monodic material
was subject to different kinds of transformations, either conservative (colouring) or more
destructive (resynthesis).
17
Gonçalo Gato
Analysis and resynthesis

The folk tunes I used were selected from sources on the Internet.2 To characterise
them, I developed analytical procedures that obtained both the set of melodic intervals
contained, and the set of rhythmic cells used in melodies and percussive accompaniments.
These two procedures (rather like dissections) produced materials I could then elab-
orate algorithmically, drifting away from the folk reality but, nevertheless, maintaining
a relationship with it. As an example, consider the folk melody Senhora do Almurtão
shown in Figure 2, which I transcribed from an online video of the Portuguese folk group
Adufeiras de Monsanto.3 There are clearly two parts, the second starting in the twelfth
bar. The analysis of the melodic intervals produces Table 1. One can say that, although
the ascending major second is only present in the second part, both parts share the same
set of interval moduli: 0, 1, 2 and 5 semitones. This set can be considered a characteristic
of the melodic line.
Figure 2. Transcription of the folk melody Senhora do Almurtão.
Part Melodic Intervals

1 0, -1, -2, +5
2 0, -1, -2, +2, +5
Table 1. Intervallic analysis of Senhora do Almurtão.
In much the same way, so can a set of rhythmic cells define a characteristic of a rhythm.
A process of polymerisation can then be created by concatenating cells (monomers) in
any given degree of random ordering, the results of which are similar to improvisations by
folk performers. Such processes can be carried out both in the rhythm or pitch domains.
The initial woodblock part of the score (Figure 1), for instance, is generated from a
random combination of rhythmic cells from the percussion instruments (adufes) played
by the Adufeiras de Monsanto. Figure 3 shows the OpenMusic patch implementing
such random rhythmic cell concatenation.
In the electroacoustic realm, cross-synthesis implies the combination of analysis data
from two sounds [4]. By analogy, one can term cross-resynthesis the process by which the
initial low brass melody of the piece (Figure 4) was obtained: intervals were combined
2A música portuguesa a gostar dela própria: https://vimeo.com/mpagdp.

3 Video directed by Tiago Pereira, Monsanto, Portugal, 2011: http://vimeo.com/62887865.
18
Figure 3. Rhythmic polymerisation from cells and pauses.
from the previous interval set (Table 1)—altered to include ascending and descending
versions of each interval—with percussive rhythmic cells from another folk source: a Viola
da Terra performance,4 the cells of which are featured in Figure 3. The combination
process is shown in Figure 5.
Figure 4. Initial brass melody with rhythmic cells in parenthesis.
Figure 6 details the patch intervallic_melody_range, which generates a number of

pitches starting from the set of intervals, constraining the result to a specified interval
range. Note that if the exact interval sets, taking into account interval direction, were
to be used, the resemblance created would perhaps have been too close. Instead, both
the order and direction of the intervals are selected randomly, therefore creating varied
and fresh melodies at each evaluation.
4 DécioLeal toca Viola da terra, http://vimeo.com/39351351. Video directed by Tiago Pereira, 2012.
Viola da Terra is a local string instrument from Açores.
19
Gonçalo Gato
Figure 5. Folk cross-resynthesis. Rhythmic material from a folk source is combined with
intervallic material from another. The resulting melody was subsequently transposed two octaves
down and its durations were doubled.
Figure 6. The patch intervallic_melody_range, generating a sequence of pitches in a given

range, from a set of intervals.
One of the beautiful and elegant aspects of algorithmic composition of this sort lies
in the fact that it enables the composer to shift his or her attention from algorithmic
design to the aural selection of materials, provided that he or she is sure the algorithm
reproduces the desired compositional technique. Drawing an analogy to the common
20
use of the piano to try out compositional ideas, algorithmic tools constitute a kind of
intelligent piano; a musical laboratory. This significant and underlying aspect is one of
the main reasons why I use OpenMusic.
Rhythm-focusing heterophony
Starting with a folk melody, I wanted to build a texture that would gradually bring into
sync various parts by progressive rhythmic diminution. The voices would start “unfo-
cused” by virtue of the different magnitudes of augmentation. They would then gradually
diminish their durations until all voices became synchronised, creating a homorhythmic
texture. The process is depicted in Figure 7.
most augmented voice
complete sync
homorhythm
neutral voice
diminution of durations
Figure 7. Rhythm-focusing process by diminution.
The algorithm I used is shown in Figure 8. A folk melody called Chamarrita Preta,
performed by the Grupo Folclórico da Casa do Povo de Bandeiras,5 is initially analysed
for its sequence of relative durations (using tree2ratio). It is then multiplied by 16
because a 16th note, or semiquaver, is the smallest duration present in the melody. This
makes all durations integers, with a minimum value of 1. The sequence of durations
is reversed, doubled, and fed into an iterative augmentation process of different mag-
nitudes: the leftmost mapcar function calls the interpolation lambda function, which
interpolates the initial sequence with five increasingly magnified versions. The focusing
process is thus calculated in retrograde form: the initial sequence of durations, perfectly
superimposed on the different voices at the beginning, is subject to augmentations of
different magnitudes, gradually bringing the rhythms out of sync (as if reading Figure 7
backwards). Once these augmentations are calculated, the voices are retrograded again
(retrogade_rhythms abstraction) so that the result becomes a gradual diminution of
the durations, culminating in absolute homo-rhythm with the original melody (unison
doubling). The end result of the algorithm is the poly object shown at the lower right.
5 https://vimeo.com/39572090. Video directed by Tiago Pereira, 2012.
21
Gonçalo Gato
Figure 8. Algorithm for melodic focusing. The bpf in the centre shows the durational contours
of melodies. The bpf-lib in the lower left shows the progressive augmentation of the durations of
each voice. The patch window on the right shows the interpolation abstraction. Note that only
the notes of the bottommost voice of the poly, the most augmented voice, are visible because
the remaining voices were subject to smaller augmentation magnitudes and so start later in the
score.
The durations were doubled in order to allow the minimum duration to be subdivided
into integer values. As the augmentation process uses only integers, it eliminates the
possibility of tuplets (which would create quantisation and playability issues). The
decision to use only integers is both operational, to prevent quantisation problems,
and aesthetic, not to allow tuplets. I define operational decisions as those taken for
purely technical reasons, while æsthetic decisions are those taken according to principles
underlying artistic creation. Hence, algorithm design is here directed by a musical goal,
which in turn determines what constraints to apply. While programming, I continuously
tested the musical results by using the playback feature so that I could modify the visual
code and/or the initial parameters to achieve a better sense of gradually coming into
sync or focus. In this way algorithms can be used to fine-tune a compositional technique.
22
sync
œ œ œ œ œ #œ œ œ œ œ œ Œ œ œ #œ œ œ œ œ
& #œ œ œ œ
œ œ
& <#>œ ™ œ œ œ œ œ ™ œ œ™ œ œ œ œ œ #œ œ œ œ Œ œ œ #œ œ œ œ œ
& œ™ œ ˙ œ œ œ™ œ #œ œ™ œ œ œ œ œ™ œ œ œ ™ œ œ Œ ≈ œ œ œ #œ ™ œœœ œ œœœœ

sync
& <#>œ#œ ™ œ œ œ ™ œ œ œ œ œ ™ œ ‰ ‰ ™ œr œ œ œœœ œ œ™ œ œ ˙ œ œ œ™ œ #œ œ™ œ

J #œ œ
& <#>œ œ œ œ™ œ œ œ ™ œ œ Œ ≈ œ™j œ™ #œ œ œœœ œ œ™ œ œ ˙ œ œ œ™ œ #œ œ™ œ
& œ
œ œ œ œ Œ ‰ œj œ œ™ #œ œ œ œ œ™ œ œ œ™ œ œ˙ œ œ™ œ œ œ™ #œ œ œ
sync
#œ œ œ œ
& œ œ œ œ œ œ Œ œ œ #œ œ œ œ œ œ œ œ œ #œ œ œ œ
& œ œ œ œ #œ œ œ œ œ œ Œ œ œ #œ œ œ œ œ œ œ œ œ #œ œ œ œ
& œ œ ™ œ œ ™ œ œ™#œ œ™#œ œ™ œ œ ™ œ œ™ œ œ

™ œ œ ™ ≈ ‰ ™ œr œ œ ™ œ œ #œ œ œ
J #œ œ™ œ œ œ œ œ œ œ™ œ œ œ œ
sync
& <#>œ œ œ œ™ œ œ œ ™ œ œ Œ ≈ œ œ œ #œ ™ œ œ œ œ œ ™ œ œ ˙ œ œ™ œ œ œ™#œ œ œ œ œ ™ œ œ œ™ œ œ
& <#>œ œ œ œ™ œ œ œ ™ œ œ Œ ≈ œ œ œ #œ ™ œ œ œ œ œ ™ œ œ ˙ œ œ™ œ œ œ™#œ œ œ œ œ ™ œ œ œ™ œ œ

œ ‰ ™ ‰ œj œ œ™
& <#>œ œ ™ œ œ œ™ œ œ
œ
R #œ œ œ œ œ œ™ œ œ œ™ œ ˙ œ œ™ œ œ œ™#œ œ œ œ œ ™ œ œ œ™ œ œ
complete sync
& œ
œ Œ œ œ #œ œ œ œ œ œ #œ œ œ œ œ œ Œ œ œ
œ œ œ
& œ œ Œ œ œ #œ œ œ œ œ œ œ œ œ #œ œ œ œ œ œ Œ œ œ
& œ
œ Œ œ œ #œ œ œ œ œ œ #œ œ œ œ œ œ Œ œ œ
œ œ œ
œ ‰ ™ ‰ œj œ
& œ œ œ ™ œ œ™ œ œ #œ œ œ œ œ Œ
R #œ œ™ œ œ œ œ œ œ œ œ œ
œ ‰ ™ ‰ œj œ
œ ‰ ™ ‰ œj œ
œ œ œ œ œ #œ œ œ œ œ œ Œ œ œ Œ Ó
& #œ œ œ œ #œ
& #œ œ œ œ #œ
& #œ œ œ œ œ œ œ œ œ #œ œ œ œ œ œ Œ œ œ Œ Ó
#œ
& #œ œ œ œ #œ
#œ
#œ
Figure 9. Last bars of the passage showing the synchronisation points of the voices.
23
Gonçalo Gato
Once I was satisfied with the result, I exported it as a MusicXML file and imported it
into notation software (see Figure 9). I then selected the part of the sequence beginning
when all voices are present, and ending with all voices playing the same melody at the
same time (the point of complete sync). The texture was scored for 2 oboes, 2 clarinets,
and 2 alto saxophones. As I conceived it, the instruments needed to be woodwinds— 7
I wanted
° a soft
88
∑
and stable ∑
dynamic ∑ Œ
and some timbral
. heterogeneity—and have compatible
5 6
bœ. œ#œ#œ.
œ. 3 3 3
Picc. #œ œ#œ#œ œbœbœnœœœ‰ œ ‰ #œ œ ‰ #œ. ‰ ‰ Ó ∑ Ó ‰
registers. In Figure 10, one can see œ the gradual. . synchronisation of all six parts in the
&
. . 3 ff 3
ff
final score (some of them >œcome b œ b œ nœ#œ into sync with

œ bœ œ bœ b œ bœ one another before
œ bœ bœ œ nœ b œ œ b œ b œthat point, as shown
bœ œ b œ b œ b œ n œ bœ nœ #œ.
& œ œ œ œ ™ nœ ™ œœ bœ œ œ™ œ ™ nœ œ™
in Figure 9). The shaded parts in the figure are not involved in the process: they
Fl. 1
œ œ Œ Œ nœ. ‰ Œ
œœ 3 3 3 3 3 ff 3 3 3 3
3 3
were composed manually (based >œ™ #œ ™

on algorithm-generated pitch reservoirs discussed further œ 3
nœ œb œ 3 3 3
below)& and∑ work as a complement to theœ™

rhythm-focusing texture.
∑ Œ œ™ bœ œ œ™ œ™ ‰ nœ#œ œ œ œ#œ œ œ œ #œ œ#œ œ. nœ œ œ. œ œ. ‰ œ œ#œœbœ
Fl. 2 J . . . 3
ff 3 3 3
Ob. 1 & œ™ œœ œ- œœ ™ œœ™ œ ˙ œ œ #œ œ- œ œ œ œ- Œ œ œ- #œ œ œ- œ œ ˙ œ œ #œ œ- œ œ
-œ œ œ ™ œ œ œ™ œœ -œ œ ‰™ œ-
Ob. 2 & œ ˙ œ œ™ œ œ œ™ #œ œ ‰ œ œ- #œ œ™ œ œ œ- œœ ™ œœ™ œ ˙ œ œ #œ œ œ
œ #œ . .
Cor. Ing. & œ œ™ œ bœ œ ™ œ œ œ œœ Œ œ œ œ œ #œ Ó Œ œ. nœ œ. œ œ. #œ
œ œ™ œ #œ œ #œ œ
œ ™ #œ
f œ
f fff
3 3 3 3 3 3 nœ bœ bœ b œ b œ nœ bœ bœ 3
3 3
3 3 3 œœ
Req. Mib & œ b˙ ˙ nœ bœ œ œ œ Œ Œ œ bœ nœ
nœ #œ œ nœ <n>œ Œ nœ#œ œ œ nœ nœ#œnœ #œ. œ. œ. #œ. nœ. œ. œ. œ œ œœ œbœnœ
nœ œ ˙ ˙ nœ
3
f
Cl. 1 & œ ˙ œ œ™ œ œ œ™ #œ œ
-œ œ œ ™ œ œ œ™ œœ œ- œ ‰™ ‰ œ œ- #œ œ™ œ œ œ- œœ ™ œœ™ œ ˙ œ œ #œ -œ œ œ
#œ œ- œ œ œ -œ #œ œ-
Cl. 2 & #œ œ œ- œ œ ˙ œ œ Œ œ œ- #œ œ œ- œ œ ˙ œ œ œ œ
3 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
j . .œ . . œ. . . œ. . œ œ œ Ÿ
Cl. 3 & œ œ™ œ™
œ œ bœ ™ œ œ œ œ œ™ ‰ ∑ ∑ #œ. œ. œ. œ#œ . œ. œ. œ œ œ. œ#œ œ œ œ œ œ ˙ #œ
f 3 3
mf ff fff
œ œ œ™ bœ œbœ™ œ bœ bœ œ bœ œbœ™ ™ bœ œ #œ
? bœ ™ œ™ J œ™ bœ œ œ™ œ™ œœ bœ ™ bœ nœ Œ œ #œ #œ nœ œ
Cl. B. J ‰ #œ ™ œ œ
J fff
f f
#œ -œ œ œ œ -œ #œ -œ
Sax. Alto 1 & #œ œ œ- œ œ ˙ œ œ Œ œ œ- #œ œ œ- œ œ ˙ œ œ œ œ
Sax. Alto 2 & œ™ œ˙ œ œ™ œ œ œ™ #œ œ œ- œ œ ™ œ œ œ™ œœ œ- œ ‰™ ‰ œ œ- #œ œ™ œ œ œ- œœ ™ œœ™ œ ˙ œ œ #-œ œ œ œ
˙ # œ# œ œ #œ #œ œ # œ.
Figure
? 10. Focusing in˙ the context of the score (inœ C). Shaded‰ ‰parts are not involved œ inœ the
œ œ b˙ ˙ bœ nœ ˙
œ #œ #œ bœ œ ˙
nœ #œ #œ œ #œ #œ œ œ œ. . . œ œ bœ œ nœ #œ œ .™ . . . . . . .
Fg. 1 ≈ œ #œ
3 3 3 3 3 3 3 3
process. The dashed line indicates the point from which the texture becomes homo-rhythmic.
3
f fff
œ b˙ ˙ œ œ ˙ b˙ œ œ œ. œ. œ. . . . . . œ. œ
Fg. 2
¢
? Œ Ó ∑ ∑ ? ‰ ‰ ‰ #œ #œ ‰ œ #œ œ ≈ œ œ œ œ œbœ
3 3 3 3 3 3 3
3
f fff
Virtual fundamentals
1.
° ∑ ∑ ∑ Ó Œ ‰ Œ
Tpa. 1,2 & w #ww
# œœ ˙˙ # œœ.
fp f mp f fff
While& I was∑ exploring∑ techniques for dealing with the# œ folk

‰ Œ material,#œ I faced the challenge
3.
Tpa. 3,4 ∑ Ó Œ Ó™ w
# œœ ˙˙
of creating complementary music6 for the tune Chamarrita Preta (as seen, for instance,
œ. Œ #˙ ™
fp f
mp f fff
in the
? shaded
Tbn. 1 ∑ parts ∑ of Figure
∑ 10).
Ó How
Œ can œ ˙one create
œ.
‰ Œ a harmony
∑ Œ that
#œ ™ is different
>œ œ
J
from, but related to, the melody’s pitch collection, and that enables some freedom and mf-pp f pp f
? ∑ ∑ ∑ Ó Œ #œ ˙ #œ. ‰ Œ ∑ Œ Œ #œ œ >œ œ
Tbn. 2
mf-pp f 3 pp f
6 See the first section of this chapter for a definition of musical complementation.
Tbn. B.
? ∑ ∑ ∑ ∑ ∑ ∑ Ó Œ ˙
¢
3 pp f
24 œ bw Para o Xil.
° Aœ
¡
Mar. & ∑ ∑ ∑ ∑ ∑ Ó œ œ w
f
Prato Susp. ¢/ ∑ ∑ ∑ ∑ ∑ ∑ w7
pp ff
greater dissonance? How could that harmony be compatible with diatonicism? I chose to
explore techniques from spectral music to obtain two solutions, which were individually
programmed and tested in OpenMusic:
1. Find the virtual fundamental of the diatonic pitch collection. Complementary parts
play harmonic partials of that fundamental.
2. Consider the diatonic melody’s pitches as the same partials of changing funda-
mentals to be calculated (for instance, suppose all pitches are seventh partials
of different fundamentals). Then, obtain other partials of those fundamentals by
transposing the pitch collection by an interval corresponding to an integer-ratio in
frequency.
Techniques related to virtual fundamentals have been around for some time. Tim
Howell discusses their use in Magnus Lindberg’s Kinetics (1988-1989):
The acoustic properties of [these serially derived] chords are [then] analysed in
relation to the overtone series, as group of partials issued from one fundamental.
That spectral analysis allows all chromatic constituents to be viewed in relation to
an underlying fundamental, which may be deployed in their re-harmonisation; this
can transform initial dissonance into sonorous chords. [2], p. 239.
Exploring the same technique, I tried to achieve a degree of harmonic amalgamation

between the folk melody and the added partials, while at the same time avoiding obvious
consonant combinations. The algorithm I used is shown in Figure 11.
Figure 11. Algorithm to find the virtual fundamental of a pitch collection and calculate the
first n harmonic partials.
I was after a bell-like sonority that would transfigure the folk melody’s diatonic
harmonic field. This perceptual shift was rehearsed and tried-out using the playback
features of both notation software and algorithmic software (OpenMusic). The various
calculated harmonic series were tested by playing back their superimpositions on the folk-
melody-based heterophony. From the eight possibilities shown in Figure 11, the second
25
#œ - - #œ -
mp dolce
"Chamarrita Preta" (trad. Portugal)
∑ ∑ ∑ ∑ ∑ #œ œœ œ œ œ™ œ œ ™ œ- œ œ‰Œ ≈ œ™ œ™ œ œ
Sax. Alto 2
¢& œ œ™ - œ
- #œ œ œ œ œ œ™ œ œ œ
mp dolce
Glockenspiel
l.v. sempre
° ˙ #˙ ˙ œ #˙ ™ œ
W.B. / ∑ ∑ ∑ ∑ ∑ & ˙ ˙ #˙ ˙ œ™ J
p
Gonçalo
& Gato
Pand.
∑ ∑ ∑ ∑ Œ
Triângulo
˙—™ Para a Mar.
∑ ∑ ∑ ∑ ∑
p
Para o Estalo
Bombo / w— ∑ ∑ ∑ ∑ ∑ ∑ ∑ ∑ ∑
Prato Susp. ¢/ ∑ ∑ w7 w7 w7 l.v.
∑ ∑ ∑ ∑ ∑
ppp f
chord of the quarter-tone-based set was selected. I then arranged some of the calculated
partials
= in the brass section, making use of the easy accessibility of quarter-tones on the
trombones (Figure 12).
36 D
° -œ œ œ ™ œ œ œ™ œ œ -œ œ ‰ ™ ‰ œ œ œ™ - ™ -
Ob. 1 & ˙ œ œ™ œ œ œ™ #œ œ - #œ - œœ
œ œ™ œ™ œ œ œ™ œ œ™ œ œ œ œ#œ ™ œ œ œ™ œ œ œ œœ œ œ
fundir-se com os metais
œ- - -
Ob. 2 &
œ ‰ Œ œ™ œ- œ#œ œ œ™ œ- œ œ œ œ œœœ œ œ œ™ œ™ œ œ œ#œ ™ œ™ œ œ œ œ™ œ™ œ œ œ œ™ œ™ œ œ œ ‰™ ‰™ œ œ œ- œ #œ ™ œ™ œ œ œ- œ œ ™
#œ œ -œ œ œ œ -œ œ
Cl. 1 & #œ - œ
œ œ œ œ™ œœ œ ˙ œ œ œœœ œ œ œœ Œ ‰ œ œ œ™
- #œ - œ
œ œ œ œ™ œ œœ ˙
- ™ - ™ -œ
Cl. 2 & ‰ ™ œ œ œ- #œ œ œ œ- œ œ œ™ œ œ œ œ™ œ œ œ œ#œ ™ œ #œ œ™ œ œ œ œ œ œ œ œ ≈Œ œ œ- #œ œ œ œ- œ œ ™ œ œ œ™ œ˙ œ œ œ™ œ #œ œ™
Sax. Alto 1 & ˙ œ œ #œ œ- œ œ œ œ- Œ œ œ- #œ œ œ- œ œ ˙ œ œ #œ œ- œ œ œ œ- Œ œ œ- #œ œ œ- œ œ

-œ -œ
Sax. Alto 2
¢& œ œ™ œ œ œœ œ œ#œ ™ œ œ™ œ œ œœœ œ œ™ œ œ™ ≈ Œ ‰ œ œ œ™
- #œ œ™ œ œ œ- œ œ ™ œ ™ œ œ œ œ™ ˙ œ œ™ œ™ œ œ œ#œ ™
3.
° 3
Tpa. 3,4 & ∑ ∑ ∑ Ó Œ ‰‰
#œ w w w w
pppp mf
con sord.
Tpt. 1 & ∑ ∑ ∑ ∑ ∑ Œ ˙ ˙ w w
3
pp
senza sord.
Tpt. 2 & ∑ ∑ ∑ ∑ ∑ w w w
pp mp
senza sord.
Tbn. 1
? ∑ ∑ ∑ ∑ Ó ‰ µœ œ ˙ œ ˙ w w
3 3
pppp mp
senza sord.
? ∑ ∑ ∑ ∑
Tbn. 2 Bw w w w
pppp mf
senza sord.
3
Tbn. B.
? ∑ ∑ ∑ ∑ Œ Œ
¢ Bœ ˙ w w w
pppp mf
° ˙ œ #œ ˙ œ œ™ œ˙ œ™ j j j #˙ ˙ œ œ™
Glock.
¢& J œ™ J
œ ˙ œ™ #œ ˙ œ™ œ ˙ ˙ J
Figure 12. Score implementation

Figure 12. Score of the calculated partials
implementation (in C).
of the calculated partials (in C).
To sum up, the same technical procedure used by Lindberg in Kinetics is used in
To sum up, the same technical procedure used by Lindberg is used in A Vida é
A Vida é Nossa but to carry out the opposite perceptual transformation accounted for
Nossa, but to carry out the opposite perceptual goal. As the folk tune’s pitches begin to
by superimposed
be Howell. As the onfolk
the tune's
added pitches
partials,begin
their to be superimposed
aural on the added
perception is changed: they partials,
become
theiraaural
just part perception
of a larger issonority.
changed:Furthermore,
they becomeasjust part
they doofnot
a bigger
conform sonority.
preciselyFurther-
to the
more, as series,
overtone they dotheynot render
conform theprecisely to the overtone
whole sonority series, they The
slightly inharmonic. render the iswhole
result that
the initial diatonic sonority (arguably very consonant) is changed into a new globally
sonority slightly inharmonic. The result is that the initial diatonic sonority (arguably
inharmonic sound entity. An analogy with a chemical reaction is almost irresistible:
very
A + Bconsonant)
→ C, where is changed into heterophony,
A is the folk a new globally B isinharmonic sound entity.
the added partial collectionAn and
analogy
C is
with a chemical
a new sonic entity. reaction is almost irresistible: A + B → C, where A is the folk heter-
ophony, B isnow
Turning theto
added partialsolution
the second collection and C isabove,
mentioned a new Isonic entity.a simple algorithm
developed
Turning now to solution 2 mentioned above, I developed a simple
to carry out and test the harmonic superimposition of the original algorithm
pitch to
collection
(of the folk melody) with a transposition given by an integer frequency ratio. The
carry out (and test) the harmonic superimposition of the original pitch collection (of
algorithm is shown in Figure 13. The results were approximated to semitones so as to
the folk
allow melody)
easier with aintransposition
playability the transposed given by an
parts. Theinteger-ratio
output was in frequency.sequence,
a harmonic The al-
gorithm
with eachischord
shownserving
in Figure
as a 13. The
pitch approximation
reservoir from which usedthewas the unfolds.
music semitone This so aswasto
allow easier
carried out byplayability
hand (see in the intervening
Figure parts. The
10 above, shaded output
parts) and was
can bea harmonic
seen as ansequence,
example
of mixed
with eachcomplementation:
chord serving as aitpitch involvevs manual
reservoir fromimplementation
which the music of unfolded.
algorithmic results
This was
(the harmonic progression serving as pitch reservoirs) carried out so as to coexist with
carried out by hand (see Figure
automatically-implemented 10 above,
algorithmic shaded
outputs (theparts) and can be heterophony).
rhythm-focusing said to create an
example of mixed complementation, involving manual implementation of algorithmic
results (the harmonic progression serving26 as pitch reservoirs) carried out so as to co-
exist with automatically-implemented algorithm outputs (the rhythm-focusing heter-
ophony).
Figure 13. Algorithm for the calculation of pitch reservoirs related by integer frequency ratios.
Proliferating melodies
Some time before working on A Vida é Nossa, I programmed an algorithm that would
generate a series of melodies by gradually developing an initial one. Each melody would
randomly retain a melodic fragment of the previous one and add notes at the end based
on its universe of intervals and durations (therefore stimulating cognition by means of
auditory memory). The number of notes to add each time was controlled by the Fibonacci
sequence. I developed and refined the algorithm, fed in a folk melody, and decided
to display the results from subsequent cumulative evaluations7 in a poly object, which
superimposes the several voices obtained (see Figures 14 and 15). This was initially
meant only to organise the results in a logical manner, but I could not resist playing it
back!
7 Usinga feature of omloop called “accumulation”, meaning that a process can be applied several times
by acting on the previous result each time.
27
Gonçalo Gato
Figure 14. The algorithm for the proliferation of a melody. See Figure 15 for the contents of
proliferate_mel_fibo.
Finding I was pleased with the aural results, I made the æsthetic decision to turn
this “accident” into one of the core techniques of the piece. It created a kind of het-
erophonic/canonic texture: imitations were created from the same melodic fragments on
different voices at different times. This also leads to heterophony if the onsets are close
enough.8 The superimposition creates harmony from the folk melody’s pitch set but also
chromaticises it because of the newly calculated pitches. In this sense, this procedure can
be classified as an algorithmic elaboration: it cumulatively proliferates melodic segments,
intervals, and durations into an n-part texture.
After several evaluations of this indeterministic algorithm, one solution was selected
or “cherry-picked” (Figure 16). One important characteristic, which was very noticeable
aurally, is that voice activity occurs by waves: increases and decreases of the local density
of short durations, separated by moments of relative relaxation through longer durations
in the form of tied notes. It also features a light and cantabile quality in the melodies
that occur from bar 4 onwards. I was struck by the ability of this algorithm to create
such melodic lines.
8 György Ligeti’s Lontano (1967) is a good example, as heterophony therein is also created by canons.
28
Figure 15. The inside of the abstraction proliferate_mel_fibo.
Different outputs from the melodic proliferation algorithm were used at various points
in the score. The technique thus created one of the main defining syntaxes of the piece.
The orchestrations that followed automatic implementation were carried out manually
(see Figure 17 for an example), along with some changes to improve playability and
musical clarity.
Algorithmic processes frequently result in elements that could be considered unmusi-
cal (or undesirable, for various reasons). In the case of this piece, I have had the chance
to hear it performed several times and could subsequently revise it to improve dynamic
balances among the voices and eliminate certain segments that sounded superfluous. For
me, these manual interventions play a key role in CAC, and show how I, as a composer,
am unwilling to sacrifice musicality even if that means carrying out “destructive” (but
musically constructive) actions on algorithmically generated materials.
29
Gonçalo Gato
Figure 16. Selected output from the patch in Figure 14. Rhythmic notation and accidentals
are shown in raw, unprocessed form. The brackets are used as visual aids to show retained
segments, which begin the next voice.
Important CAC-related reflections arose during and after the compositional process.
If algorithm design can be considered creative work and an integral part of composing—
as it naturally is—then it can allow space for much discovery to take place, without
being reduced merely to straightforward programming of a given technique or process.
It is the interaction between the algorithmic tools and the composer’s thought, mani-
fested through decisions, which makes CAC such a rewarding activity. Decisions restrict
possibilities, helping shape the workflow, but also make certain phenomena—like some
accidents—more likely to happen. This resonates with a famous statement by Pierre
Boulez [1]:
Un univers musical sans loi ne peut exister : il s’agit, sous un autre nom, de la
cohérence si chère à Webern; mais la loi seule ne permet pas à l’accident d’exister,
et prive ainsi la musique de la part la plus spontanée de ses moyens d’expression.
Furthermore, there is fundamentally no reason why algorithmic composition should

not feature, or deal with, the same æsthetic issues as purely manual composition. This
can mean reacting against the excessive rigidity of techniques and/or mechanicity of
their results, either through careful algorithm redesign, or through the use of subsequent
manual procedures (intervention, elaboration, or complementation) assisted by æsthetic
judgment.
30
? ‰ ‰ Œ ‰ Œ ∑ ∑ ∑ ∑
Tba.
¢ j j j j j j j ™ œ œ bœ ˙™
œ bœ œ œ œ œ œ œ nœ #œ œ #œ œ œ #œ œ #œ #œ ™
f n
3 3 3 3:2q
° 3
Vib. & Ó Œ Œ #œ Œ Œ ‰ Ó ##œœœœ œœœœ ‰##œœœœŒ Œ œœ Œ Ó
œ œ ##œœ
Ó
œ #œ##œœ
Œ Ó Œ #œ#œ #œ #œ ‰ Œ Ó #œ #œ #œ œ#œ ‰ Ó
b œœ œœ œ œœ œ œ #œœ mp
œœ
p mp
œ œ
4 Para a pandeireta
Mar. & ∑ ∑ w bw ‘ ‘ ‘ ˙ ˙ w bw ‘ ‘
W.B. / œ œœœ œ œ ™ œœ œ Œ ‰ œ œ œ‰
Folk material
Œ œ œœœ transformations and elaborations in A Vida é Nossa
œ œ œ œ œ œœœ œ œ œ œ œ œœœ œ œ œ œ Œ œ œœœ œ Œ œ œœœ œ œ ™ œ œ œ Œ ‰ œ œ œœ œ ‰ Œ ‰ œ œœœ œ ‰ Œ Ó
Glockenspiel Para Sinos Tub.
Sinos Tub.
¢& ∑ ∑ Ó Œ ‰ j ™ bœ œ œ7 ˙7 Ó ∑ ∑ ∑ ∑
œ œ œ œ œ œ™ J
pp p
=
A
22 . Animato (q = 120)
° ‰œ ‰Ó ∑ ∑ ∑ ∑ ∑ ∑ ∑ ∑
Picc. &
œ œ œ œ œ #œ #œ [ ]
. ‰ ‰ œ. ‰ ‰ bœ. œ œ ˙ œ ˙ Œ #œ œ #œ œ œœ œ œ œ‰
Fl. 1 & ‰ bœ ‰ Ó ∑ ∑
J J Œ Ó
p ff p ff cantabile
b œ. œ.
œ œ œ œ œ œ#œ œ œ œ#œ ™ ˙ #w œ™ #œ.
Fl. 2 & ‰ ‰‰ ‰ ∑ ∑ ∑ Œ œ œ™ J J J Ó
p ff p
œ œ œbœ œ- \ œ. œ. œ. œ
Ob. 1 & ∑ Ó Œ bœ œ™ bœ bœ bœ œ œ œ bœ nœ œ œ œ w ˙ œœ ‰Ó ∑ ‰ œ bœ
J
p pp ppp mf
j
Ob. 2 & ∑ Ó Œ bœ bœ œj bœ œ nœ œ™
J w ˙™ Œ ∑ #w œ™ œ œ œ#œ#œ w
J
p pp
œ œ bœ bœ ™ ˙ w ˙ nœ œ œ œ œ œ #œ
Cl. 1 & ∑ Ó Œ J Œ J #œ ™ w ˙ œ Œ ∑
p mp pp
Cl. 2 ∑ Ó Œ bœ œ w w ˙ œ Œ Ó Œ œœ œ ‰Œ Ó Ó Œ #œ œ œ œ #œ #œ ˙
&
p pp p
Cl. 3
¢& ∑ Ó Œ bœ œ bœ œ œ ˙ w ˙™ Œ ∑ ∑ ∑ ∑
p
1.
°
Tpa. 1,2 & ∑ ∑ ∑ ∑ Ó Œ œ. œ. ˙ œ œ œ œ œj œ™ ˙ w ‰ œœ œœ œœœœœœœœœœ
. . . . . . .
pp n pp
3.
∑ ∑ ∑ ∑ ∑ ∑ Ó j
Tpa. 3,4 & #˙ œ™ #œ œ œ-bœ#œ w
"Senhora do Almurtão" (trad. Portugal)
pp
con sord. [ > ]
Tpt. 1 & ∑ Ó Œ œ. œ. b˙ œ œ bœ bœ œj b>œ œbœbœ œ w w ∑ ∑ ∑
J
mf cantabile p
con sord.
Tpt. 2 & ∑ Ó Œ œ. bœ ˙ bœ bœ bœ œ bœ œnœ ˙ w ˙™ Œ ∑ ∑ ∑
p
con sord.
Tpt. 3
¢& ∑ ∑ ∑ ∑ Œ ‰ œ œ œ bœ bœ w w ∑ ∑
p
Para os W.B.
~~~~~~~~~~~~ l.v.
° œ bœ
‰œ œ œ
3 œ #œnœ œ œ Ÿ˙ œ ‰ #œ#œ ‰ Ó ‰#œ#œ#œ œ œ œ
> l.v.
‰ nœbœbœ œ ‰ Ó
Vib. & #œœ œœ #œ #œ Œ nœbœbœbœ œ œ œ œ œ œ œ œ œ œ œ œ bœbœ œbœ œ œ œ œ œ œ œ œ œ œ‰ ∑ œ œ #œ#œ œ#œ#œ œ#œ
p 5
#œ
3 5
pp p pp mp pp pp
° Pandeireta
ø ø ø
Mar. & ∑ ∑ / œ œ œœ œ œ œœœ ‰ œœœ Œ œ ∑ ∑ ∑ ∑ ∑
pp
Bombo
Caixa Para o Bombo
W.B. / œ œ œœœ œœœ œœœ œ ‰ Œ Ó A*w
— ∑ w7 ∑ ∑ ∑ ∑
pp p
Sinos Tubulares 3 3 Para o Prato Susp.
Glock.
¢& ∑ ∑ bw ˙ bœ ˙ ˙ œ b˙ w ∑ ∑ ∑
mp pp
Figure 17. Score excerpt showing the implementation of the output (boxed). Notated in C.
Colouring melodies
I have always admired the way Olivier Messiaen colours melodies, whether they are
from birdsong, plainsong, or of his own composition. Béla Bartók was one of the first
composers to take this influence from Claude Debussy, as Malcolm Gillies observes about
the Bagatelles, Op. 6 (1908):
[...] streams of parallel 5ths and 7ths (no. 4), of 4ths (no. 11), of tritones (no. 8),
or of piled-up 3rds (nos. 7, 9, 10) [...] The influence of Debussy, about whose
works Bartók had recently learnt from Kodály, also lies behind several of the pieces,
notably in the use of parallel chords, and in no. 3, with its unchanging semitonal
ostinato. [6]
Stravinsky was certainly influenced in this way by Debussy too. The way Messiaen
approaches this technique is frequently more dynamic; the timbres change as the melody
progresses:
When I reproduce a birdsong, every note is provided with a chord, not a classified
chord but a complex of sounds which is designed to give the note its timbre. [3]
31
Gonçalo Gato
A particularly good and simple example is the slow first movement of Éclairs sur l’Au-
Delà: Apparition du Christ glorieux (1988-91). As the pace is slow, one can “taste” each
chord as either timbre or harmony. The fast and staggering third movement “L’oiseau-
lyre et la Ville-fiancée” shows, on the other hand, the same technique at work in a more
virtuosic manner where timbral perception dominates aurally.
I wanted to automate the colouring of melodies according to a given harmonic pro-
gression (or timbre evolution). A melody and a chord progression of the same size would
be input, plus a function (bpf) controlling the positions that the melody’s pitches would
take inside each chord (anything between the lowest and highest pitches). This algorithm
enabled me to test the technique with various melodies, timbral evolutions, and registral
evolutions. The OpenMusic patch is shown in Figures 18 and 19.
Figure 18. Patch used for melodic colouring. Timbral evolution is shown here as an
interpolation from a chord to a unison in 12 steps.
The output, used to build the main climax of A Vida é Nossa, is shown in Figure 20.
As we can see, sometimes the added pitches are higher than the melody, sometimes lower.
This was controlled by the bpf contour.
This algorithm is applied in other instances throughout the piece. For example at
the beginning the low brass melody (Figure 4) is coloured in a similar way, but this
time by using timbres that are subsets of the harmonic series. I have used this kind
of procedure in the past,9 calling it spectral enrichment. This aspect of modularity in
algorithmic composition is well documented and very appealing to me as it means that the
composer can easily create variations on processes by merely patching together different
9 In Vectorial-modular (2011) for instance, an analysis of which can be found in [5].
32
Figure 19. The abstraction melody_timbre_with_bpf (left) and its internal loop (right).
combinations of a limited set of subroutines. Accordingly, a process of calculating

harmonic partials was used and connected to the same melody_timbre_with_bpf patch
used before. The resulting algorithm is shown in Figure 21, and the output in Figure 22.
The implementation of the algorithm was automatic, but received subsequent manual
interventions and elaborations. In Figure 23 one can see the respective excerpt from the
final score. Interventions included doubling the durations and discarding the first part
of the algorithmic harmonisation. Subsequent elaborations comprised octave doubling of
the lowest voice plus reharmonisation of the beginning. The reason for this reharmonisa-
tion was that I wanted a denser sonority to follow the timpani crescendo just before (see
Figure 1 above). This æsthetic decision draws attention to the importance of musical
context in CAC.
33
The output, used to build the main climax of the piece, is shown in Figure 20. As
Gonçalo Gato we can see, sometimes the added pitches are higher than the melody’s pitches, some-
times lower. This was controlled by the BPF contour.
Initial melody
Figure20.
Figure 20. Melodic
Melodic colouring
colouring used
used in
in the
the main
mainclimax
climaxofofthe
thepiece.
piece.
Other instances of the application of this algorithm exist throughout the piece.
For example, just at the beginning, the low brass melody (Figure 4) is coloured in a
similar way but this time by using timbres that are subsets of the harmonic series. I’ve
used this kind of procedure in the past,11 calling it spectral enrichment. This modular-
ity aspect of algorithmic composition is well documented and very appealing to me as
it means that the composer can easily create variations on processes by merely patch-
ing together different combinations of a limited set of subroutines. Accordingly, a
process of calculating harmonic partials was used and connected to the same ‘melo-
dy_timbre_with_bpf’ patch used before. The resulting algorithm is shown in Figure
21, and the output in Figure 22.
The implementation of the algorithm output was automatic but suffered
subsequent manual interventions and elaborations. In Figure 23 one can see the
respective excerpt from the final score. Interventions such as durations doubling and
discarding the first part of the algorithmic harmonization were carried out. Octave
doubling of the lowest voice plus re-harmonization of the first part constituted subse-
quent elaborations. The reason for re-harmonizing was that I wanted a denser sonority
to follow the timpani crescendo (see Figure 1 above). This aesthetic decision draws
attention to the importance of musical context in computer-aided composition.
11
In the piece Vectorial-modular (2011) for instance, an analysis of which can be found at the EarReader
website (Gato 2013).
Figure 21. Spectral enrichment patch.
34
Figure 22. Spectral enrichment of a melody. The initial melody is shown on the bottom staff,
with odd-number partials added: 3, 5, 7, 9, 11, 13 and 15.
Figure 23. Score implementation of the algorithm shown in Figure 22 (in C).
Conclusion
This chapter worked as a journey through various compositional procedures used to
compose A Vida é Nossa. Various elaborations and transformations of folk material
were analysed and different types of score implementation were discussed, along with
manual procedures.
By giving special care to the use of appropriate terminology, I hope to have clarified
the procedural multiplicity of the manual/machine interface in CAC. In fact, general
æsthetic issues of compositional craft are found to be omnipresent, always informing
decisions and thus bringing CAC very close to pure manual composition. This intellectual
framework allows for an increased scope of actions in procedures such as algorithm
redesign and manual intervention, elaboration or complementation.
35
Gonçalo Gato
References
[1] Pierre Boulez. “Le système et l’idée”. In InHarmoniques, 1. Le temps des mutations.
Paris: IRCAM-Centre Pompidou/Christian Bourgois Editeur, 1986.
[2] Tim Howell. After Sibelius: Studies in Finnish Music, chapter: “Magnus Lindberg –
Rediscovering Balance”. Aldershot: Ashgate, 2006.
[3] Olivier Messiaen. Musique et couleur: Nouveaux entretiens avec Claude Samuel.
Paris: P. Belfond, 1986.
[4] Curtis Roads. The Computer Music Tutorial. Cambridge, MA: MIT Press, 1996.
Online
[5] Gonçalo Gato. “Vectorial Harmony”. The Ear Reader, 2013.

http://earreader.nl/archives/628.
[6] Malcolm Gillies. “Bartók, Béla (1881 - 1945), composer, ethnomusicologist, pianist”.
In Grove Music Online. Oxford University Press, 2001.
http://oxfordindex.oup.com/view/10.1093/gmo/9781561592630.article.40686.
[7] François Lesure, Roy Howat. “Debussy, Claude (1862 - 1918), composer”. In Grove
Music Online. Oxford University Press, 2001.
http://oxfordindex.oup.com/view/10.1093/gmo/9781561592630.article.07353.
Acknowledgements: I wish to thank my doctoral supervisor Julian Anderson and co-supervisor Paul
Newland. My first OpenMusic steps were taken with Carlos Caires from the Escola Superior de Música
de Lisboa, who introduced me to the world of Computer-Aided Composition, and so deserves a special
mention and thanks. Also, I wish to thank my first composition teacher Eurico Carrapatoso for showing
me with remarkable elevation and elegance how composing always goes hand in hand with clear thinking.
This research was funded by a doctoral grant from the Fundação para a Ciência e Tecnologia (FCT),
Portugal, with funds from the POPH (Programa Operacional Potencial Humano) and Fundo Social
Europeu (European Union). I have also benefited from scholarships from the City of London Corporation
and the Guildhall School of Music and Drama, United Kingdom.
36
Tri Bhuwana for 12 voices and
Balinese gamelan:
Three worlds/tripartite world
Philippe Boivin
Bottom up, from essence to diversity, the TRI BHUWANA three

worlds are: BHUR, BUWAH and SWAH. In the Hindu-Buddhist
cosmology of Bali, they are present in every existing structure
since there is identity between the whole and its parts.
Kati Basset
Following my appointment to Aix-Marseille University, two musical encounters led me to

elaborate the project to compose Tri Bhuwana for 12 vocal soloists and gamelan. First,
there was the discovery that Marseille possesses a gamelan, the Gong (Kebyar) Bintang-
Tiga (see Figure 1), a fascinating Balinese group of instruments, under the management
of percussionist Gaston Sylvestre, a leading light in 20th century contemporary music.
Then Roland Hayrabedian, artistic director of the vocal and instrumental ensemble
Musicatreize, suggested that I take part in the Space Odyssey project being set up as part
of Marseille-Provence, European Capital of Culture 2013. Its objective is to encourage
exchanges between amateur and professional musicians through a policy of commissioning
new works. Both ensembles share a will to stimulate new musical compositions by
inventing pathways that get round the traditional barriers between different audiences.
So it seemed right to include both groups of musicians in the same project for a new
composition: Tri Bhuwana for 12 voices and Balinese gamelan [6].
The research programme I have developed on OpenMusic allowed me to model
traditional Balinese music textures in order to create a new material that could solve
the antagonism between equal temperament scales (for the Western choir) and relative
tuning of the gamelan based on vibration (frequency of gong beats).1 The algorithmic
nature of Balinese music, the concentric overlapping of temporal strata, together with
the specific constraints of the collective body playing the gamelan, have been efficient
models to formalise the writing of vocal parts. The models’ consistencies stem from the
holistic nature of the project. This chapter will detail the steps of the questioning inspired
by distant cultural and geographical time-space models to open new paths, respecting
tradition better to deviate from it.
1 Mostof the functions used in the patches presented in this chapter have been programmed as part of
my own OpenMusic library.
37
Philippe Boivin
Figure 1. Gamelan Bintang Tiga in concert on the beach of l’Estaque (Marseille).
Systemic modes
My compositional procedures are all based on the same pattern generation system. The
core of this system uses five basic modes developing various probabilities of repetition of
any two elements a and b. A mode starts from two or more basic cells (small sequences
of a and b) and combines them randomly to create longer sequences (see Figure 2). This
random generation allows one to obtain an infinity of Markov-like chains while conserving
the proprieties of each mode.
Figure 2. Systemic modes: generation.
In order to enrich this sequence generation system I use a process of successive

ramifications from either “side” of the sequence (that is, from the sub-sequence of either
a or b). In Figure 3, the terms of the sequence created through mode 2 (level 1) are used
to replace its own b elements (level 2). Note that a different sequence might have been
generated using another sequence of the same mode, or a sequence of another mode.
Free transcoding of the elements of the successive ramification levels can then easily
be done to obtain a new chain of three or more elements that may be used for rhythmic or
melodic purposes (Figure 4). In reality, such transcoding is bound to more sophisticated
algorithms implementing strict musical rules.
38
Tri Bhuwana for 12 voices and Balinese gamelan: Three worlds/tripartite world
Figure 3. Systemic modes: ramifications.
Figure 4. Systemic modes: transcoding.
Rhythmic generation
In Bali’s traditional music, a musical piece
can be recognised by its general musical
outline, differently treated in each melodic
part of the collective instrument. The
parts roughly correspond to pitch ranges
that are linked to different treatments of
time: long values in low strata, short
ones in high strata (stroke frequency
multiplied by 2 and by powers of 2).
Ethnomusicologist Kati Basset’s concen-
tric notation of the Dhongdhing system
(Figure 5) illustrates this principle of
binary multiplication of stroke frequency
in the different strata, starting from the
centre [2].
I developed a function called gam-ryt,
Figure 5. Dhongdhing system in concentric
which reproduces the principle of nested
notation (Kati Basset 2009).
stroke levels while adapting them to Tri
Bhuwana’s specific musical context. In Figure 6, this function uses three basic durations:
quarter note, half note, and dotted half note to generate four voices containing slower
and slower stroke levels.
The <RYT-mode> argument enables systemic modes finely to articulate alternation
between short values and long ones (1 vs. 2/3), as well as distribution between binary
(2) and ternary (3) long values. In this example, the same mode (2) is used for both
operations. The list of durations thus obtained is then restructured by an algorithm
regrouping either the single short values with the preceding long one or the succeeding
short values only, up to a maximum of two quarter notes.
39
Philippe Boivin
Figure 6. Stroke levels.
In Figure 6 the sequence generated by the systemic modes is:

(2 1 3 1 1 2 3 1 3 1 1 1)
The grouping steps are then:
((2 1) (3 1 1) (2 3 1) (3 1 1 1))
((2 1) (3 (1 1)) (2 (3 1)) ((3 1) (1 1))
These steps provide the structure of voices 1 and 2. From the nested list is then deduced
the hierarchy of stroke levels (voices 3 and 4):
((2 1) (3 (1 1)) (2 (3 1)) ((3 1) (1 1)))
((3) (3 2) (2 4) (4 2))
(3 5 6 6)
The returned rhythmic texture is radically different from that of Balinese structures,
which are founded on strictly binary ramifications. However, the musicians readily
adapted to it, as it is respectful of both the individual role of each member of the
gamelan and the general sound balance to which they are accustomed.
Melodic generation
The Bintang Tiga gamelan plays along a pelog 2 melodic scale of five notes corresponding,
given a few corrections, to the Phrygian mode transposed to C]. Clearly, however, the
2 One of the two basic scales of gamelan music.
40
notion of pitch is quite relative in Balinese music. The tuning of the instruments may
greatly vary, not only from one village to another, but also among different instruments
within a gamelan, or even from one octave to another on a given instrument. Besides,
all blade instruments (for example the gender) are played in pairs, matched to produce
beats whenever a single note is played by two musicians at the same time (Figure 7).
Figure 7. Pelog scale used by Bintang Tiga gamelan.
This is a musical world that is not equally tempered, but extremely rich; vibration
prevails over the principle of absolute pitch founding our present Western system. The
spectral analysis tools available in OpenMusic have proved particularly useful to study
these subtleties, often quite difficult to hear (Figure 8).
Figure 8. Spectral analyses.
As with rhythmic generation and with a view to keeping a coherent approach, systemic
modes were used again, this time to generate melodies on the five notes of the pelog
scale. An algorithm controlling interval succession (Figure 9) ensures melodic outline
and colour, while leaving out over-connoted “tonal” phrases. This algorithm first controls
alternation between seconds and other types of intervals, then selection between thirds,
fourths and fifths. The gam-mel function is able to generate melodies of <tot> notes
whose outline may be modified by shifting the <low> inferior limit not to be overstepped.
Moreover, it also cuts this melody into several small loops to the return of the <start>
note.
41
Philippe Boivin
Figure 9. Melodic loop generation.
This operation of segmentation permits the endless repetition of the same loop or
the free linking of several corresponding loops together. As it happens, the principle of
repetition of more or less long cycles, punctuated by low gongs, is one of the bases of
Balinese traditional music. Moreover, redundant melodic formulas can be avoided within
a melody that may appear slightly too long as it is limited to a combination of the five
notes of the scale. Figure 10 shows how melodic-rhythmic blocks associating loops and
various stroke levels can now be built.
Figure 10. Melodic-rhythmic blocks.
42
As may be seen in Figure 10, every stroke level finds its pitch material in the main
melody, whose corresponding notes are sent down into the lower octaves according to the
range of the selected instrument. The group of four gongs playing the block’s last stave
is the only one to use its own pitches, distributed in relation to the number of strokes in
the sequence according to an alternation principle called colotomy: Wadon (0) is struck
only once at the start of the sequence, Lanang (1) once in the middle, Kempur (2) and
Kletong (3) are distributed on the remaining strokes (see Figure 11).
Figure 11. Gong colotomy relative to number of strokes.
In this example, the first loop returns a balanced sway visible in the third voice
that is distributed across 10 beats (3+3+2+2) and the second one a resolutely ternary
structure (3+3), two rhythmic morphologies that can never be found in Bali’s traditional
music. Even so, these novel textures fully respect this music’s usual acoustic features in
a renewed context.
What is also significant is that the rhythmic structure of the stroke levels is similar
in the two loops, although it is shortened in the second one (Figure 12). This constitutes
a basic principle to develop material with limited means in Tri Bhuwana. Rhythmic
repetition introduces organic coherence in the melodic variation brought when passing
from one loop to the other. In the following example, the three stroke levels are shown
in line so that the loops may be compared.
Figure 12. Truncated rhythmic structures.
In addition, rhythmic variations enable the building of a new type of block, keeping
the same melodic loop. In Figure 13, at the marking RytSel.2, the regular flow of the
main melody is interrupted to give it an irregular rhythm (the rhythm of the second
stroke level played by the Calung). The resulting time dilatation in the melody, as well
as the shift of meeting points, deeply change the perception of the initial object to reveal
new aspects. Tri Bhuwana constantly plays on this dual repetition/variation concept
thanks to the modification of a few generation parameters. This allows the development
of long and non-redundant sequences from a few notes only.
43
Philippe Boivin
Figure 13. Succession of loops and rhythmic profiles.
Proliferations
A particular playing technique called kotekan makes it possible to develop new figures,
sometimes very fast, between two successive sounds, whatever the stroke level used.
Thanks to a collection of orally transmitted formulas, Bali’s musicians know a great
many ways to pass from one note to another logically, by filling time and space at a
given speed. The four main kotekan families used in traditional music (Nyok-cag, Norot,
Telu and Empat) were analysed at length and then modelled by algorithms that are able
not only to reproduce their operating modes, but also to generate new ones.
Figure 14 first illustrates the principle of internal expansion of kotekan Nyok-Cag.
Expansion exp.2 inserts an intermediate note after each pitch of the mel melody, accord-
ing to rules based on the observation of traditional practice. It usually is a neighbour
note, superior or inferior to the expected following note. Taking exp.2 as a basis, exp.4
reproduces the same principle, and so on, each time doubling execution speed. It is
interesting to note that the algorithm tends to get more complex with the iterations in
order to improve the plasticity of the melodic lines. For instance, the choice of a simple
treatment (marked ) 1 or of a note “enjambment” ( ) 2 may be made according to a set
of cumulative rules: redundancies to be avoided, an ambitus to be respected, a number
of different notes not to be exceeded, etc.
Figure 14. Kotekan Nyok-Cag: internal expansions.
The very long resonance of the gamelan instruments must be taken into consideration.
The usual way to play the blade instruments is by muting the note just played, with one
hand, at the same time as the other hand strikes a new one with a mallet or a small
leather-covered wooden hammer. So as to play fast, there must be two musicians, one
completing the other’s part, sometimes with impressive velocity. As shown on Figure 15,
the technique of distribution of such melodic garlands between two players called Polos
and Sangsih is much simpler to write down than to apply, at least for Western musicians.
44
Figure 15. Kotekan Nyok-Cag: distribution between two musicians.
Kotekan Norot is much simpler to program, as there is only one option to set it.
First, developed from the starting note, is a sequence of superior neighbour notes that
ends three impulses before the expected following note. The latter is then struck twice
and followed by its own superior neighbour note (see Figure 16). So the predictability
of the formula is very high, since the expected note is always announced; moreover, it is
accentuated through being doubled, as shown in the lower staff of Figure 16. This type
of kotekan is particularly interesting as it can run at slow speeds as well as fast ones, but
also with possible chordal versions, such as have been developed in a novel way in the
composition of Tri Bhuwana.
Figure 16. Kotekan Norot: expansion and distribution.
Kotekan Telu is a clever mechanism: a cell of three adjacent notes of the pelog scale
is repeated several times in such a way as to obtain a succession that necessarily ends on
the expected following note (see Figure 17). The modelling process has proved extremely
complicated, due to a number of qualitative parameters to be dealt with carefully: full
melodic analysis must be carried out before selecting the cells in order to hold the same
notes as long as possible; the start of the process must be set accurately so as to retain
elegance (see additional parenthesised pitches in Figure 17); the cells must be oriented
along the melodic line, etc. The distribution between the two musicians is remarkable in
terms of both instrumental playing and sound effect: only two in three notes are played
by each, the central one always being doubled, so naturally accentuated.
Kotekan Empat is a type of extension of kotekan Telu, with the addition of a higher
fourth adjacent note. As shown in Figure 18, the additional pitch is set above the lowest
note of the three-note cells. The accentuation resulting from the simultaneous strokes
then creates a new “virtual” rhythm typical of these interlocking structures [5]. Just
making Sangsih play the two superior sounds and Polos the two inferior ones will no
doubt ensure velocity and fluidity, provided that they get along well!
These diverse techniques of expansion inspired from traditional practice can be mixed
at will to create varied and renewed polyphonic textures. Figure 19 shows an alternation
45
Philippe Boivin
Figure 17. Kotekan Telu: irregular expansion and distribution.
Figure 18. Kotekan Empat: addition of a fourth sound to kotekan Telu.
of three different kotekan, as tested in a study preceding the composition of Tri Bhuwana.
Here, both kotekan Nyok-Cag and kotekan Empat follow the continuous melody in regular
quarter notes of the first stroke level played by the Panyacah (*1). The kotekan Norot
of the last two measures is adjusted on the airier rhythmic line of the second stroke level
handled by the Calung (*2).
Figure 19. Study of alternation of three types of kotekan.
46
I was led to design and program many

functions in OpenMusic, which made it
possible for any type of processing to be
applied to any type of voice by selecting
relevant parameters (algorithm, stroke level,
speed, etc.). During the actual phase of
composition of Tri Bhuwana, several high-
level parameters were generated by the
work’s macrostructure itself, based on a set
of permutations of five elements (Figure 20).
This structure was interpreted to derive
musical materials using the main function
gam-gongan shown in Figure 21. Figure 20. Tri Bhuwana’s macrostructure.
Figure 21. Parameters generated by the macrostructure. The gam-gongan function outputs
the rhythmic structures, pitch values, as well as other parameters such as the tempo, modal
scale, link note with the next section, etc.
47
Philippe Boivin
Using this method and as shown on Figure 21, the validity of more and more complex
combinations could then be checked, which would have been difficult without the capacity
to listen almost instantaneously to the instruments sampled (a sound bank of over 200
samples I recorded blade by blade with several playing modes). This illustrates that, by
meeting the specific needs of a given work, Lisp programming can be an integral part
of the creative process of a composer, as it has indeed been for me over the past forty
years. It is then time for pencil and rubber to refine, develop, reorient, and even delete
the results returned.
Gamelan and voices

Given the novel nature of this project and the lack of previous references to support
it, finding a vocal material likely to fit in the complex world of the gamelan and its
unequal temperament was a challenge. The first path explored was that of microtonal
scales developed from spectral analyses of combinations of gamelan sounds, as shown in
Figure 22.
Figure 22. Harmonic fields developed from spectral analyses.
This method, though appealing at first sight, was soon left aside for several reasons,
among which uncertainty as to the ability of singers to hold such intonations against the
unsettling background of the gamelan.
A more realistic approach gradually stemmed from the many simulations carried out
with synthetic sounds. A specific modal system seemed to allow the choir to sing in its
own equal temperament, taking as a reference diapason the low C] of the large Wadon
gong (427.5 Hz). The idea was to have the two sound spheres coexist, the gamelan
creating an impressive broadening of acoustic space for the voices in some cases, and
vice-versa in other cases.
In order to provide frequency areas that are common to both ensembles, the modal
scales used include all the sounds of the pelog scale. As shown in Figure 23, each of these
sounds can be paired with one of the five sounds of the pentatonic scale of A transposed
to C, which forms a second level for the generation of the modal scales. The two missing
sounds F] and B make up a third level, which completes the total chromatic range.
Each section of Tri Bhuwana is always associated with a polar note, which is both a
starting point and a link between the different loops used in the sequence. The section
features a modal scale of nine or ten sounds built in three stages, considering the polar
note of the section and that of the next section as shown in Figure 24. To the pelog scale
are added the sounds from the pentatonic scale not associated with the two polar notes,
then the possible insertion of a mobile degree with a view to filling the intervals superior
to major second.
48
Figure 23. Augmented pelog: building elements.
Figure 24. Augmented pelog using notes 2 and 3 as polar notes (development in three stages).
The fifteen scales used in this opus are all built on this single unifying principle so
that through the interplay of repetitions, transformations and loop strings, each section
asserts its own modal colour. Besides, transitions from one section to the other are made
in a gradual way, through the linking of the polar notes—those of the current section
and of the next one—used to set the different scales. As shown in Figure 25, only A]
turns into B when one goes from mode 3-3 to mode 3-4.
Figure 25. Augmented pelog: the 15 scales of Tri Bhuwana. (The numbers indicate the
positions of the two polar notes for each scale.)
From a set of four pelog loops and three modal loops corresponding to section C1 of
the piece, Figure 26 shows how to isolate a melody (A), then redistribute it as a vocal
quartet in an expanded spectral layout (B), and filter it to let the gamelan punctuate it
on the pelog pitches it contains (C).
49
Philippe Boivin
Figure 26. Material of a section of Tri Bhuwana.
Modal loops 1 and 3 actually use defective forms of the ten-sound mode, whose
complete version is only displayed in loop 2 (see Figure 27). This option makes it
easy for several intermediate modal colours to be found within a given section. Even
if the generation system is completely automated and linked to the macrostructure of
the piece, such subtleties clearly highlight the total liberty left to the composer to handle
these various materials according to his own inspiration, so as to meet musical needs of
a superior nature.
Figure 27. Defective scales within a given section.
Moreover, the capacity to explore in depth the qualitative aspect of vocal generation
while conserving an organic link with the gamelan offers the last but not least advantage
of returning a material that is immediately ready to use. Just like the entirely modelled
kotekan techniques, this material may in turn be subjected to very sophisticated algorith-
50
mic processing, this time adapted to vocal writing: for instance, modal transpositions,
recurrences and inversions, concatenations, imitations, harmonic overlaps, rest control,
and so on. These functions make up an extensive library. They may be applicable to a
group of voices as well as to isolated voices. The flexibility of graphic programming—the
capacity to encapsulate several generation levels and to gradually refine the returned
results—makes it possible to build draft sketches that can generate musical sequences
lasting several minutes, as shown in Figure 28.
Figure 28. Draft sketch of full section.
Beyond technology
This project has been a step forward in my understanding of musical languages, first and
foremost of my own. But I must now come to the heart of the matter.
It took several months for me to get attuned to the world of the gamelan [3]. I actually
had to learn how to write for this collective body, whose operating mode was utterly
unknown to me, being so far from that of a Western orchestra that is mainly focused on
the individual. While collecting relevant literature [4] and developing the architecture
of the Lisp program at the same time, I produced several musical sketches that were
immediately tested by the musicians. This to-and-fro process allowed me gradually
to grasp the usual practices, then much rarer ones sometimes reserved for initiated
musicians, for instance the art of Kendang or of Gender. Only then could I start designing
my own devices of composition for the gamelan.
However, the true question came up later when it came to choosing the title and the
texts with Kati Basset, who helped me chart my own course thanks to her judicious
51
Philippe Boivin
advice [1, 2]. She reminded me that the practice of the gamelan was first and foremost a
ritual and social gesture, far from the concept of art or pure aesthetics defining Western
concerts where music has its own intrinsic value. From the start, the work was conceived
as a triptych combining operating forces. The Balinese cosmology founding Tri Bhuwana
has been a means to give deeper sense to the speculative choices made at the beginning
of the project.
SWAH is the world of essences, principles and oneness, the origin of all things.
BUWAH is that of form and objects perceived by our senses in their everyday use.
As for BHUR, it is the world of lower reality, devoid of spirituality and tending
towards shapeless disintegration.
Kati Basset
In Tri Bhuwana, each of the three worlds is here associated with such diverse texts as
poems about meditation, knowledge, or death in Swah, verse by Jean de Baïf or François
Villon in Bhur, and even a half-genuine, half-mock version of a University governance
report for Bhuwah. Beyond the technical considerations described in this chapter, my
ability to give musical life to the basic principles of this tripartite world became the
main stake of this project. The leads were manifold and could induce treatments of
the material that I would never have imagined without the support of this cosmological
guideline. That is how willingly unfinished musical features such as armless kotekan or
headless melodies quite naturally found their way into the score, echoing the Bhur infra-
world, peopled as it is of Buta-Kala with incomplete bodies, as illustrated in Figure 29.
Figure 29. Bhur world: Buta kala with incomplete bodies.
A last example sheds light on an original way of envisioning the text/music relation-
ship from an algorithmic outlook. It is the crossing of a text that is sacred in Java and
Bali, the “Syllabary of consonants” (visible on the left-hand side of Figure 30), with the
vibratory matter of the aksara sound-letters [e u a i o] used by Balinese musicians to sing
the names of the five notes of the pelog scale. The two voices shown in Figure 30 recite
the syllabic text which is associated with two modal transpositions of these vowels. A
new text appears, whose meaning is in no way altered by the shift effected: Once upon
a time, there were two emissaries of like power. They killed each other at the top. It is
worth noting that due to the strict logic of the device applied, the two sopranos felt very
comfortable singing this completely invented “Balinese” language.
52
Figure 30. Vowels on the Syllabary of consonants.
Clearly, such a long-lasting immersion in the world of Tri Bhuwana has opened my
mind to new dimensions, in particular a broader relationship to time, together with a
more physical approach to sound vibration. As I was proceeding in the writing of the
piece, I felt a deep drive to let go, to forsake the frantic nature of some of my composer
habits. Indeed, tirelessly repeating the five evocations of the name of Shiva so as to
elaborate a complete sequence (as in Section B2 of the piece) cannot be insignificant,
even with the use of variation. In this way, Tri Bhuwana will no doubt leave its deep
and long-lasting mark on my subsequent work, as was already clear in the pieces that
followed it. Even if Revenantes, a piece for female voices and cello, seems far remote
from anything Balinese, this opus composed just after Tri Bhuwana is still deeply under
its influence [7]. I have no doubt that from the spiritual experience I underwent with
Tri Bhuwana has stemmed a genuine renewal of my musical language and composition
methods.
53
Philippe Boivin
References
[1] Catherine Basset. “Au delà des apparences. Morphologie des esthétiques et cosmologie
à Bali”. Le Banian, 9, 2010.
[2] Catherine Basset. “L’univers du gamelan: opposition théorique et unicité

fondamentale”. Archipel, 79, 2010.
[3] Edward Herbst. Voices in Bali: Energies and Perceptions in Vocal Music and Dance
Theater. Hanover, NH: University Press of New England, 1997.
[4] Michael Tenzer. Gamelan Gong Kebyar: The Art of Twentieth-Century Balinese
Music. Chicago: The University of Chicago Press, 2000.
[5] Wayne Vitale. “Kotekan: The technique of interlocking parts in Balinese music”.
Balungan, IV(2), 1990.
Online
[6] “Tri Bhuwana for gamelan and voices, creation in 2013”. Project presentation video
(6’32), 2013. http://www.musicatreize.org/fr/tri-bhuwana-mpx16.html.
[7] Philippe Boivin. “Choir and gamelan / Choir in prison”. Lecture recorded at IRCAM,
2015. http://medias.ircam.fr/x21af85.
Acknowledgements: The writing of this article has benefited from a work residence at the Villa Médicis
French Academy in Rome in August 2015. I am deeply grateful to Kati Basset for her expert advice and
unflinching support throughout the long musical and spiritual progress inherent to this project. I would
also like to thank Nicole Mausset for her translations and rereading of this text.
54
Programming modular progressions
in OpenMusic
Matthew Lane
This text presents the author’s modular approach to designing and using OpenMusic
patches to produce multi-dimensional progressions in instrumental music.1 Musical
passages (in one or several instruments) are passed from process to process with tailored
parameters for each process and instrument, much like an assembly line working on
a single base material. Individual progressions gradually change the passage in each
instrument, altering combinations of pitch, time, structure, and stochastic elements,
before passing the result to the next progression. The chapter introduces the overlying
organisational structure, as well as the justification for this system, and guidelines for
moving between these patches and composing by hand. Examples that make use of these
patches are presented in four compositions: Sliding Apart, Melodious Viscosity, and two
movements from Short Pieces on Falling.
Introduction
Computer-assisted composition (CAC), and specifically the use of OpenMusic, is es-
pecially prevalent in my music related to the development of progressions. I define a
progression as a directed alteration of elements of a musical passage over time. This
applies both to continuous progressions (one long line that continues to change), and
sequences (progressions where the same idea is repeated while evolving at each instance).
Sequences are essentially changing repetitions, and have long been in use, but have until
the 20th Century been mostly confined to diatonic (and sometime chromatic) modulations
of the base element (model). In the 20th Century, especially with the music of Olivier
Messiaen, rhythm began to change in a formalised way.2 Eventually, every element of
music became a possible parameter for progression, including pitch, rhythm, articulation,
timbre, speed, and all their subcategories.
My compositional process could be summarised as follows: creation of an original
idea (seed), creation of related ideas, triage of ideas, development of remaining ideas,
interpretation/reinterpretation of developed ideas where necessary, formal organisation,
1 Thischapter is an extended version of a paper presented at the Journées d’Informatique Musicale

conference in Montréal in May, 2015 [2].
2 Thatis, in a gradual and clearly directed way. It is true that rhythms in Bach, for example, occasionally
change during sequences, but almost never in a consistent, directional, and quantifiable way. For more
on Messaien’s rythmic processes see The Technique of My Musical Language, chapters III-IV [3].
55
Matthew Lane
and linking. I have used CAC in nearly all of these processes, but the present focus will
be on development. Developing different parts of a piece using the same OpenMusic
patches, with the distinct traces and colours they leave on the music, helps to create
unity between sections and to shape the æsthetic of the piece.
This text will refer throughout to four pieces: Melodious Viscosity (2013) for wind
quintet; Sliding Apart (2013) for flute, clarinet, piano, violin, and cello; and movements
“Rings” and “Float” from Short Pieces on Falling (2015) for flute, clarinet, piano, violin,
and cello. The latter in particular is an exploratory set of over a dozen small pieces
to examine which uses of computer-assisted composition I hope to extrapolate in larger
projects over the next few years; the two selected movements represent the compositional
processes I found the most appealing.
Theory and programming

Programming structure
Because progressions and sequences invariably play a role in every one of my pieces, to
remain useful and portable, patches have to be flexible and rearrangeable: they require
sufficient parameters for control and variety, but with a simplicity to render them efficient.
For this reason I use modular patches, each one for a different treatment, applied in
sequence. In some cases, order of application makes no difference, such as a patch that
progressively changes the pitch of all the notes and another that removes half the notes.
But in other cases order is crucial, such as a patch that stretches selected durations and
another that randomises attacks based on temporal location. The system is summarised
in Figure 1. The musical passage is passed downwards through several “applicators”,
each applying a given function to the score objects with given parameters.3 This process
can involve numerous applicator/function pairs.
Progressions can be organised continuously or in discrete segments. A continuous
progression is like a glissando, where a treatment is progressively applied over time or
over a number of notes to an entire passage, with no regard for where notes fall in phrases
or musical cells within the whole. Discrete progressions, however, can be thought of as a
staircase where processes are applied to segments of music in order to form a progression;
within individual segments (on each step of the staircase) the progression remains static
or “pauses”. Figure 2 shows how the results of a simple transposition differ between
these two types of progression. Essentially, discrete progressions allow one to maintain
the integrity of the musical cells within (shown as bars in Figure 2). While chord-seq
objects for continuous progressions can be passed in their entirety to the applicator, the
initial data for discrete progressions must be passed as a list of cells, so that the applicator
can keep track of where the divisions are. If we consider repeating cells as columns in
a table, and the different voices as rows, this means that music for discrete progressions
will be sent as a matrix of rows and columns (Figure 3-b), while passages for continuous
progressions may be sent as rows only (Figure 3-a).
These two ways of organising progressions can then each be divided into two treat-
ments in my compositional system, each corresponding to a specific series of tools.
3 The applicator uses “lambda” functions applied to the passage for any number of voices at a time.
56
Figure 1. Summary of progression process.
Figure 2. Continuous vs. discrete progressions.
Figure 3. A list of voices compared to a matrix of cells.
Applicator patches for the first treatment apply the functions directly to chord-seqs, and
entire chord-seqs are passed to the progression function through the applicator. Where
necessary, such as with discrete progressions, the applicator contains a mechanism to map
break-point functions (bpfs) or constant values to the arguments required for processing
each chord-seq. Applicator patches for the second treatment allow functions to be applied
to one note at a time, which I term musical atoms (including pitch, onset, duration,
velocity, MIDI channel, and offset information), and the application mechanism maps
different points on a temporal bpf to different arguments for the function being called.
Values can thus be sampled from a bpf for each individual atom in a line, and used as
arguments for functions that alter atoms.
57
Matthew Lane
Figure 4 shows how values from a bpf are mapped to a musical passage depending
on whether they are mapped continuously or discretely, and whether they are mapped
by chord-seqs or by atoms.
mapping continuous progressions discrete progressions
No temporal mapping unless inside

Points on bpf mapped to each cell,
the progression function. The con-
based on position of that cell within
stant is used as an argument for the
the whole passage.
function applied to the chord-seq.
chord-seq
!
Points on bpf mapped individually to Points in the bpf mapped to each note
each note in the passage based on the based on the position of the note’s
note’s temporal position. parent cell in the passage.
atom
Figure 4. Continuous and discrete progressions using chord-seq and atom mapping.
Figure 5 shows the modular framework of these four systems implemented in Open-
Music. The music is passed from an original motive at the top to the complete altered
progression at the bottom. At the beginning, a multi-seq
!
class (A) contains an imported
motive. (B) shows bpfs for controlling the transposition, one bpf for each voice. The
next level (C) differs for continuous or discrete progressions. For continuous progressions,
the patch createbasepolypattern
!
multiplies this motive, creating and joining any number
of copies of it for any number of lines. Thus it produces one long chord-seq with several
iterations of a motive for each instrument. The patch fillmatrix from multiseq, for discrete
progressions, is similar but does not concatenate the repetitions, instead holding them
in a matrix so the cells can be treated individually. Step (D) is the application of a
patch to one or several musical passages—the primary place where these four methods
differ. For the purpose of this example, the same list of bpfs is used for all of the
progressions, one for each voice of the input data. In the first method on the left
(continuous progression/chord-seq mapping), the chord-seqs and bpfs are passed one at
a time to the patch chord-seq transpose by BPF via applydifferentlytoseveralscoreobjects,
58
Figure 5. Two types of progressions, with two ways of applying each.
which does little more than a the Lisp mapcar.4 In the second method (continuous
progression/atom mapping), the applybyatoms function maps the values from bpfs (or
constants or lists) based on the onset time of each note, and the lambda patch note
transpose only deals with one note at a time. For transpositions this is trivial, but for
more complex progressions this allows the programmer to focus only on how an argument
affects one note at a time, and saves reprogramming the bpf time-mapping each time.
Applybpfstomatrix is essentially another elaborate list iterator, working on the matrix of
chord-seqs instead of a list and sampling the supplied bpfs (or other values) to produce
the lambda function arguments. Applybpfstomatrix by atom is similar, but instead of
passing chord-seqs it processes individual notes as atoms—being careful, however, to map
the same values onto all the notes from one musical cell. This ensures that cells can be
treated independently: for example when transposing a cell this way intervals in the cell
will remain intact, but when using the applybyatoms method for continuous progressions
each interval will be altered as the transposition curve is applied (see Figure 2 for an
example).
Looking more closely at the resulting chord-seqs in Figure 5, the first and second are
identical, as are the third and fourth. So in this case application by chord-seq and by
atom yield the same results. However, the left two continuous transpositions differ from
the right two discrete transpositions: in the continuous case, even within the first three-
note cell, the initial intervallic relationships are distorted compared to the initial cell at
the upper-left of the patch. In the discrete case, entire three-note cells are transposed
4 Unlikemapcar, where each argument has its own input, applydifferentlytoseveralscoreobjects has one
input for the score objects and one for a list of all other arguments combined.
59
Matthew Lane
together, maintaining the size of the melodic intervals in the initial cell: these intervals
are repeated starting again at the fourth chord, the seventh, and the tenth.
Any number of these applicators with corresponding functions can be applied to the
lists/matrices of chord-seqs. Finally, several further functions are used:
• present list as multi-seq (E) takes the matrix of chord-seqs and reduces it to one
multi-seq, concatenating the cells in each voice.
• mergegroups (F) combines certain lines together as part of one instrument. For
example, in some cases one instrument (like a piano) may play several lines at
once—this allows them to be merged together into one staff.
• assignchannels (G) assigns each line its own MIDI channel, imperative for audi-
tioning the result through a sequencer.
A note on atoms
Atoms5 are a concept to simplify how functions are applied to chord-seqs where a
process is being applied to each note inside the chord-seq independently. Applicators like
applybyatoms manage the entire process of listing the atoms (“atomising”), passing each
atom to a function, and mapping the appropriate arguments for their transformations.
This includes mapping the y-values of bpfs supplied as arguments to different atoms
based on their onsets. Once inside a function, another function atom-›comps splits up
the elements of the atom (pitch, onset, etc.). Figure 6 shows a simple example of how a
transposition function would work in the context of applybyatoms.
Figure 6. applybyatoms maps bpf values to a function that alters each note individually.
5 The “atoms” used here are not the same as Lisp atoms, but retain the name for the same reason: they
are indivisible. They represent, for this way of working, the smallest piece of information that can still
be useful to a musical progression.
60
Types of progressions
The progressions used in my music can be grouped into three categories: pitch functions,
time functions, and stochastic functions, although with a significant degree of crossover.
Pitch functions include those that gradually raise/lower pitch or gradually expand or
compress a melody’s range. Time functions might include those that gradually slow
down/speed up a passage, apply a curve or mathematical function to the speed of a
passage, or gradually alter the lengths of the notes. Stochastic functions, usually used
in conjunction with pitch or time, include those that gradually increase/decrease ran-
domness in time or pitch, but might also include processes that gradually but randomly
remove, repeat, or freeze notes.
Most of these functions work well using atom-application tools because the various
values that affect change are applied on a temporal basis (as progressions generally are),
making the bpf mapping mechanism very useful. Some progressions, however, require
the function to consider overall low and high values for pitches or note lengths across an
entire chord-seq. This is poorly handled with atoms because they allow a function only
to manage one note at a time, disregarding other notes. In these cases, functions are
constructed to apply to whole chord-seqs using instead applicators applybpfstomatrix or
applytoseveralscoreobjects.
Progression programming in my compositional process

Advantages
The primary advantage of conceiving and programming progressions in this manner is
the amount of material easily generated, and the possibility to hear quick approximate
results by changing various parameters. For example, imagine a patch that first gradually
stretches timings, then gradually raises pitch as a function of time, then a function that
removes notes as a function of time, and finally a function that increases the range of
a motive as a function of its location in the passage. While the result of this process is
easily musically imaginable, the creation of such a passage, lasting perhaps thirty seconds
at a fast tempo, could take an afternoon. Now imagine that the end result were not quite
satisfactory, and one wanted to change the first step of the algorithm to stretch the timing
slightly more over the length of the passage. This process affects each of the following
processes in sequence, and rewriting this by hand could take another afternoon; whereas
verifying the workability of this alternative would take seconds in OpenMusic.
Stemming from this flexibility are formal advantages for composing, including the
possibility of efficiently creating variants of a passage. These can either be parametrically
different passages that may require only a slight shift in parameters, stochastic variations
requiring only a new evaluation of the same patches, or variations of processes that
apply a similar colour to different source materials. Because of the nested nature of
the patches, it is also possible to create progressions of progressions (to any number of
layers), retaining a formal direction in each level of nesting.6
6 Thisnesting (possibly recursive) can be used to generate many coherent musical structures, including
and especially fractal structures. For more on this see [1].
61
Matthew Lane
A variant of this idea is the possibility easily to create progressions that go elsewhere
from the original instance. For example, the same progression could occur many times,
each transposing to a different tonal area, possibly over a different period of time. Finally,
one of the most interesting benefits of composing in this way, and one that can be
extrapolated to OpenMusic in general, is the occasional artefacts from certain processes,
the “tool marks” left by the software that often provide the unique gems from which entire
pieces can be built.
A fork in the road:

Further programming or interpretation of data into music
The output data usually provides a clear-cut progression with an obvious direction, but
like some ideas developed using CAC, it could sound overly mechanical played raw.
Here, the composer is left with a choice: interpret the data and begin to compose with
it by hand, or develop further processes in OpenMusic to render data closer to the
musical intention. In this sense, interpreting the data refers to applying in some way
a compositional sense of intuition to the computer-generated output. It is crucial to
understand where and why the data falls short of being effective musically, whether for
the purpose of improving the programming process or simply for the sake of efficiently
pinpointing what needs fixing by hand. While this discussion could be much elaborated,
below are two main areas of interest that have come up in my work.
One of the principal decisions often relates to data with randomness. What we
perceive as randomness and seek occasionally in our music is rarely true randomness,
but rather even distribution with a degree of randomness. Four voices stacked on top
of one another are occasionally bound to line up, creating a synchronicity that our ears
perceive as ordered. In the same way, in a sequence of random note values, there will
occasionally be several of the same value in a row, creating an ordered impression where
there is technically no order. In many cases in OpenMusic, several evaluations are
required before obtaining a result that sounds truly random. It is possible (although
it involves further programming) to use constraints systems to control the amount of
repetition of an element in a random sequence or to control or restrict the output of
several elements that follow a pattern (for example, four random numbers that result in
the sequence 2, 4, 6, 8).7 Another possible approach is to avoid randomness altogether:
instead to create sequences with a random-sounding distribution, and apply Markov
chains to produce variants, lowering the probability of undesirable patterns. Currently,
I have dealt with this issue in two more primitive steps, keeping closer to the music and
further from programming at this stage. First, the patch is evaluated several times to
find a result that sounds appealing. Second, the result is reworked by hand (interpreted),
altering rhythms or sequences by ear or by searching through the preceding material for
patterns, notes, or note-values that have not yet appeared. For longer passages, it can
even be worth performing several evaluations and taking pleasing segments from each.8
A second issue is that progressions produced in OpenMusic often imply an explicit
direction and stick to it, even in cases where intricate bpfs control the progression instead
7 Several constraints programming systems exist for OpenMusic, e.g. OMClouds: see [4].
8 OpenMusic is also conducive to other forms of random distribution, using libraries like OMAlea.
62
of straight lines or simple curves. Once a progression becomes too predictable, it loses
much of its dramatic value. Even randomness, which adds a degree of instability, can
itself become predictable. One approach to dealing with this is by interrupting the
progression. I have developed tools to incorporate freezes and silences in some cases based
on direction changes in bpfs, but these also engender their own type of predictability.
Randomly (or partially controlled randomly) placed freezes, pauses, or other processes
such as flashbacks to previous sections, occasional scrubbing backwards and forwards in
the sequence, or elements as simple as sudden octave transpositions are all possible, but
this kind of formal randomness is rarely an æsthetic I seek while composing. Thus, I prefer
interpreting the OpenMusic data by hand and instead focusing on listening. As soon as
I recognise the intention of a progression, that means there is not long to change it before
it loses its musical interest. From there, the question is whether the progression should be
interrupted and continued again, interrupted and restarted, or stopped outright. When
a progression returns for the second or third time, as a recurrent idea in different parts
of a piece, the third option is often best. Next, the nature of the interruption comes into
question. It may be a slightly altered continuation of the progression already in progress
(for example, an upward scale that momentarily doubles in speed or becomes an upward
glissando). It may also be a sort of stop or break, such as a freeze or a silence. Or it may
refer to completely different material, although usually material that has already been
introduced, suggesting a flashback. Finally, only once the first interruption is written is
it possible to move forward and determine where and if another interruption is needed.
To be musically useful, this process must take into account the material and time that
precedes it.
Even beyond these two issues, the data may still lack the flow or spontaneous feeling I
seek in my music. Regardless of how far the patches go, the data itself is rarely musically
suitable for my needs. For this, more drastic measures are required: it is here that
reinterpretation becomes relevant.
Reinterpretation
What I call reinterpretation is a process like the interpretation of data discussed above,
but a step further towards impulsive intuition. The means are varied, but the principle is
the same: become a musical performer and improviser of one’s own data and document
the process. In some cases, this involves playing through the passage on an instrument
and recording or transcribing (especially the rhythm). This opens the possibility to move
back and forth in time, add occasional notes, or alter harmonies where the ear suggests.
In cases with many lines undergoing processes simultaneously, I may take what strikes
me as the leading line, reinterpret this, and then rebuild the other lines around this one.
Some computer-assisted tools for reinterpreting the passages are also useful, including
scrubbing and transcribing the result in a notation program or sequencer, or simply
tapping the tempo or chord changes and re-transcribing the new rhythm. These are not,
however, recipes to successful reinterpretation and require discretion to determine their
utility in each case.
Most important, about both interpretation of data and reinterpretation, is the time
and “space” required away from the computer-aided composition environment. Other-
wise uncomfortable passages become familiar and comfortable sounding, and undesired
rough edges become softened in our memory. Programming in OpenMusic can involve
63
Matthew Lane
laboriously going over the same passages many times for hours or days on end. This is
why, for me, it is crucial to return to the output data several days after the evaluation
and ideally to return to it in a new context. The time allows me to rehear the musical
ideas in a fresh way and to be surprised, a crucial element to my music. Either playing
the output on an instrument, in a sequencer, or in one’s head can work, but I avoid
returning to OpenMusic. For this reason, I tend more and more to work on multiple
parts of a piece at the same time, allowing me always to have something to which to
dedicate my attention (a principle of Toyota’s “lean production”, see [5]).
Patches at work in compositions
Use and interpretation in Sliding Apart

CAC processes were used in nearly every section of this piece. Between sections, the
main fast theme breaks down by dying and fading away in several dimensions. As soon
as the motive begins to become obsessive, it breaks down rhythmically, slowing down at
the same time, and putting the different lines out of sync. Certain notes go missing from
the pattern, and the pitch slowly drops. Some elements are completely controlled in this
progression, while others have stochastic influences. The musical goal is to create a sense
of sudden falling apart in order to break away from a section that is otherwise extremely
rigid and rhythmically precise.
The progression takes place within one
large patch (see Figures 7, 8, and 9), in which
the general structure is to be noted before
looking at the processes in detail. First,
because the progressions are only functional
on single-note lines (and not chords or other
forms of harmony), the violin and cello
double-stops (two-note chords) are broken
up into two lines each.9 Also, while every
instrument is passed through every process,
not every process affects every instrument:
certain processes are simply controlled so as
to have no effect on some of the instruments.
The beginning of the patch, as seen in
Figure 7, imports the initial cell, creates
ten iterations of it, and adjusts the tempo
before passing the list of chord-seqs onto the Figure 7. Sliding Apart patch (1).
treatments in Figure 8.
The first significant treatment, a rhythmic expansion (rhythmextendcompress), works
in ratios: the initial ratio (1 for all the instruments) is attached to the lambda patch,
while the destination ratios are attached to the applydifferentlytoscoreobjects patch.
9 The score order in Figure 7 is: flute, clarinet, piano, violin line 1, violin line 2, cello line 1, cello line 2.
64
Any argument that is the same for all in-

struments can be connected directly to the
lambda-mode patches. While this process
does not apply to the strings at all (their
ratio moves from 1 to 1), it does mean the
flute becomes progressively slower, as do the
clarinet and piano to lesser degrees. In con-
junction with the next process (melodyrisefall-
overallinterval), which lowers each line over
a specific interval, the highest-pitched instru-
ments slow down and fall the most in pitch,
contrary to our usual conception of inertia.
Versions of this patch used in other parts of
the piece include an additional process after
rhythmextendcompress, to equalise the lengths
of the different lines by padding them so
that time-based processes could be applied
similarly to all lines.
The stochastic processes are randomize-
timegradient and gradualnoteremoval—their
interest lies not in being random, but in cre-
ating a gradient in randomness. Randomize-
timegradient takes four arguments: the
amount of randomness in possible millisecond
Figure 8. Sliding Apart patch (2).
deviation at the beginning, the amount at the
end, and the temporal start- and endpoints for
the process. Here, for example, the process
begins 10% of the way through the passage
(0.1) and finishes half way through (0.47).
Time in this function does, however, remain
linear, meaning notes and their endings still
retain the same order. Gradualnoteremoval
also contains a random element, removing
more and more notes, but not necessarily
the same ones on each evaluation—meaning
different evaluations render different material
from which to choose. Both of these processes
contribute to the sense of falling apart in
the lines. Note that randomizetimegradient
applies more strongly to the strings, while
gradualnoteremoval is not at all applied to
the strings, meaning the string attacks end
up being much more unpredictable, but they
maintain all of their notes. Dynamic gradation
simply creates a progression of velocities to
help in the simulation. Figure 9. Sliding Apart patch (3).
65
Matthew Lane
The interpretation of data for this piece involved multiple steps of reworking the
output data by hand. The major changes from the output data were the inclusion of
an interruption and another process (also generated in OpenMusic) cross-fading with
the ongoing progression. The interruption is a combination of two events: a “freeze”
in the wind instruments and piano, while the strings continue the progression by their
downwards fall (Figure 10, m. 32). The cross-fading process, a rising, somewhat spectral
piano chord progression, helps offset two problems in the first progression. It counteracts
the loss of energy created by the falling-apart progression, and it distracts from the
increasing predictability of where the first progression is headed.
Figure 10. Sliding Apart score segment, showing the gradual deceleration and downward
motion of the flute and clarinet lines along with increased randomness in both the violin and
cello lines. The small downward slashes after some notes represent short glissandi.
Use and interpretation in Melodious Viscosity

The wind quintet Melodious Viscosity also uses CAC in several places, primarily for
progressions. One particular progression (which returns in different parts of the piece)
is the most salient example and a relevant look at how many of the same processes from
Sliding Apart can be used to different ends. The passage (seen in Figure 13) builds
toward a climax, so many of the same processes as in Sliding Apart can be seen in
reverse. There is also a process within a process, as the concatenated base segments of
66
material also each undergo decreasing randomisation, hence the importance of the patch
createbasePatternWithMultipleEvaluations (Figure 11), which allows the random element
to work to its full potential, creating slightly varied instances before submitting them to
the global chain of processes.
Figure 11. Left: patch createbasePatternWithMultipleEvaluations. Right: Main patch for

Melodious Viscosity (part 1).
The most important process is gradualnoteremoval (see Figure 12), again in con-
junction with adjustdurations in order to fill out the space emptied by the removal of
notes. Each instrument is given separate parameters for when to enter. There is also
an acceleration treatment through rhythmextendcompress, but in the end I deemed it
simpler for the players to read an accelerando in the score; although the OpenMusic
process helped me gauge the efficiency of the process as a whole.
Figure 12. Melodious Viscosity patch (parts 2 and 3).
67
Matthew Lane
Figure 13. Melodious Viscosity for wind quintet, score excerpt (concert pitch).
A crucial part of the interpretation of results in this piece involved performing many
evaluations. After that, bars were chosen from different sets of results that filled certain
auditory criteria, including a desire for a feeling of evenly distributed randomness, attack
points that did not sound like errors, and interesting harmonic coincidences. Then a
partial reinterpretation was done, primarily of what I perceived to be the leading lines:
bassoon (on the bottom staff of Figure 13) as well as the highest-pitched line present.
The bassoon’s first notes for several measures were altered to make the passage feel
less harmonically static. Finally, touches of colour were added: occasional harmonic
contributions by instruments where the texture was too thin, and occasional performance
techniques momentarily to draw attention away from the progression. This progression
occurs in two section of the piece, and the CAC framework allowed the creation of two
independent, but closely linked, progressions.
Use and interpretation in Short Pieces on Falling: “Rings”

The base material for the short piece “Rings” is entirely developed as a progression in
OpenMusic. This was the first piece where progressions were mapped from bpfs, and
exceptionally the base material was developed in OpenMusic rather than imported.
The homophonic moving chords carry the distinctive harmonies of a ring modulation,
68
becoming clearer as the lines move higher and all of the instruments become consistently
present. Short rhythmic segments change speed, reflecting their slopes on the generating
bpf. All of this is juxtaposed with sudden stops in the fast and difficult-to-predict lines.
First, as shown in Figure 14, a rhythm
is developed from the hand-designed bpf
(createrhythmfrombpf - linelength). The
steeper the slope of a bpf segment, the
faster the notes repeat. Then the same bpf
shape is used to apply pitches to the existing
rhythm (apply notes to rhythm chord-seq).
Third, the chord-seq is “frozen” in places
(notes are suddenly held) based on changes of
direction in the bpf (and therefore melody),
using the functions get direction changes in
BPF and multi-freeze. Proceeding to part 2
in Figure 15, the function bpf-multiscale-y
scales the y-axis of the bpf for application
to different functions: while the same bpf
is mapped to rhythm and midicents, the
range of values needs to be different for each
(a difference of 2000 for 1 2/3 octaves in
midicents, and 5000 ms as an appropriate
rhythmic slope).
At this point, the result is still a single
melody (it enters Figure 15 via the connec-
tion at the top left), but is then passed to
applybyatoms to apply a ring-modulation one
note at a time, using the given note from
the chord-seq as the carrier and the mapped
value from the bpf as the modulator. Note
that the shape of the line has changed Figure 14. Patch for “Rings” (1).
somewhat due to the process of “freezing” in
certain places, and the function multi-freeze has provided a new bpf with “freezes” in the
same temporal locations as they have been placed in the chord-seq. Between the carrier,
modulator, upper sideband, and lower sideband10 there are now chords of four notes
throughout the progression, and the function split chords to melodies splits these into
four new lines (the opposite of the function merge). The final significant function, applied
to the four chord-seqs using the applytoseveralscoreobjects applicator, is noteremovalbybpf,
which uses a bpf with y-values from 0 to 100 to determine stochastically the percentage
chance that a given note stays or is removed. In this case, it is arranged so that as the
melodic line moves higher there are more notes retained. Finally MIDI channel numbers
are applied to each line, and the output is displayed in a multi-seq.
10 Ringmodulation is originally a signal processing function, creating additional frequencies by adding

and subtracting two given frequencies. If Fc is the carrier (the chord-seq frequencies in this case), and
Fm is the modulator (the frequencies mapped from the bpf), then the two additional frequencies are
Fc − Fm and Fc + Fm .
69
Matthew Lane
The interpretation process for this piece

differs from the others in that I did not
use the final result as a whole passage
anywhere in the piece, but rather elected
to use it as the structure for the work:
repeating segments of it while juxtaposing
more Romantic-sounding piano lines against
the rather harsh fragments of the original
progression (see Figure 16). The inter-
ruptions do not follow any mathematical
pattern, and cannot, in the end, be predicted
by any factors in the progression, giving the
piece a sense of continual surprise within
a context of a clearly-defined trajectory.
Compare Figure 16 and Figure 17 to see how
some of the source material was chosen, split
up, and repeated (segment a in mm. 10-
12, segment b in mm. 13-15). In many
cases, for variety, several evaluations of one
passage were used, differing slightly due
to the stochastic noteremovalbybpf process
(Figure 15). For example, note the dif-
ferences between motives in measures 10
and 11. Otherwise, aside from removing
an occasional note that fell out of range,
the lines were not significantly altered to
preserve the raw quality of the original data.
Finally, a note on quantification: the
output of OpenMusic was exported in MIDI
format because, at the time, MusicXML files
caused problems upon import to Sibelius
notation software. Because the rhythms
encoded in the MIDI files were mathemat- Figure 15. Patch for “Rings” (2).
ically derived from the outset, and could
correspond to any value in milliseconds, quantification in the notation software produced
rhythmic artefacts when imported, which I chose to embrace as part of the composition
of the piece. Figure 16 shows a few examples of these. The flute, clarinet, violin, and cello
are all based on the same rhythm throughout, although sometimes with missing notes.
Note at the beginning of m. 11, for example, how the upward passage is a combination
of sixteenth notes and quintuplet-sixteenths. Similarly in m. 13, what were originally
even durations in milliseconds became a combination of eighth-note triplets, sixteenth-
note triplets, and sixteenths. These artefacts finally provided much of the colour for the
piece.
70
Figure 16. “Rings” score excerpt (concert pitch).
Figure 17. Segments of source material for “Rings” (concert pitch).
71
Matthew Lane
Use and interpretation in Short Pieces on Falling: “Float”

“Float” is an example of the use of discrete progressions, where individual cells within
lines had to maintain their integrity in order to give the formal structure meaning. First
a matrix is created with the required number of cells of the original material for each
instrument. Figure 18 shows what this matrix might resemble schematically, with cells
as columns and instruments as rows; Figure 19 shows how this matrix is processed in the
OpenMusic patches. The first treatment applied to the matrix is a transposition. The
bpf is sampled to determine the transposition for each cell based on its position in the
progression, and then the transposition is applied to each cell as a whole. By maintaining
a constant transposition for a whole cell, and then jumping to a new transposition (as
opposed to moving there gradually), the intervallic relations in each cell are maintained.
Figure 18. Representation of a matrix of instruments (rows) and cells (columns).
Figure 19. Patch for Float (3 parts).
72
Next, only a portion of each cell is preserved, again using the same bpf mapped
to the select with intervals function and applied to individual cells. This mapping and
application of the function is done using applybpfstomatrix once again. As opposed to
a standard select function, select with intervals forces the selection only at certain time
intervals in the passage. For example, this ensures that if it received the mapped starting
and ending points of 212 and 492 ms, it actually would select from 200 to 500 ms (for
an interval of 100 ms). The function also contains safeguards to ensure that the final
resulting chord-seqs are in fact the exact length requested, even if there is empty space
at the end, and takes not only entire notes, but cuts portions of notes where necessary
at the beginning and end.
The most characteristic aspect of the progression is the repeatsegment function applied
to each cell. This function cuts out a segment from somewhere in the chord-seq and
appends it a certain number of times to the end. In this case, due to the mapping
by applybpfstomatrix, the length of the portion to extract and the number of times it is
repeated are both controlled by the same bpf shapes that have controlled everything else.
Thus, as the bpf moves higher, creating higher transpositions, intensity is also accrued
by shortening the cells unexpectedly and by obsessively repeating their final segments.
Finally, the speedchange function quantifies the note lengths so that the notation software
(in this case Sibelius) recognises them as eighth notes (for example 500 ms at a tempo
of 60). Figure 20 shows how one cell might change as it passes through all of these
functions.
Figure 20. How one cell transforms as it progresses through various functions.
Because the structural integrity is maintained from the original cell provided, this
type of progression is already much more logical to the ear, and finding a convincing
progression involves mostly work on the initial bpf and some of the scaling factors.
The work of interpreting this data into a score is less destructive, involving primarily
the addition of dynamics, to sculpt a curve of intensity reflecting the direction of the
passage, as well as articulation. Both of these additions help punctuate clear divisions
between the original cells in OpenMusic, so that the beginning of each new cell is clearly
recognisable. Figure 21 shows a segment of the final score.
73
Matthew Lane
Figure 21. “Float” score excerpt.
74
Further tools
While the applications for this modular system have not been exhaustively explored, some
simple additional tools have drastically multiplied the contexts in which these patches
are effective. These include, for example, a patch to constrain musical objects to a given
mode and a patch to contract several instruments’ parts to a common temporal length.
The former passes through all the notes in a chord-seq, and if they do not belong to
the supplied mode, alters the notes so that they do. The later takes a list of chord-seqs
(corresponding to different instruments for example) and adjusts the final note of all but
the longest so they are all of equal length. While working with matrices of chord-seqs,
where all the elements in a row will eventually be concatenated, this can be crucial in
ensuring that the instruments (the different rows) line up with each other.
Other OpenMusic and sequencing resources also could allow more complex and
layered progressions. One possibility is to use multiple channels corresponding to different
playing techniques of one instrument, to allow gradual progressions where the stochastic
chances of each techniques change throughout the piece. Another is to use 3dcs (three-
dimensional curves) to map the y- and z-axes to different elements of a progression.
CAC consistently raises the question of how much “composing” the software does.
At present, no data developed in this system could be used as music in its raw form,
but analysing why helps us move closer to musical results both with OpenMusic and
by hand. The more we can formalise solutions to the problems associated with musical
data from OpenMusic, the better we are prepared to formalise solutions for issues in
our own music by other means. It becomes irrelevant whether a composer creates a
patch to rectify a problematic passage or finally composes it by hand: once the patch
is created, it means that the composer understands perfectly the cause and solution to
the problem in small steps. Consequently, should a composer never even use a note of
data taken from OpenMusic, the process of programming these progressions remains
inherently valuable. And while CAC programming may be pushed farther to interpret
our musical ideas, they remain our musical ideas, and not the computer’s.
References
[1] Michael Gogins. “Iterated Functions Systems Music”. Computer Music Journal,
15(1), 1991.
[2] Matthew Lane. “Modular Progression Programming in OpenMusic”. In Actes des
Journées d’Informatique Musicale. Montréal, 2015.
[3] Olivier Messiaen. The Technique of My Musical Language. Paris: Alphonse Leduc,
1956. Trans. John Satterfield.
[4] Charlotte Truchet, Gérard Assayag, Philippe Codognet. “OMClouds, a heuristic
solver for musical constraints”. In Metaheuristics International Conference. Kyoto,
2003.
[5] James P. Womack, Daniel T. Jones, Daniel Roos. The Machine That Changed the
World: The Story of Lean Production—Toyota’s Secret Weapon in the Global Car
Wars That Is Now Revolutionizing World Industry. New York: Free Press, 1999.
75
Musicalising sonification:
Image-to-music conversion using
OpenMusic
Luiz Castelões
David Cope [4] defines sonification as the production of sound “by translating (called nor-
malising), in one way or another, traditionally non-audible data to the audible frequency
range.” He goes on to say that “[s]ince the data typically used for data sonification has
little if any inherent musical logic, the normalised output from sonification generally has
little musical use” (p. xi, italics are mine).
Departing from Cope’s definition, as sonification becomes more and more widespread
and trivial (concerning the technologies involved) the compositional challenge at hand
shifts from the act of converting between different realms (image-to-sound, number-to-
sound), which characterises any given sonification, to effectively obtaining consistent,
original musical results through such operations.
Strategies for turning sonification into music (and finding possible intersections be-
tween musical logic and other logics in the context of sonification) have been a major
field of interest for me as a composer and scholar and will constitute the main focus of
this article, particularly with regard to image-to-music conversions.
Introduction
Image-to-sound conversions (in a broader sense than image-to-music conversions)1 con-
stitute a widespread type of sonification. One could even propose that a modern/contem-
porary tradition of image-to-sound conversion for compositional purposes began devel-
oping even prior to computer-aided composition—a tradition that includes Scriabin,
Villa-Lobos, Messiaen, and Xenakis. I will use the word “conversion” to refer to this
type of sonification, instead of Cope’s “translation”, because I believe that the latter
implies some shared meaning between the data to be sonified and its corresponding
1 Even if, in a general sense, music can make use of any sound (Cage’s “anything goes”), that is not
the same as to state that music is any sound for every composer and every musical work. In practical
terms, the amount of effort and intent that most composers put into their compositional processes
reveals that their music seeks certain sounds (potentially different for each composer, or each work), a
much more specific territory than simply any sound (“whatever”, “n’importe quoi”). This is also valid
for improvisatory, aleatoric, and “chance” approaches, which in their own ways also produce certain
sounds as opposed to any sound.
77
Luiz Castelões
sounding result (which is not necessarily the case), while the former is more neutral with
regard to obtaining a shared meaning. Also, throughout this study, whenever image-to-
sound sonification results in music it will be called “image-to-music conversion” so as to
differentiate it from Cope’s understanding of sonification (as having “little musical use”).
Clearly there is not a unique solution to the problem of obtaining music from image-
to-sound conversions, but rather multiple strategies that can serve musical solutions:
not least because of the premise that music is culturally based and individually based.
Therefore making music may constitute a different task (and, as such, imply different
methods) for different (groups of) individuals.
Following the model of a case study, I will present some compositional strategies for
making music out of sonification that have been employed in two original works: the elec-
tronic soundtrack of VIA (2013) and 3 Transcrições (2011) for solo piano. These works
highlight two separate aspects of the aforementioned compositional problem: contour-to-
music conversion and colour-to-music conversion. Although this investigation focuses on
describing a list of strategies for obtaining image-to-music conversions in these concrete
compositional examples, it is also part of a broader perspective on how to musicalise
(humanise, personalise) sonification.
Contour-to-music conversion in the electronic

soundtrack of VIA
VIA is an interdisciplinary project2 integrating dance, architecture, design, music, video,
and smartphone technology. The musical task at hand consisted of supplying 1-minute
bits of music to accompany choreographies that took place in several different streets of
downtown Rio de Janeiro. Each choreography was videoed in situ and accompanying
musical segments were added later. The overall result was delivered via satellite to mobile
phones as the audience walked past the streets where the dance had taken place.
Due to the project’s tight deadline, I had to make the practical decision not to
compose a work of instrumental chamber music, for which the final stage of carefully
transcribing the computer results to a readable score would take a longer time than
available.3 So the final musical segments would have to be produced for electronics,
not for acoustic instruments. Furthermore, I decided to work according to Cope’s CGC
(Computer-Generated Composition, i.e. generating an entire composition from one or
more compositional algorithms) instead of CGA (i.e. generating musical materials that
will later be worked upon to become a finished composition) [4] because the former was
likely to achieve musical results within a shorter period of time—as well as presenting an
interesting algorithmic challenge. I chose OpenMusic as the main tool for the project’s
image-to-music conversions due to my previous experience with its visual programming
environment and with using it to convert a variety of 2D and 3D images (see [2] and [3]).
2 Headed by semiotician João Queiroz and dancer Daniella Aguiar and begun in 2013. For more
information on VIA please visit the website: http://via-locativeart.com/.
3 Following Joshua Fineberg’s invaluable advice, I always transcribe CAC (Computer-Assisted
Composition) or CGA (Computer-Generated Assistance) results by hand, a slow process that however
allows for an intimate control of the final results in the score.
78
Musicalising sonification: Image-to-music conversion using OpenMusic
These preliminary decisions left me with three possibilities for a sound source:
(a) sound synthesis (via OpenMusic and CSound, for instance); (b) exporting the
generated OpenMusic scores to other software in order to generate sound; or (c) using a
sound preview provided in OpenMusic to render the musical results (typically via MIDI
playback).
The possibility of assembling bits of ready-made timbres (a sort of collage approach)
seemed closer to the project’s approach to video-dance, based on editing short images
of dance taking place on the streets: with pedestrians passing by, pigeons flying around,
street vendors, etc. So (a), sound synthesis, was discarded for its usually more abstract
approach, whereas (b), OpenMusic scores exported to other software, and (c), MIDI
rendering, were tested numerous times using hundreds of conversions. Eventually most
of the results used in the project’s final version were the product of the built-in MIDI
rendering within the OpenMusic environment.
Converting image contours to musical parameters

In general, the image-to-sound conversion used for this project followed the visual logic
already present in traditional musical writing: that is, the x- (horizontal) axis supplied
rhythm-related data, subdivided into onsets and durations, whereas the y- (vertical) axis
supplied pitch-related data, within a variable pitch space that was flexibly normalisable.
The process sought not only to sonify in Cope’s sense, but also that this sonification
generate satisfactory musical results (obviously according to the subjectivity of the
composer). The patch shown in Figure 1 illustrates the overall process of conversion
used in VIA. The straightforward and flexible approach to pitch is shown by an om-scale
function tied to two note objects to normalise the image contour’s y-axis values within
any desired pitch range.
As constraints on pitch range were not demanded by any acoustic instrument, this
flexibility was instead intended to allow empirical testing of the best values for the desired
sounds and metaphorical “sound spectra”.4 For each different range of normalisation one
obtains a different sonic (harmonic, intervallic, spectral) profile.
Other musical parameters, such as intensity and instrumentation/orchestration, were
generated based on the same x- or y-axis input data, but with different algorithms, as
shown in the following sections.
Instrumentation and orchestration

But why use default MIDI sounds for compositional purposes in 2013?5 The answer
is threefold. First, the use of these sounds nowadays provides a sort of “vintage”
sound profile. Second, they allow for a compositional approach based on the reuse and
recombination of clichéd sounds, referencing collage and pop-art approaches. Finally,
they represent an enticing compositional challenge: as they are not originally meant to
4 Given that each MIDI note often only contributed a tiny “brick” to the overall result, the task is much
like “instrumental synthesis”, albeit carried out with MIDI samples instead of acoustic instruments—
instrumental synthesis taken back to the studio, so to speak.
5 Fora reader growing up in the 1980s or 1990s, such sounds (or their earlier counterparts) already
sounded old-fashioned then!
79
Luiz Castelões
Figure 1. Overall view of one of the VIA’s patches.
be presented as a final music product, but only as a sketch in the early stages of any
compositional process, their use as music represents a personalised, distorted way of
exposing the compositional tool at hand.
In spite of all of these arguments, to my ears most of these MIDI sounds were simply
unbearable either sustained or in isolation. Therefore it was necessary at least to develop
algorithms to avoid two (un)musical situations: solos and sustained sounds.
Avoidance of solos, here taken to mean timbres played alone or with significant
prevalence over other timbres, was addressed and implemented by means of two simple
steps analogous to traditional music approaches: instrumentation (the choice of sound
sources, in this case particular sets of MIDI presets) and orchestration (the specific use
one makes of these sources). I structured the OpenMusic Instrumentation patches so
that any image-to-music conversion would use 15 different instruments.6 In rough terms,
15 is large enough to avoid undesirable solos. This is easily feasible by assigning different
MIDI programs (or instruments) to the different MIDI channels (see Figure 2).
The Orchestration sub-patch, in turn, divides the total pitch range in different slices
(“bands”, or registers), then assigns subsets of three MIDI channels (or instruments)
to each range-slice, and randomly chooses one MIDI channel/instrument of each subset
6 The MIDI standard provides 16 different channels, but one (channel 10) is usually reserved for
percussion.
80
for each note. This is similar to basic acoustic orchestration (e.g. if a certain pitch is
within the range of a flute, an oboe, and a clarinet, then it might be played by any of
them). Several different versions of orchestration sub-patches were developed and tested.
Figure 3 displays a version that segments the total pitch range into five different registers.
Figure 2. Left: sending out instrumentation settings to the MIDI system (one MIDI program
for each channel). Right: inside the Instrumentation sub-patch (random selection of programs).
The Instrumentation sub-patch is locked when an interesting combination of MIDI instruments
is reached.
Figure 3. Left: Orchestration sub-patch controlling the MIDI channel assignments in Figure 1.
Right: Inside the first omloop.
81
Luiz Castelões
In order to address the second point, avoiding sustained MIDI sounds, I systematically
favoured small values for note durations as I normalised the image contour values.
From a more global standpoint, given that resonances are not particularly attractive
in these sounds, I decided to adhere to a music of attacks by limiting myself to the very
beginning of each MIDI sound, its “richest” portion. The most straightforward way to
implement this in OpenMusic is by assigning a short maximum value for <ldur> (the
slot controlling the durations) in the chord-seq. Figure 4 illustrates the use of the Durs
sub-patch, scaling down the score durations according to a small maximum value.
Figure 4. Setting the durations from image contour data. (N.B. this figure hides the other
connections to the chord-seq shown in Figure 1).
By reducing the participation of each sound to a minimum, the focus of hearing is

shifted from the sounds themselves to the complex networks resulting from the conversion
of visual image contours to music. Each sound contributes only a tiny brick, a grain, so its
individual lack of interest becomes significantly less noticeable. The overall task is more
like random granular sound synthesis than writing notated music, especially taking into
account that I also often compressed the inter-onset values to a minimum (see Figure 5).
In the few cases where I used recorded sound samples (instead of default MIDI sounds)
to render a MIDI file generated in OpenMusic (by exporting it to Finale or Reaper), I
selected sounds of acoustic musical instruments used in a free, non-idiomatic way—given
that the music was not intended to be written, read, or performed. The result is a sort
of instrumental music produced acousmatically, one that employs tempo and gestural
profiles impossible for actual musical instruments—a contrabass and a contrabassoon
that play at the speed of Paganini’s frenzied violin, for instance.7
7 In
“Estudo Capanema” from VIA, audio available at:
https://soundcloud.com/lecasteloes/estudo-capanema-2013-para.
82
Figure 5. Scaling onset values in VIA (e.g. to a total duration of 33 seconds). (N.B. this figure
hides the other connections to the chord-seq shown in Figure 1.)
Image-to-music relationships
The implementation of these algorithms in OpenMusic was carried out so as to allow
for maximum flexibility in the normalisation of the image data, like freely compressing
or expanding a picture before printing it. In musical terms, this means that I was able
comfortably to change normalisation values as I converted each photograph as many
times as I wished, and hear the corresponding musical results, until I was fully satisfied.
These at-times-subtle changes in the normalisation values should not be underesti-
mated because they often have a great impact on musical parameters (harmony, timbre,
melody, gesture, texture, tempo, etc.). In this way, they may be the most essential factor
(along with the above-mentioned instrumentation and orchestration strategies) in ob-
taining music from sonification—so much so that no further compositional manipulation
of the input data (such as cutting, looping, superimposing, etc.) was necessary in order
to obtain musical consistency in VIA.
In the end, hearing is the decisive factor in VIA’s image-to-music conversion method.
It determines whether (1) there should be changes in the instrumentation and orches-
tration sub-patches, (2) there should be changes in the normalisation values, or (3) a
musically consistent result has been reached, in which case the process comes to an end
(see Figure 6).
83
Luiz Castelões
Figure 6. Image-to-music conversion method in VIA
Music-to-video relationships
In the early stages of our project I discussed with VIA’s team the best way to apply
music to video-dance for our purposes. We quickly rejected the possibility of adopting
a previously composed soundtrack (totally independent from dance and images) or, on
the other hand, a soundtrack that strictly followed the dancer’s bodily movements—
“Mickey Mousing”. Instead, we chose to convert the visual contours of photographs from
the locations where the dance had taken place. I then redrew these visual contours “by
hand” (using the computer mouse) by placing each photograph as a background picture
in a bpc (break-point curve) object, but always allowing a certain degree of freedom
(Figures 7 and 8).
Figure 7. Redrawing visual contours in a bpc (photograph from Beco das Cancelas).
This freedom was a means not only to personalise the whole process (as another person
would inevitably redraw it in a different way) but also to obtain consistent musical results:
as I repeated the task over and over I gradually began to perceive general connections
between the intricate visual contours and the corresponding resulting sounds, thereby
allowing myself to interfere musically in the conversion process—a consequence of what
Mikhaïl Malt calls a solfège de modèles, which in this case would mean forehearing the
potential musicality beneath the images.8
8 “Qu’entendons-nous par ‘solfège de modèles’? Nous ne parlons pas d’un solfège au sens d’un catalogue
de modèles statiques, soit d’un solfège issu d’une typologie figée. Nous parlons plutôt du développement
84
Figure 8. Redrawing visual contours in a bpc (photograph from Museu de Arte do Rio).
Further compositional questions

Five further questions were posed and tentatively answered during the compositional
process of VIA:
(1) What is the space of musicality intended by this project? Space of musicality refers
here to the musical space one wishes to occupy within a certain work. Specifically, starting
by imagining musical extremes (e.g., highest degree of stability opposed to the highest
degree of instability), one may predetermine the musical/auditory territory—for instance
characterised by a high degree of repetition, or by the alternation between repetition and
contrast, or by athematicism, etc. For the current project, a space was chosen that offered
a middle ground between the high degree of communication or redundancy of pop music
and the high degree of information and contrast of contemporary “art” music—that is,
a space that is intelligible and enjoyable by non-musicians, young people, or specialists.
(2) What is the magnitude of music making? This magnitude (especially as compared
to the amount of data generated by image contours) denotes the amount of data that
is necessary to produce each space of musicality, or each musical work. In any given
creative project that involves image-to-music conversion, one may observe a significant
discrepancy between the amount of data generated from images and the required, or
desired, amount of data to create a certain musical space. In the specific case of VIA,
where the resulting musical excerpts were to be no longer than one minute, such a
discrepancy did occur occasionally: the amount of data generated by image contours
was of a significantly larger magnitude than the amount of data required, or desired, to
create the intended space for this project. This motivated the development of criteria
d’un savoir, de capacités intellectuelles et cognitives de la part du compositeur qui lui permettraient
soit de contrôler et maîtriser le résultat musical issu d’un modèle génératif quelconque, soit d’établir
consciemment le lien entre des représentations graphiques et/ou textuelles de certains logiciels musicaux
avec un résultat musical. Solfège au sens d’aptitude à relier le comportement de deux espaces de
caractéristiques différentes en tenant compte des particularités de chacun.” [5]
85
Luiz Castelões
intended to reduce the amount of data, including the use of only the opening musical
excerpt that resulted from a given image-to-music conversion, or the simplification of
the image contour done by hand in such a way that the conversion generated less
data. In conclusion, creative work using image-to-sound conversion implies adjusting
the magnitude of image-derived data to the magnitude of the intended musical resulting.
(3) If this magnitude is “complex”, or at least “complicated”, how does the composer
relate to this complexity, and what are the best ways to obtain it and orient it toward
an intended musical goal? VIA’s space of musicality ideally sought a level of musical
complexity comparable to handmade crafts or to human activities (even though, perhaps
paradoxically, the work’s music was entirely realised by means of a computer). In
VIA, the strategy for obtaining the desired complexity consisted in starting from data
that was already complex and working from the macro- to the micro-level, instead of
obtaining complexity by working all the way up from small building blocks. We did not
seek complexity per se, but rather the kind of complexity that possessed musical logic.
Nevertheless, it is necessary to remark that:
- The totality of the characteristics that make up such (musically logical) complexity
have not been thoroughly formalised through the image-to-music algorithms devel-
oped for VIA and would constitute an incommensurable challenge; instead, such
desired complexity is simply “encountered” and “chosen” by ear among the several
musical results furnished by the converting algorithms. This complexity is partially
analysed, understood, and managed, and these stages contribute to the continuous
refinement of the original algorithms, an ongoing and potentially endless work in
progress.
- In VIA the composer does not control the whole process from the start, but
rather exerts gradual and partial control over the encountered, and chosen by ear,
complexities generated by image-to-sound conversions.
- The production of musical results in this case is therefore significantly open to
unforeseen structures.
- The final musical result is as much, or even more, “found” than “created”—in
this context, music composition becomes the intellectual-auditory discovery of a
pre-musical object, an abstract object with a highly musical potential.
(4) How does the magnitude of music making relate to independently developed
images of choreography, as sound and image are united within an audio-visual product?
There is certainly a re-signification that takes place as the finished musical component,
which—let us suppose—was satisfactory enough when heard alone, joins the video com-
ponent, which in turn contains the images that served as abstract data sources for the
musical component. In VIA there was limited space for feedback between music and
image; however exchanges of early mixes allowed us to learn about the particular twofold
association between music and video as well as between music and dance. One of the
main conclusions drawn from this dialogue was that music and choreography seemed to
be more clearly related when the musical component was dense and intricate. In early
mixes that had soundtracks with more empty spaces, more silences, the artificial aspect
of the audio-visual collage became more apparent and it looked and sounded like an
unintentional mistake, a technical flaw. This finding has further motivated the search
for complexity within the musical component, as previously delineated in (3).
86
(5) How can one obtain human-like musical results while making music exclusively
through a non-human device such as a computer? How can one make computer music that
sounds as if humans had performed or even improvised it? Here the computer’s role as a
musical tool is emphasised, but without allegiance to a machine æsthetic: one in which the
“watermark” [1] of the employed technology becomes too apparent, rendering the musical
work that employs it obsolete even before the technology’s own programmed obsolescence.
Strategies for obtaining a human-like yet machine-made sonic profile include:
- ongoing refinement of the musical problems’ formalisations and the implementation
of algorithms that provide human-like musical results (which inevitably involve
sound-to-sound intensity variation and frequent tempo fluctuation);
- emphasis on complexity (the more dense and intricate the algorithm output is, the
less often one perceives the small-scale limitations of the technology employed);
- the search for asymmetrical rhythmic sequences, obtained effortlessly and at will
as one copies an image by hand (or with a computer mouse);
- the search for varied instrumentation models (e.g. with 15 different musical instru-
ments) combined with the use of aleatoric orchestration, which simulates spontane-
ity and surprise; and
- orchestration by registers, simulating instrumental chamber music practices.
Colour-to-music conversion in 3 Transcrições

3 Transcrições (3 Transcriptions) for solo piano9 was my first attempt at making music en-
tirely from colours. The entire piece was derived from three colour-to-music conversions—
i.e. algorithms to convert colour data to quantifiable musical parameters. Again I chose
OpenMusic to run all of the compositional algorithms; no piano was used during the
compositional process. Except for a single pitch added in the second movement (the B[6
in m. 61), the three movements stem entirely from colour-to-music conversions of a single
photograph found by chance on the Web.
What the photograph portrays—a colourful van—played no role in the compositional
process or in the resulting musical piece. It was chosen simply for its colour variety
and because it generated satisfactory musical results. To this day, I cannot detect any
meaningful (æsthetic, emotional, conceptual) connections between the source photograph
and the resulting music. It is an exact example of how sonification may work not as a
“translation”.
Colour systems: RGB, HSV, CMYK

The conversions were carried out through OpenMusic’s get-RGB function, which ex-
tracts the red, green, and blue component for each pixel of a picture, as well as through
two more functions that I implemented (with the invaluable assistance of mathematician
Talita Sant’Ana): RGB-to-HSV and RGB-to-CMYK (see Figure 9).10
9 Composed in 2011 and premiered in 2014 by Brazilian pianist Grazi Elis.

10 Detailed
explanations of equations, algorithms, and patches developed for this piece can be found in [3].
The colour-to-music functions can be downloaded at http://www.ufjf.br/comus/cac_patches/.
87
Luiz Castelões
These two new functions allowed me to run

and hear conversions with HSV and CMYK
colour systems, enhancing my experience in
colour-to-music conversion.
The different colour systems (RGB, HSV,
CMYK) utilise contrasting methodologies to
codify colours: RGB codifies colours by
mixing numerical values of Red, Green, and
Blue; HSV by mixing Hue, Saturation, and
Value (Brightness); and CMYK by mixing
Cyan, Magenta, Yellow, and Key (black).
Therefore they produce different lists of
values as they codify colours, and colour-to-
music conversions based on the same image
using these different systems will produce
differing musical results. Figures 10 (a), (b),
and (c) illustrate these differences with the
results of converting the same image using
the RGB, HSV, and CMYK colour systems,
respectively.
To further complicate matters, the region
of the total colour space covered by each
of these systems is not identical. So, even
though it is possible to find intersections
and convert between any two of these sys-
tems with some level of precision (using
the RGB-to-HSV and RGB-to-CMYK func-
tions), when one separates and distributes
the original values of each system among
various musical parameters for sonification
purposes there is no hope that they will Figure 9. Inside the RGB-to-CMYK
coincide. function: an omloop which converts a list
Regardless of any evaluation of the effec- of RGB points to CMYK values.
tiveness of each colour system (based on its
total colour space, intuitiveness, etc.), I consider this methodological diversity extremely
fertile for music composition because it widens the array of possible musical results
derived from colour-to-music conversions. We may suppose that each of these conversions
probably leaves its characteristic mark, although my research has not gone so far as to
describe this mark in musical terms. This diversity also demonstrates how arbitrary the
relation between colour and sound is in the context of non-synæsthetic conversion (or
whenever the composer and the listener do not share a common type of synæsthesia).
In the specific case of 3 Transcrições, only the RGB conversions have been used in
the final version of the score. The choice was purely based on the quality of the resulting
music.
88
Figure 10. Examples of colour-to-music conversions. The same image is converted (a)
using RGB: R=midicents, G=onsets, B=dynamics; (b) using HSV: H=midicents, S=onsets,
V=dynamics; (c) using CMYK: C=midicents, M=onsets, Y=dynamics.
From colour data to musical parameters

The approach to distributing colour data among musical parameters in 3 Transcrições is
in many ways similar to what I did with image contours two years later in VIA; that is,
each list of numerical values is mapped to one or more musical parameters.
Normalisation variables are left loose enough to allow for satisfactory musical re-
sults to emerge. In the first movement R=pitch, G=rhythm (onsets+duration), and
B=dynamics; in the second movement, R=dynamics, G=pitch, and B=rhythm; and in
the third movement, R=rhythm, G=dynamics, and B=pitch (see Figure 11).
What distinguishes this compositional process from VIA’s is the particular strategies
used for adapting the magnitude of image-derived data to the magnitude of music
making: in VIA, I redrew the photographs’ contours in order to simplify, musicalise,
and personalise contour-to-sound conversions, whereas in 3 Transcrições I selected only
a small number of image lines (four) to be converted and then slightly filtered them, as
will be shown in the next section.
Selection of image lines and subtractive synthesis

In order to adjust the amount of colour data output by get-RGB to the amount of data
I needed to compose a solo piano piece, I made two decisions:
First, I selected the lines of the photograph that were to be converted, given that
the chosen picture had 183 lines that were not all needed to make music. This was very
easy to implement in OpenMusic: I determined the position of the lines to extract by
simply dividing the number of lines in the image by the number of voices (+1, so that
the converted lines would never be at the borders of the photograph). This way, the
extracted lines would also be equally distributed across the image.
89
Luiz Castelões
Figure 11. Overview of the patch used for the second movement of 3 Transcrições. Selection
of four horizontal lines from the picture and colour-to-music conversions.
Second, I carefully “filtered” the colour-to-music conversions by ear, very much as

one does when performing subtractive synthesis. My intent with this filtering stage
was twofold: (1) to further musicalise the sonification, and (2) to make the conversions
playable by a human being (i.e. humanise the sonification). This filtering stage was
necessary because even though the converting algorithms had been refined as closely as
possible to obtain a finished composition (Cope’s CGC), the preliminary colour-to-music
conversions functioned only as a sort of raw material, a block of marble that had to
be sculpted further (representing a firm positioning on the side of CGA). This stage
was done entirely by ear and has not been formalised in the converting algorithm. The
various steps of this filtering used in the first and third movements of 3 Transcrições are
displayed in Figures 12 and 13.
Figure 12. Filtering stage in the 1st movement of 3 Transcrições.
90
Figure 13. Filtering stage in the 3rd movement of 3 Transcrições.
Once I reached satisfactory musical results by ear I went on to transcribe everything

into a readable/playable musical score. This was done by hand, gesture by gesture, by
carefully looking at a chord-seq editor grid—a slow, painstaking, yet rewarding process
(see Figure 14). Transcribing music from a chord-seq implies not only notating pitches
and durations (clearly visible on the screen), but above all hearing through what is visible
in order to identify and write down what is invisible (and, therefore, does not allow itself
to be transcribed by a machine). This includes the length and musical character of each
musical gesture, the articulation of each sound, as well as the clearest, most musical way
to fit individual durations into a larger rhythmic framework. Finding a middle ground
between loyalty to the intended sound (since this piece first emerged as sound, not as
a written score) and human playability was the most challenging aspect of the musical
transcription (Figure 15).11
Figure 14. chord-seq of the beginning of 3 Transcrições used for musical transcription.
11 Theaudio recording of 3 Transcrições can be heard at:

https://soundcloud.com/lecasteloes/3-transcricoes-2011-para-piano-solo-elis.
91
Luiz Castelões
Figure 15. First system of 3 Transcrições, after transcription.
References
[1] Rodolfo Caesar. Círculos Ceifados. Rio de Janeiro: 7 Letras, 2008.
[2] Luiz E. Castelões. “Conversão de imagens para sons através de três classes do
OpenMusic”. Revista DAPesquisa, 8, 2011.
[3] Luiz E. Castelões, Talita De Oliveira, Yago Franco. “Conversores de parámetros del
color a parámetros sonoros cuantificables usando los sistemas RGB, HSV y CMYK”.
Sonic Ideas, 7(14), 2015.
[4] David Cope. “Preface”. In Jean Bresson, Carlos Agon, Gérard Assayag (eds.) The
OM Composer’s Book 2. Editions Delatour France/IRCAM-Centre Pompidou, 2008.
[5] Mikhaïl Malt. “Concepts et modèles, de l’imaginaire à l’écriture dans la composition
assistée par ordinateur”. In Bruno Bossis, Anne Veitl, Marc Battier (eds.)
Musique, instruments, machines: Autour des musiques électroacoustiques. Actes
actes du séminaire du MINT (vol. 2). Musicologie, Informatique et Nouvelles
Technologies/Observatoire Musical Français/Université de Paris-Sorbonne, 2006.
92
On “slow” computer-aided
composition
Julien Vincenot
This article presents some developments in the field of computer-aided composition,

gathered inside the author’s library jv-components for PWGL, available in part for
OpenMusic. Aspects are discussed of three pieces that rely heavily on paradigms
of interpolation and constraint-based generation: Ascidiacea, silent_data_corrupt, and
Mémoire de l’eau. In parallel, a broader reflection is opened about the role of computer-
aided composition (CAC) in learning composition today, and the benefits brought by
“deferred-time” composition environments in a time of speed and immediacy.
Introduction
I began to consider myself a composer relatively late in my life as a musician, first
approaching the discipline through electroacoustic composition and improvisation. I now
count myself among the increasing number who learned composition with the help of
computer music tools, and CAC in particular. I completed my studies in the composition
Cursus of the Conservatory of Montbéliard, France with Jacopo Baboni Schilingi and
Frédéric Voisin. In this unusual context, CAC was not merely one more tool to master—
something “on the side”—among all those that composers need to know today. Within
this class we were encouraged to think all together with CAC. The software, PWGL
in this case, was not only a “toolbox” for the composer, but a hyper-expressive means
to learn how to conceive and describe our musical ideas and, above all, a way to share
them. Patches were for us a medium to exchange and confront ideas, and to solve musical
problems lying in front of us that we might not have been able to express in words at
first.
CAC made us entirely reconsider our way of thinking about detail while creating
music: written detail does not necessarily have to be realised by the composer because
its properties can be declared algorithmically, modelled before it even exists. On the
other hand, much more space and time is left for compositional detail: to make artistic
decisions on a multitude of time scales and musical parameters, as is more often the case
in electroacoustic composition.
Some may see in this a crutch, a “fig leaf”. On the contrary, I consider CAC a
prosthesis, the “augmented arm” I need to extend my level of consciousness, to multiply
my abilities for thinking, and to realise my musical ideas. In some situations making a
specific patch is the only way to help me better understand an idea I ultimately write
93
Julien Vincenot
out by hand. Without the bridge that CAC offered me I never could have reached the
level of confidence and freedom I feel today with musical writing; indeed my music would
hardly exist.
In the following sections I present some developments I have made with PWGL and
OpenMusic1 since 2009, which are gathered today in the jv-components library.2
Metamorphosis of numeric profiles

When confronting CAC for the first time it is important not to take the available
techniques for granted but instead to try to question them and adapt them to ones own
musical language, which sometimes implies reformulating them entirely. We have today
at our disposal a large number of libraries—some available for OpenMusic, PWGL,
and other environments as well—that are, first of all, a reflection of the musical thinking
of those who conceived them.3
When I started to study CAC I became especially interested in the Profile library.
It offered on one hand a graphical approach to musical parameters—through the use of
curves or break-point functions (bpfs), very seductive for someone familiar with envelopes
in Max or audio sequencers—and on the other hand the concept of musical interpolation.
In the 1980s and 90s many composers confronted this idea or variants of it, sometimes
following very different approaches: Tristan Murail, Brian Ferneyhough, Kaija Saariaho,
Magnus Lindberg, Marco Stroppa, and others. The commitment of the Spectral School
to the concept of process is of course a precursor, and interpolating chords or spectra has
become a common introductory example of CAC. I was fascinated by these techniques
and the music that resulted; however I always had troubles recognising their effectiveness
in perceptual terms when applying them myself.
Among the techniques I encountered, interpolations based on the regular sampling
of a line drawn between two values were the most problematic because they ignore the
notion of musical function. In the most basic situation, an isolated note inside a list of
pitches representing a chord will not be connected with the closest one in the next chord,
but the one that has the same position in the list. Hence the unavoidable generation of
leaps or crossings, notions absent from the identity of either chord (see Figure 1).
1I discovered computer-aided composition with PWGL, and was immediately seduced by the possibilities
of constraint programming. During my year in IRCAM’s Cursus, I felt instantly on familiar ground
when I discovered OpenMusic, as both OpenMusic and PWGL share the same family tree with
PatchWork [2]. Today, I continue to use PWGL on a daily basis, for reasons of habit and comfort,
but also because constraints are fundamental to the way I work. Yet I find OpenMusic especially useful
to work in conjunction with other IRCAM software such as Audiosculpt, for instance to extract data
from SDIF files, and it is now natural for me to have both programs in my toolbox.
2 This library was originally a part of my masters research on musical morphogenesis and transposition
of non-musical materials into music with the help of CAC [8]. Some of these applications are already
available in OpenMusic; the rest of the sources are completely open and can be adapted to any Lisp-
based environment.
3 Emblematic examples are Esquisse (and later OMTristan, initially developed by Tristan Murail and
Laurent Pottier in PatchWork, then ported to OpenMusic by Rozalie Hirs); Combine developed by
Brian Ferneyhough and Mikhaïl Malt for PatchWork, then OpenMusic; or Profile initially developed
by Jacopo Baboni Schilingi, Nicola Evangelisti and Mikhaïl Malt for OpenMusic, then adapted and
augmented as JBS-Profile in PWGL.
94
Figure 1. Interpolations of two chords in five steps with different list ordering: (60 62 67) to
(49 60 70) vs. (62 67 60) to (49 60 70).
If we take the example of lists of values evolving through time, representing for
instance melodies or more complex objects, it is even more clear that the forms of the
start and endpoint are not taken into account during the process. Even if the notion of
an element or point in those lists is clearly defined by the algorithm, this is not the case
for the relations between these elements (in the Gestalt sense [6]), for possible groupings
that we might take for granted, and especially for functions (musical, formal, etc.) that
these elements can hold in our perception within a given object. Interpolating such forms
in the standard way amounts necessarily to making a sort of “cross-fade” between them
(see Figure 2).
The first requirement to achieve a convincing metamorphosis is to find how to match
characteristic points or features—holding particular functions, including perceptually
salient features—of two distant morphological structures.4 For instance, we might want
to transition from an elementary form, containing few salient features, to another one
much more complex and articulated. In this case it is necessary to determine which zones
precisely will be matched and which details can emerge in a more unpredictable way (see
Figure 3).
The first tool I present addresses this issue. It is an algorithm for supervised interpo-
lation inspired by morphing techniques in image processing. The most obvious example
is a transformation of one human face to another that maintains, throughout the steps
of processing, consistency in the location of the different parts of the face—contributing
to the realism of the result. This coherence is ensured by the fact that different features
(eyes, nose, ears, etc.) are preliminarily identified in the images by a supervising operator
who remains active during the process.
4 Comprising all the properties that can used to describe the identity of an object as well as its form
(elements and relations between them, functions, contrasts, directionalities, intensities, etc.) and that
can be identified at various scales in space and time.
95
Julien Vincenot
Figure 2. Simple interpolation of two curves, regularly resampled, in seven steps. OpenMusic
patch (top) and the seven interpolation steps (bottom).
Figure 3. Matching features arbitrarily between two morphological structures.
In our case, dealing with bpf profiles, we need to start with a first step of seg-
mentation. This step can be achieved in different ways: automatically—unsupervised
classification, detection of transients or unstable zones, etc.—or by hand, in an arbi-
trary or exploratory way—by putting markers along the objects we want to morph (see
Figure 4).
96
Figure 4. Placing markers along profiles for segmentation prior to interpolation (for this
example, both profiles are cut into 11 segments).
At this point the profiles can be interpolated according to the raw segments that are
defined by the markers. But it is also possible, and even more interesting, to gather
some of these first-level segments into subgroups. The interface I developed in PWGL
(shown in Figure 5) allows one to compose, through the use of parentheses around indices
representing the segments, the zones that will be put into correspondence during the
interpolation process. This operation is indispensable when the two initial profiles of
the interpolation contain a different number of segments: in this case the problem can
be bypassed by creating an identical number of subgroups. Therefore it is possible for
the user to elaborate precise, realistic metamorphoses or totally unexpected ones by
determining in parallel the nature of the segments and the groups to which they belong.
Figure 5. Placing markers along a profile for segmentation and grouping segments prior to
interpolation (in PWGL). In this example, a profile is cut into 11 segments then regrouped in 5
parts. The two bpfs at the bottom show the second and fifth groups, gathering segments with
indices (1 2) and (7 8 9 10) respectively.
The engine of the supervised interpolator that I developed then iterates on the
successive subgroups determined by the user. Two standard interpolators5 are employed
to deal with both the form and the length of the profiles through different phases
of resampling. Each step of the metamorphosis is then flattened—the subgroups are
discarded—and the result can be visualised directly as a set of bpfs.
5 Using the well-known interpolation function from PatchWork, OpenMusic, and PWGL.
97
Julien Vincenot
In Figure 6 we can see two possible interpolations made with this approach, corre-
sponding to different groupings of the segments in the same pair of profiles. The first, on
the left, might be more obvious if we take into account the relative similarity of features
in both profiles. In the second most of the target profile emerges, progressively, from a
single segment of the starting profile. Both propositions are valid in the sense that they
were actively “composed” with the system.
Figure 6. Two possibilities of metamorphosis based on the same segmentation but with different
groupings.
I experimented with this process for the first time while writing Ascidiacea for solo
bass flute (2010), whose formal skeleton strictly follows the metamorphosis of a relatively
simple figure—a re-encoding, as a bpf, of a musical phrase written by hand—toward a
more complex morphology extracted from arbitrary non-musical material.
In this context, my approach to the profiles was a bit different from what we might
be used to when manipulating such graphical objects: the amplitude of the curve was
not taken as the continuous expression of a given musical or sonic parameter, but as a
discrete representation of contrast inside a musical object (see Figure 7). The value 0
represents a given category of musical materials sharing some common aspects; value 1
another category, and so on. After evaluating the metamorphosis itself intuitively, as
shown previously, I discretised all the successive steps into a number of categories.6
The piece strictly follows the plan given in Figure 8, which must be read one line
after the other: the first line being the initial phrase (shown in Figure 7), the second
following it immediately, and so on. The plan is an interpolation between the phrases 1
and 7. The connected fragments, belonging to different phrases, are shown by vertical
dashed lines. Similar categories (obtained after discretisation of the profiles) are labelled
with the same letter. Starting from this, I determined the nature of the different musical
materials, their evolution throughout the piece, and their mutual relations: resemblance,
discontinuity, amplification, dissipation, budding, hybridation, corruption, absorption,
etc.
6 This operation was performed with the class-num (previously class-1 ) function from the Morphologie
library. This clustering function, similar to k-means, allows the classification of a given number of points
(in any number of dimensions) by optimising progressively the position of n centres whose quantity is
determined by the user. For instance, if I want to classify the list (8 5 8 4 5 2 3 2 9 10) into three
categories, the function will easily stabilise itself around the following classes: (0 1 0 1 1 2 2 2 0 0).
98
iv
Ascidiacea
morphological creature for bass flute
Julien Vincenot
ó ñ ó ï ó ï
“anomaly” only key clicks
? ??
bisb. più forte possib.
???
bisb. ad lib.
(timbral trill)
(slow) (moderate)
q ! 50-60 (fast) 3 fingerings 2 fing.
, ≈
pizz. (molto secco)
q!
œo o
freely alternate between idem
n. vibr. p. vibr. n. vibr.
o
pitch and air sound
nœ #œ
& Lœ œ™ ÍÍÍÍÍÍÍÍÍÍ œ ÍÍÍÍÍÍÍÍÍÍÍÍÍÍ œ O #O nœ ÍÍÍÍÍÍÍÍÍ ˙ ÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍ œ ‰
Jœ nœ œ™ ˙ B> #> n> O ¿ ¿ ¿ ¿ ¿ ¿ ¿ ¿ ¿
ÆJ
bœ b>
Æ
Bass Flute
œ >
> > >
ppp p mf p ppp p mp o
3
2 2
1 1 1
ó ï ñ ó ï
0 ¿
bisb.
?? ¿ 0
?? ? ???
µ ≈
bisb. (slower possib.)
, œo
æ ≈ its re-encoding
2 fing. 3 fing.
# œo
sempre frullato
6
œæ ≈ œ #œ O
inhaling 5 n. vibr. vibr. n. vibr.
™ ÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍ œ ÍÍÍÍÍÍÍÍÍÍprofile.
œ™
r
2
#œ
bœfi œ™
#œ
#. #. #. #. #œ
œ ÍÍÍÍÍÍ œ ÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍ
5
& Figure 7.#œ nœ Initialj phrase #œ of Ascidiacea for bass flute and
œ œ œ ÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍ œ™ ÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍ œ ÍÍÍÍÍÍ œ ÍÍÍÍÍÍÍÍÍÍÍÍÍ œ™
as a nœcontrast O
J
The gesture with pizzicati was added later to the score as an intentional
p pp p
3
mp
“anomaly”. The final poco
pp mp pp
sfz
silence seems also to contradict the properties of class 0—a choice I cannot remember or explain
today.
ñ Phrase 1ï ñ ó
only key clicks
Ÿ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
ff
??? ? ??
bisb.
ad. lib.
, , ≈ ,
p. vibr.
q™!
vibr. n. vibr.
°
# œo
frullato
#œo ææ ææ #œo
3
&
12 seconds
nœ ÍÍÍÍÍÍÍÍÍÍ ÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍÍ œ ÍÍÍÍÍÍÍÍÍÍÍÍÍ œ™ <µ œ> ‰ #œ
#O nœ #O œ™ œ < µ œ> ¿ ¿ ¿ ¿ ¿ ¿ ¿ ¿ ¿ ¿
>
#œ
µœ
#œ œ #O bœ #. #. #. #. #. #. #. #.
RÔ >
mf o poco
sfz
p ppp
poco
sfz
pp
fij
t. ram
ì ì
pizz.
Phrase 2
¢& 16 seconds
¥
> nO
mf #
ff
the pause is not necessary if the tongue ram

can be achieved without interrupting the key clicks
(this applies for every similar situations)
Phrase 3
19 seconds
Phrase 4
14 seconds
Phrase 5
14 seconds
Phrase 6
22 seconds
Phrase 7
23 seconds
Figure 8. Formal plan of Ascidiacea, showing the connections between similar classes of musical
materials throughout the piece.
The elements of the initial sequence—gestures in grace notes, sustained notes, har-
monics, key clicks—were therefore used as a starting point to determine the material of
the following phrases. The different instances of materials belonging to the same class
were consequently written with similar instrumental techniques. Beside these primary
elements, others appeared as the initial figure became more complex, giving me the
possibility to imagine other categories of materials or to hybridise existing ones. Outside
of these recurring categories a certain number of “anomalies” were added locally, not
directly related to the structure given by the interpolation.
99
Julien Vincenot
Constraint-based music generation

We will come back later to the problematic of musical metamorphosis. Now a small
detour to a domain that is fundamental to the field of computer-aided composition:
the generation of music using constraint-based systems. We can describe this as the
exploration, with a search engine, of a set of possible values (or candidates) called a
search-space. Each evaluation of the engine builds up a result, picking up candidates
from the search-space, that necessarily respects the constraints (or rules) given by the
user. Some of these rules might be strictly applied (hard rules or true/false rules) while
others might on the contrary be more flexible and applied as much as possible according
to a specific weight (soft rules or heuristic rules). A common example is shown in Figure 9:
starting from a chromatic scale of possible pitches, generate a melody of n notes, using
only specific pitches and intervals, following as much as possible an arbitrarily drawn
profile, and containing as few repetitions as possible.
Search space: 15 notes chosen within a chromatic scale

from MIDI note 48 (C3) to 72 (C5)
& œ #œ #œ œ œ œ nœ œ œ nœ
œ No rule
? œ #œ œ œ
Follow a drawn
& œ
œ #œ œ œ œ #œ
profile (heuristic)
? œ œ œ # œ œ œ œ # œ
œ
& œ #œ œ #œ No repetition
? œ œ œ # œ n œ œ #œ n œ œ #œ (heuristic)
Allowed intervals
œ
& œ #œ œ œ
? œ œ #œ n œ #œ œ n œ œ #œ #œ (1 2 5 7, up or down)
& œ œ œ œ Allowed pitches

? œ œ œ
œ œ #œ œ
#œ œ œ (0 2 4 6 7 9 11,
modulo 12)
œ
Figure 9. Several melodic generations computed with the PWGL constraints solver (PMC
engine) adding rules progressively.
The first significant applications of constraints in CAC were PWConstraints7 and

Situation [3]. Today there is a renewed interest in such systems for musical applica-
tions, in particular with the bach library for Max (bach.constraints and cage.lumbricus)
and the heuristic approach of the Orchids program (and before it Orchidée [5] and
ATO-MS). I first approached this family of tools with the PMC (Pattern-Matching
Constraints) engine of PWGL. The first step was to understand how to use and “feel” the
function of existing rules, especially heuristic ones, to obtain a meaningful result. That
7 PWConstraints consists of two engines, PMC and Score-PMC. Both were developed for
PatchWork [7] then expanded into PWGL. PMC was brought to OpenMusic by Örjan Sandred
as part of the OMCS library.
100
is, to try to understand something one might want to hear but without knowing how to
write it: only how it should and should not be. In this sense constraints are exploratory
means to create new musical materials. But the true challenge was to formalise new rules,
to learn the language to write them (in this case an interface in Lisp), and above all to
realise the logic behind the very notion of a rule. This means understanding the difference
between musical properties (of single elements) and relations (between elements) and how
to constrain both on a local, detailed scale, and on a more global, morphological scale.
The jv-components library contains numerous examples of rules I designed over the
years. Some of them concern interval structures, others control metronomic modulations
(inspired by the works of Elliott Carter and José Manuel López López). A few of
them focus on instrumental idiomaticity, allowing a given result, already constrained by
many musical rules, to adapt its complexity to the technical restrictions of one specific
instrument: maximum spacing between fingers, impossible notes, multiphonic chaining,
etc. The last category of rules I investigated are morphological rules, which allow one to
generate results that reproduce the global behaviour of another musical object or model,
with a variable degree of similarity, after the extraction of its morphological properties.
Why is it necessary to valorise these techniques today? The first reason is they are
yet not well known and especially not taught in composition classes. Second, the speed
of curent machines trivialises a number of usual techniques to the point where they are
now computable in real time. On the contrary, using constraints, especially as the level of
exigency increases, remains relatively incompatible with real time and relies intensively
on machine resources. Despite an interactive dimension—with constant feedback between
the definition of the search space, the setting of rules, the result, and its perception—
constraints still allow access to a form of “slowness”, generating material with a “rarity”
which otherwise seems to have disappeared. This rarity, even if it is controlled by the
user who defines the rules to be applied, often leads to unexpected and surprising results.
Another aspect of constraint systems worthy of interest is that they can help to
understand and control the notion of musical style, whether it concerns music from the
past, today’s music, or especially one’s own style. In particular it is possible to extract,
from a given musical entity, not necessarily the rules that were used to write it but the
rules that might have been used.8 This kind of knowledge can be precious to understand
the writing techniques and formalisations that a composer learned or designed him or
herself—his or her craftsmanship—and what resulted from accident and artistic intuition.
The extraction of rules from a given musical object is also a powerful asset for music
composition. Of course, in the field of creation, it may not be much more interesting
to generate hours of music “in the manner of” than it was previously to do it by hand.
But it is more expressive to pull such rules out of their initial musical context and apply
them to a new one where the properties they convey are more meaningful. This applies
also within the work of a composer. I noticed for instance, from one piece to another,
that I had a tendency to reuse a certain number of rules in specific settings while others
disappeared after a while. Rules have become a way to remain consistent within my own
work, even while keeping for each new project a significant margin for experimentation.
8 The information might be lost, or maybe the composer had no conscious notion of any rules while
composing. In this respect analysis is necessarily a posture of construction of a specific point of view,
or even a posture of creation.
101
Julien Vincenot
In my work today the privileged system for musical exploration and composition with
constraints is the Cluster-Engine library conceived by the composer Örjan Sandred.
It succeeds his previous libraries OMRC (for OpenMusic) and PWMC (for PWGL),
themselves following premises from Laurson’s PWConstraints. This new engine is
remarkable for various reasons. It allows one to control several musical voices in parallel
by using a network of semi-independent constraint engines—one for each parameter of
each voice, including metrical structure, pitches, and rhythms—that can solve a musical
problem collaboratively. Therefore the rules, through the use of accessors of various
scopes, can be applied to each one of these parameters, between parameters within a
single voice, and of course between voices. Backtracking 9 procedures were also optimised,
allowing a spectacular gain in speed and flexibility compared to previous approaches. As
Sandred explains:
The intricate network of dependencies that exist in a score makes it hard to develop
a good strategy for how to find an acceptable solution. After having struggled
with some of the existing constraint solving systems, it became necessary for me
to develop a new, more efficient constraint solver that could handle more complex
musical formalisations. I needed a system that could approach the problem from
an angle where the nature of a music structure is taken into account. [10]
An example of this awareness is found in the vast possibilities left to the composer while
determining the search space. Here it is constituted by separate domains, one for each
parameter. This way we can define which time-signatures, notes or chords, or durations
values are allowed. But much more interesting is the possibility to insert already well-
formed musical entities or motifs. A list representing a rhythmic motif, a pitch motif, or
even an intervalic motif (freely transposable by the engine) might be, as much as single
isolated values, picked up and controlled by rules thus reinforcing the emergence of a
musically consistent result.
In Figure 10 we can see the generation of a simple result in PWGL using the cluster-
engine for one voice with such motifs. Only one rule is applied, r-only-m-motifs provided
by the library, which makes sure that only transposable pitch motifs will be selected to
build the result (except for the initial starting point, MIDI note 60 in this case). The
rhythm domain includes two isolated values (1/4 and -1/8) and five different motifs.
The pitch domain defines a mandatory starting point and five transposable pitch motifs,
starting with the “m” symbol and followed by a series of intervals. It is interesting to pay
attention to the last two motifs that each comprise only one element, a list of intervals.
This allows one to define transposable chords, which could also appear within a larger
list of intervals. For instance the motif (m (-2 5 5)) starting from MIDI note 60 would
yield, after a descending major second, the chord (58 63 68).
I used this engine intensively for the composition of my recent pieces silent_data
_corrupt for alto saxophone and live computer (see Figure 11)10 and Mémoire de l’eau
9 Backtracking is the capacity for a constraint engine to move backwards during computation and rewrite
a part of a solution that might contradict the given rules. Within the Cluster-Engine, backtracking
was optimised by the use of backjumping: instead of erasing the solution step by step to try and find
the error (classic backtracking), the engine “jumps” directly to the variable which failed a specific rule,
corrects it, and then resumes the computation.
10 silent_data_corrupt(2014) was presented at the end of my year in IRCAM’s Cursus, in collaboration
with saxophonist Nikita Zimin.
102
Figure 10. Rhythm and pitch domains using motifs (in PWGL).
for accordion and live computer (see Figure 12).11 In both cases I had in mind to generate
virtuosic passages inspired by improvisation in general and free jazz in particular. To
reach this result I paid specific attention, even before defining the rules, to the elaboration
of complex search spaces that comprised freely transposable melodic motifs (and chords
in the case of the accordion). In particular, I carefully designed the representations
of my transposable chords, allowing efficient and idiomatic chaining for the accordion
player. Already at this point, without defining any rule, the results were promising: the
candidates were already determining a musical context as fully-fledged rules would do.
Figure 11. Excerpt from silent_data_corrupt for alto saxophone and live computer.
Figure 12. Excerpt from the introduction to Mémoire de l’eau for accordion and live computer.
11 Mémoire de l’eau (2015) is the result of a collaboration with accordeonist Yohann Juhel.
103
Julien Vincenot
After trying to understand my intentions more precisely and listening carefully to

successive evaluations, I empirically applied a set of rules that I would divide into
two categories. I will take the specific case of Mémoire de l’eau, which implied the
use of two voices (one for each manual), as the process was similar yet simpler for
silent_data_corrupt. The first set of rules was related to idiomatic performance:
• Restrict to allowed register for each hand;

• Avoid local repetitions as much as possible (heuristic rule);
• For rhythmic values faster than a eighth note, allow the pitch engine only to select
a chord if it is repeated one or two times with identical rhythm (equivalent of a
“bellows shake” for the accordion); otherwise choose a single note;
• Etc.
The second set of rules was dedicated mostly to rhythm, and also controlled some aspects
of the pitch engines relative to the rhythmic context:
• Forbid consecutive silences;

• Forbid the nth consecutive thirty-second note;
• Force rhythmic hierarchy between voices (the right hand must follow the left hand’s
basic grid and add details on top of it);
• In some contexts, if the current rhythmic value of the right hand is slower than a
sixteenth note, then force the corresponding pitch engine to choose a chord;
• Etc.
Metamorphosis of symbolic sequences

More recently I came back to the problematic of interpolation and metamorphosis, but
from a different angle. Instead of following the usual numerical approach of interpolation,
I investigated a purely symbolic criterion from the edit distance algorithm (also called
Levenshtein distance). I discovered this algorithm at the same time as the Morpholo-
gie library, which gathers various approaches for the analysis of sequences of symbols,
including of course musical materials.12 As noted by Voisin, edit distance is not exactly
a distance but:
[...] a measure of dissimilarity, based on the search for the minimal cost of the
transformation of one sequence into another, by means of the following three op-
erations applied step by step: substitution of one symbol by another, insertion of
a new symbol, deletion of a symbol. Thus, for example, the edit distance between
the list (C D E G A G G) and the list (E D E G A G A G) equals 3, the minimal
number of operations needed for the transformation. [9] (author’s translation)
12 Morphologie was initially developed in 1997 by Frédéric Voisin and Jacopo Baboni Schilingi for
OpenMusic, then recast for PWGL and any Lisp environment in 2009.
See http://www.fredvoisin.com/spip.php?article28
104
How does this apply concretely to our problem, the metamorphosis of musical entities?
The original edit-distance function of the Morphologie library, following the algorithm
as expressed by Levenshtein, only returns the minimal cost to transform one sequence
into another but does not specify the successive steps of the process. This is the purpose
of the function s-interpolation 13 from the library jv-components, which can deduce the
possible path. In most cases there are different possible minimal paths among which the
result is selected randomly. An exemple of a path between the two words “AMPHORA”
and “AMORPHA” is as follows:
((AMPHORA)(AMOPHORA)(AMORPHORA)(AMORPHRA)(AMORPHA))
It is interesting to note that going from one word to the other or back can produce slightly
different results. This is a consequence of the order of operations in this implementation,
which always starts from the beginning of each sequence:
((AMORPHA)(AMRPHA)(AMPHA)(AMPHOA)(AMPHORA))
This algorithm allows one to move from one sequence to another in a relatively
convincing way, whether acting on pitches, durations, or any other parameter. The
result is all the more relevant, from a perceptual point of view, as common elements are
maintained between the successive steps. The interpolation is “informed” by the simple
nature of the algorithm: it does not add, delete, or replace elements by accident but
takes into account both the content and form of the initial sequences.
I used this simple approach to generate progressions between different melodic frag-
ments for Mémoire de l’eau (see Figure 13). Unlike numeric interpolation, this approach
does not allow one to decide a priori how many steps will be returned between the
two sequences: the number of elements is necessarily equal to the edit distance plus 1.
I therefore manually selected afterwards—as one often does with classic interpolations—
the steps I considered to be more musically interesting and relevant.
The advantage of a purely symbolic approach is the possibility to morph sequences
containing any kind of musical materials, as long as they can be encoded as symbols
(melodic or rhythmic motifs, chord sequences, etc.) or sequences of symbols representing
more complex materials at a higher level. All depends on the encoding chosen to represent
these materials or forms. And if we use the appropriate predicates for testing equalities,
even lists of any length and nesting level become symbols in their own right.14
To this approach I have applied the concepts that were already expressed in my
algorithm for supervised interpolation dedicated to numeric profiles (see Figures 5 and 6).
13 Thisfunction is a translation I made from Javascript to Lisp of the algorithm proposed by Jean-Paul
Davalan on his website. The section dedicated to the Levenshtein distance is particularly informative
and allows one to experiment with the algorithm directly from a web browser.
See http://jeux-et-mathematiques.davalan.org/
14 The = function in Lisp will accept for its arguments only values of type number (mathematical
equality). Equal or equalp, on the other hand, will allow us to check for instance whether symbols or
lists (even nested ones) are identical or not. Lisp is particularly well-suited to express any properties
or relations in the form of custom predicates. The jv-components library contains some proposed
predicates based on morphological properties. For instance, the simple function close-p tests if the edit
distance between two lists is inferior to a given threshold; that is if these two lists are “close enough”
to be considered approximately equal.
105
Julien Vincenot
Figure 13. Melodic process using symbolic interpolation (Mémoire de l’eau for accordion and
live computer). The different steps of the interpolation are identified by beaming.
This previous experiment demonstrated another important point: performing a realistic

metamorphosis requires different simultaneous transformations, in parallel, on different
aspects of the materials. Given that the classic interpolation function, irrelevant in my
view for effective metamorphosis, could be used as a starting point to build a higher-
level algorithm, a symbolic interpolator based on edit distance might become itself the
building block for a more complex criterion of supervised symbolic metamorphosis.
This new approach I propose exists for now only as a PWGL patch. As previously, a
first step of segmentation is needed. This can be done arbitrarily or with the help of any
criteria of analysis that might seem relevant for the purpose. In the example in Figure 14
I segmented two melodic profiles by hand with the help of the group-list function. Similar
to numeric profiles, it is necessary to obtain an equivalent number of subgroups at the
end, but with one important exception: the value NIL (the empty list), itself a symbol,
may now be inserted among subgroups as a “wildcard” when no obvious correspondence
can be made with a fragment from the other profile. In this example I inserted two NIL
values (or 0 in the group-list function parameters), one as the sixth fragment of the first
profile, the other as the fifth fragment of the second profile. This means any subgroup in
front of a NIL value will emerge or “bud” from nothing, or “dissolve” itself to nothing,
depending on whether it is in the first or the second profile.
The next step is the choice, for every pair of segments, of the mode of symbolic
interpolation. As mentioned earlier, the process to obtain the successive sequences of
symbols is oriented from the beginning to the end of the sequences, which means we can
obtain completely different results in morphing from sequence A to B or back. Mode
0 describes the original interpolation from A to B, and mode 1 is the reverse of the
interpolation from B to A. The role of the update-modes function in Figure 14 is to
generate a random sequence of modes according to how many segments are in place, a
first proposition which can be modified later inside the text box.
The main function, supervized-interp-sym, iterates through the two lists of segments
and operates the s-interpolation for each pair in the corresponding mode (original or
reverse). As each process might end up with a different length (because of the edit
106
Figure 14. PWGL patch for the supervised symbolic interpolation of two melodic profiles.
distance paradigm), we need as a last step to resize each according to the longest
one. This implies repeating some steps between certain pairs, in the most homogeneous
manner possible, in order to obtain exactly the same number of steps for each process.
Figure 15 shows a simple example of such a metamorphosis between to melodic profiles,
highlighting the different parallel interpolations.
Figure 15. Example of a supervised symbolic interpolation of two melodic profiles.
This approach, as mentioned before, can be applied to any kind of musical parameter
or even to complex encodings aggregating multiple parameters. In the specific case
of rhythm the results are already promising, perceptually speaking, especially when
interpolating simple lists of durations. Still, in the metric domain the process often
107
Julien Vincenot
generates irregular sequences that can be difficult to notate: by adding, deleting, or

replacing symbols the hierarchy between on-beats and off-beats is often distorted, and
incomplete tuplets might appear.
The next step to improve this algorithm will be to integrate constraints in the process,
to control the musical properties of the result more carefully. I am also currently
investigating the “Levenshtein-Damerau” variation of the edit distance which considers,
among the usual operations of substitution, insertion, and deletion, the local permutation
or swapping of elements between two sequences.
Conclusions
Within the field of computer music today, computer-aided composition as we know it is
going through a phase of transition with the emergence of new tools oriented toward real-
time, interactivity, and performance like the bach library for Max [1], OpusModus,15
or the new “reactive mode” in OpenMusic [4]. All of these recent innovations are
stimulating and we cannot yet imagine the applications that will result: especially for
the development of open/interactive scores and forms in composition, and for interactive
installations in particular.
At the same time, for those who believe in musical writing, and especially in a possible
association of writing with technology, it is necessary not to allow the disappearance of
the knowledge and techniques accumulated on the side of deferred-time in the domains
of composition, analysis, and formalisation in general. For me the applications of tools
like OpenMusic or PWGL and the reasons to keep them in our arsenal are numerous.
First of all is the possibility to work in parallel with various modes of representation
(scores, bpfs, maquettes, sound files, etc.) and to organise all of them inside the
topological environment of the patch. There is also the flexibility and expressivity of
the Lisp language which, even with its age, allows custom modes of representation
and manipulation that are still unattainable with a Max patch, for instance. And
finally there is the capacity of these tools fully to exploit the machine’s computing
resources to run complex (and often richer) musical algorithms, and not only a graphical
interface. For my own work, CAC is still a privileged way to introduce materials
derived from sound or extra-musical models in composition, to try to solve the conflict
between algorithmic complexity and perception of the listener, to favour expressivity and
instrumental idiomaticity without sacrificing complexity and experimentation, to link
instrumental and electronic composition, and even to think of instrumental composition
as electronic composition.
One might remember that these tools, which were designed for the fast temporality
of music production, are still among the rare tools allowing a focus on slow temporality:
that of composition—but also of programming, interpretation, analysis, and critique—in
other words, of musical thinking.
15 http://opusmodus.com/
108
References
[1] Andrea Agostini, Daniele Ghisi. “Real-Time Computer-Aided Composition with
bach”. Contemporary Music Review, 32(1), 2013.
[2] Gérard Assayag, Camilo Rueda, Mikael Laurson, Carlos Agon, Olivier Delerue.
“Computer-Assisted Composition at IRCAM: From PatchWork to OpenMusic”.
Computer Music Journal, 23(3), 1999.
[3] Antoine Bonnet, Camilo Rueda. Situation. IRCAM documentation, Paris, 1999.
[4] Jean Bresson, Jean-Louis Giavitto. “A Reactive Extension of the OpenMusic Visual
Programming Language”. Journal of Visual Languages and Computing, 25(4), 2014.
[5] Grégoire Carpentier. “Global Constraints in Orchestration”. In Charlotte Truchet,
Gérard Assayag (eds.) Constraint Programming in Music. London: Wiley-ISTE, 2011.
[6] Paul Guillaume. La psychologie de la forme. Paris: Flammarion (1999), 1937.

[7] Mikael Laurson. PATCHWORK: A Visual Programming Language and some Musical
Applications. Ph.D. thesis, Sibelius Academy, Helsinki, 1996.
[8] Julien Vincenot. Analyse morphologique d’objets extra-musicaux et Morphogenèse en
écriture instrumentale assistée par ordinateur. Master’s thesis, Université Paris 8,
Saint-Denis, 2011.
[9] Frédéric Voisin. “Dissemblance et espaces compositionnels”. In Actes des Journées
d’Informatique Musicale. Saint-Etienne, 2011.
Online
[10] Örjan Sandred. “Computer Assisted Composition – Background”.

http://www.sandred.com/CAC.html.
109
OM-Darwin: Generative and
descriptive aspects of genetic
algorithms
Geof Holbrook
In this chapter I present OM-Darwin, an OpenMusic library that implements genetic

algorithms. Designing a patch using OM-Darwin involves a strategic division between
generative procedures (the translation of genotype to phenotype) and description of
qualities (the fitness function). This chapter looks at how those aspects relate to each
other in the composition of Future Perfect (2010) for ensemble, and how the library
facilitates the design of genetic algorithms with a simple yet extendable interface.
Introduction: two approaches to CAC

We can distinguish two strategies for creating musical material using computer algo-
rithms: descriptive and generative. By the descriptive approach, we control what prop-
erties the output must have. By the generative approach, we control by what process the
output of the algorithm is produced.
The descriptive approach requires an optimisation strategy such as an exhaustive
search, constraint programming, hill climbing, or genetic algorithms: all of which involve
evaluating and comparing a large number of candidate solutions. The generative ap-
proach is taken by a great variety of computer-assisted composition (CAC) techniques:
stochastic methods, frequency-based calculations, combinatorial processes, etc.
Of course, it is possible and in fact common for these two strategies to be combined
in a single application. In a typical patch that uses OM-Darwin, the two are divided
conceptually, and also in a visual way. Generative procedures tend to appear on the left,
and descriptive procedures appear on the right (Figure 1). Under the hood, the object
that makes everything happen (ga-engine) is nonetheless executing those procedures in
rapid alternation.
A brief explanation of genetic algorithms

A genetic algorithm (GA) mimics the phenomenon of natural selection to produce an
optimal solution to a posed problem by repeatedly combining, altering, sorting, and
filtering a population of candidates. A problem is defined in two parts: the search space
and the fitness function.
111
Geof Holbrook
Figure 1. Typical OM-Darwin patch: generation of a chord sequence on the left, description
of desired properties on the right.
For our purposes, the search space is often all possible musical sequences of a given
type, with some limits on dimensions and ranges. An example of a search space might
be “15 chords with cardinalities between 3 and 5, and a pitch range of 4800 to 8400
midicents”. In OM-Darwin, I use the term species to indicate the definition of a search
space. The library provides built-in species that represent things like melodies, chord
sequences, rhythm trees, and polyphonic passages. Custom species can also be defined
by the user.
The fitness function is a measurement of a single candidate, returning a numeric value
that represents its suitability or “fitness”. The GA makes no restrictions on the nature
of this function—it can measure any property that the user wishes to control, as long
as the output is expressed as a number. The convention in OM-Darwin is that higher
numbers are interpreted as worse than lower numbers, and a fitness of zero means the
candidate is perfect (the fitness can been seen as a penalty, or distance from an ideal
state). So, the GA searches for solutions that give as low values as possible when passed
to the fitness function. If it finds one that gives a value of zero, it stops.
Each iteration of the GA might be called a “generation” in which the population
is duplicated several times with slight alterations, sorted by fitness, and then cut back
to its original size, preserving the most suitable solutions. These slight alterations are
made either by mutation (a small change to one detail of the object) or by recombination
(combining two existing candidates into a single new one). At any given time there is
a current best solution, and the current best fitness will improve over time, especially
after hundreds or thousands of generations. It can, perhaps, be seen intuitively that this
improvement will occur; but to put it more exactly, it occurs because the system has
three required features: transmission, variation, and selection [4]. For a more in-depth
discussion of genetic algorithms and evolutionary computing, see [2].
112
OM-Darwin: Generative and descriptive aspects of genetic algorithms
OM-Darwin basic operation

The genetic algorithm in the OM-Darwin library runs inside the ga-engine object. The
instantiation of this object initialises the population and sets the species and fitness
function; because of the amount of calculation involved, I thought it better for the GA to
run in a separate process, so one can continue to use OpenMusic while it is operating.
The “g” key launches the GA, and while it is running the score display constantly updates
with the current best solution.
Figure 2. Another simple OpenMusic patch using OM-Darwin.
In Figure 2, we can say that the two rightmost inputs (the leftmost <self> input of
ga-engine is seldom used) represent the generative and descriptive sides of the algorithm,
although there is not much in the way of generative processes going on here. Each solution
is simply a sequence of chords, with dimensions and ranges given as arguments to the
built-in make-ga-chord-seq function. The fitness function (detail in Figure 3) makes a
measurement of the sensory dissonance of each chord, using a function written by Sean
Ferguson [5]. It returns a sum of differences between the measurement and a goal value,
provided as an argument. Running the GA will produce solutions whose dissonance
values get progressively closer to the goal value, as this sum of differences approaches
zero.
This example is instructive, as it demonstrates two common features of fitness func-
tions for musical objects: 1) the application of the same measurement over the elements
of a given structure (chords, in this case) and 2) a measurement of proximity to a
desired value (the absolute difference between the actual and target values). Because
these and other kinds of descriptive strategies are common, the library includes a set
of helper functions called criteria. There are different criteria for iteration over notes,
chords, melodic intervals, simultaneously sounding pairs of notes, etc. They can be given
arguments for expressing various kinds of static or dynamic goal values, specifying rates
of occurrence, or supplying a custom measurement function to be applied to individual
elements. Criteria can be combined by joining them into lists, which amounts to adding
their outputs to determine an aggregate fitness value. Looking back at Figure 1, one can
see the dissonance function combined with some other measurements, expressed using
the built-in functions, to create an aggregate fitness function.
113
Geof Holbrook
Figure 3. Inside the fitness function patch from Figure 2.
Generative mapping functions

An optional feature for a GA—the one that OM-Darwin uses to integrate with gener-
ative procedures—is the mapping function. It means that there is a distinction between
the genotype, or “code” for a solution, and the phenotype, which is the solution in its
decoded form. The mapping function converts a genotype into a phenotype, which must
be done any time a candidate is altered by mutation or recombination. The important
distinction to understand here is that the slight alterations done by the GA operate on
the genotype, yet it is the phenotype that is measured by the fitness function.
A genotype in OM-Darwin is always a list of integers of fixed length. This uniformity
is necessary for the mutation and recombination algorithms to work, without the user
having to create a specialised function to do the job. However, the mapping function
can be anything—it can produce a phenotype of any structure, with an arbitrary level
of complexity. The user can devise such a mapping function in order to create a custom
species that has a more specific structure than the built-in species (chord sequences,
rhythms, etc.)
Consider a patch that uses random elements to generate a musical sequence—perhaps
it contains several instances of the native OpenMusic functions om-random or nth-
random, which might be evaluated multiple times during one execution of the patch.
We can reinterpret such a patch as a mapping of a genotype to a phenotype, as follows:
the genotype is the list of values obtained from all of the random evaluations, while the
corresponding phenotype is the final result of the patch for that particular sequence of
random values.
The visual method for creating custom species (there is also an efficient way to do
this in Lisp code) puts this idea into practice. Any generative patch that uses random
114
values can be easily made into a species, by simply replacing all instances of om-random
with the OM-Darwin function om-gene, and replacing instances of nth-random with
nth-gene. Expressed as a function (lambda mode), and passed through define-species,
this patch can then be used to set the species argument for ga-engine, in the same way
as a built-in species (see Figure 4).
Figure 4. Basic OM-Darwin patch with custom species.
What define-species is doing, in short, is running a test on the function given to it.
It counts the number of calls to om-gene (and nth-gene), recording the inputs as well
(representing ranges of values, just like om-random). Because the convention in OM-
Darwin is that genotypes are lists of integers of a fixed length, it is important that the
patch be designed so that the calls to om-gene and nth-gene are made in a predictable
order. This is a necessary limitation of the library.
This establishes a crucial division between the generative and descriptive parts of the
algorithm. If the phenotype is defined in such a way that a certain property is guaranteed,
then that property results from a generative procedure in the mapping function. Other
properties, which have to be achieved through adaptation, are the desired outcome of the
descriptive part of the system. Seen this way, the composer has two ways of controlling
properties of the output: phenotypic controls and adaptive controls.
Take the example in Figure 4. The pitches are expressed in the genotype by their
integer rank within a harmonic series, with the mapping function translating these into
actual pitches; so, the property of belonging to a harmonic series is a phenotypic control.
That means that all candidates in the search space have this property; there is no way
for mutation or recombination to alter the fact. Other properties are alterable by the
GA if they are dependent on the specific harmonic ranks chosen—a measurement of such
115
Geof Holbrook
properties can then have an impact on the selection part of the algorithm. So, if we wish
also to control the melodic intervals of the sequence (for example), we include that in
the fitness function, making it an adaptive control.
The reversal of the assignment of these two properties to phenotypic and adaptive
controls is also possible. We might write a generative function that chooses melodic
intervals within a certain range, but then measure the membership or non-membership
of each note to a harmonic series in the fitness function. This means that there are at
least two approaches in this case, and the user must decide which is more efficient, either
by intuition or trial and error.
Melodic species in Future Perfect

A type of instrumental melody that I devised for my ensemble piece Future Perfect will
serve here as an illustration of both generative and descriptive features of OM-Darwin.
On the generative side, I invented a custom species called s-oboes that has a specific
contour and rhythmic character. The phenotype is a concatenation of melodic units, each
of which consists of a long note followed by a scalar passage leading from it. Figure 5
gives the mapping function, represented as a generative patch containing instances of
om-gene and nth-gene. Figure 6 shows an example from the final score.
Figure 5. The generator patch for s-oboes. The genes encode, from left to right along the top
of the patch: long note value, scale note values, starting note, scalar intervals, scale length, and
scale direction.
116
3 3 + 7
! ! # %! ! ! ! $ ! $! ! ! ! !
3 7
! ! ' ! % ! #! " ! !
3 3
$ ! ! $! %! ! ! '! $! ! % ! %!
* 43 !
140
( &
3
, )
,
7
# # " 5
! % ! %! +
3
# , , ! !"
3
/
5
! ! ! ! ! ! ! '!
3 3
* % !# '! ! 3 ! ! % ! !. -
143
! ! !) !
, 4 , 5
, %! %! # !3
3 5 3
" ! , ! !
5 5
, ! %! & 0 ! ! % ! ! %! ! '! '! $ !

5 ! ! ! %! % ! ! ! %!
5
! % ! % ! %!
146
* 0 # - 4 # ,
5
#" - 3
3
7 5
5
Figure 6. Excerpt from the score of Future Perfect (2010).
One of the instances of om-gene encodes the length of a scalar run, which can be
as long as nine notes; another is called multiple times to encode the melodic intervals.
Note that the latter instance is always evaluated nine times per melodic unit—but if
the run is shorter, then the extra information is discarded. If it were evaluated only
as many times as needed for each scale, then the genotype would be of different length
for different candidates, which would be nonsensical to the ga-engine box. There is a
safeguard against this: define-species makes multiple test evaluations of the generating
patch and signals an error if the genotype is not the same length every time.
The single input to this patch allows the user to decide on the number of units the
assembled melody will have. However, the ranges of the genetic functions could also be
specified as inputs to the patch. Then we would have a whole set of parameters separate
from the genotype, as we often do, which we might call species parameters. They would
be part of the definition of the species for any given trial, and could not be altered by
mutation or recombination.
Conflicting criteria in a fitness function

Genetic algorithms are an ideal strategy for reconciling competing criteria. The fitness
function used for the passage in Figure 6 is a classic example of this. It takes the form
of two simple rules, which can be applied to a sequence of tempered pitches. Given two
variables m and n, where m < n, they may be stated as follows:
1. Any consecutive subset of m notes must belong to some diatonic scale.
2. Any consecutive subset of n notes must not belong to any diatonic scale.
In order to satisfy both these rules, a melody must continually “migrate” from one
diatonic set to another in a fluid manner. It will have tonal implications, but these will
be unstable.
I applied this pair of rules, with the arguments m = 4 and n = 6, to the melody in
mm. 129-148 of Future Perfect. I also applied an additional criterion that tries to make
80% of the pitch classes belong to a C melodic minor scale, in order to lend a certain
amount of pitch focus to the passage. An analysis of the resulting passage in the score
gives an idea of the level of success of this GA trial:
Diatonic 4-note segments: 78%
Non-diatonic 6-note segments: 83%
Pitch classes in (0 2 3 5 7 9 11): 68%
117
Music engraving by LilyPond 2.19.26—www.lilypond.org
Geof Holbrook
Although the GA did not arrive at a perfect solution, the effect of these colliding
criteria is powerful enough to give the desired effect. In the best-case scenario, the GA
outputs a perfect result (a fitness of zero) in a reasonable amount of time. But it is
possible, as in this case, that a satisfactory result is not necessarily perfect.1 For a
given problem, it is actually possible that no perfect solution exists, which means that
the global optimum fitness will be some positive value. In such cases it is generally
unnecessary to find out what that value is, as there are likely many solutions which are
close enough to the optimum fitness to be acceptable, or even “new and interesting”.
However, it is also possible that the GA will fail to make significant progress toward
a decent solution. In this case there are a number of steps the composer can take: 1)
restructure the fitness function, 2) rewrite the mapping function, or 3) redistribute the
phenotypic and adaptive controls.
The distribution of work between generative and descriptive programming seems to
be entirely appropriate in the case of s-oboes, although one could imagine the phenotypic
and adaptive controls being reversed. Some method that uses diatonic fragments, along
with a pitch mapping applied to 80% of the notes, could be used to assemble a phenotype
that automatically adheres to the rules given above. Conversely, a fitness function can be
devised that measures the pattern of long and short notes and limits changes of direction
to give the quick passages a scalar quality.
The strength of OM-Darwin’s problem-unawareness

There is fourth step a user might take, if she or he is willing to get into the guts of
the algorithm: to specialise the mutation or recombination methods of the GA. There
is no visual method for doing that in the library at present, so it would require writing
new Lisp code. This would be done with the specific problem in mind—that is, the
specific structure of the phenotype and the specific properties that the composer wants.
It means that the GA will become “problem-aware”, which is in fact contrary to what
makes OM-Darwin efficient to use.
What the ga-engine needs to know in order to run, other than the length of the
genotype (always a list of integers), is really only the mapping function and the fitness
function. It does not need to know or “understand” anything about those functions; it
applies the mapping function to the genotype to obtain the phenotype, and applies the
fitness function to the phenotype to obtain a fitness value. Unless the user has the means
to tinker with internal operations, the GA is totally problem-unaware.
This gives complete flexibility to the user to design the mapping and fitness functions
as she or he pleases. Any generative patch can be combined with any descriptive patch
for use with a GA, with trivial modifications: the parameters to be optimised must be
expressed with om-gene and nth-gene (possibly in place of om-random and nth-random)
and the fitness function must be expressed as a numeric penalty.
For a difficult problem, an optimisation strategy might be replaced by a problem-
specific mathematical strategy. For an example, see [1] for a comparison of constraint
1 “The goal is to identify new and interesting solutions—normally more than one is desirable—and these
solutions must be good. However, finding global optima may be undesirable, impractical, or even
impossible” [3].
118
programming (another optimisation method) to more intricate mathematical methods

as alternative ways of creating rhythmic tiling canons. Such an approach can potentially
work better and with less computation time, but it takes some effort and expertise to
program it. In contrast, OM-Darwin allows the integration of existing patches, and a
great deal of experimentation with various generative procedures and measurements of
properties, without being concerned with the internal operations of the GA. It meets the
needs of a composer who requires flexibility, but not mathematical perfection.
References
[1] Carlos Agon, Moreno Andreatta. “Modeling and Implementing Tiling Rhythmic
Canons in the OpenMusic Visual Programming Language”. Perspectives of New
Music, 49(2), 2011.
[2] Peter J. Bentley. Evolutionary Design by Computers. San Francisco: Morgan
Kaufmann, 1999.
[3] Peter J. Bentley. “Exploring Component-based Representations—The Secret
of Creativity by Evolution?”. In Ian Parmee (ed.) Evolutionary Design and
Manufacture (Selected Papers from ACDM’00). London: Springer, 2000.
[4] Richard Dawkins. “Universal Darwinism”. In D. S. Bendall (ed.) Evolution from
Molecules to Men. Cambridge, UK: Cambridge University Press, 1983.
[5] Sean Ferguson, Richard Parncutt. “Composing in the Flesh: Perceptually-Informed
Harmonic Syntax”. In Proceedings of the Sound and Music Computing Conference.
Paris, 2004.
119
Computer-aided composition in the
creation of As I ride the late night
freeways
Matthew Schumaker
This article presents a composer case study documenting the use of OpenMusic software
in the compositional processes of As I ride the late night freeways, a work for soprano and
orchestra. The introduction lays out the work’s æsthetic environment, considering Cathy
Park Hong’s poem Census and how its theme of freeway driving can suggest composi-
tional approaches and parameters that unify the musical work. Architect Zaha Hadid’s
parametric æsthetic is also considered as a “streamlining” influence. The following section
focuses on the development of the musical processes, which take a variety of forms: from
harmonic and textural elaborations drawn from the analysis of car sounds, to a poetic
suggestion of aerodynamism, given by the algorithmic shaping of musical gestures and by
the smooth morphing of one gesture into the next. Other related OpenMusic patches
illustrate how these ideas are being developed further for use in future works.
Background and inspiration

As I ride the late night freeways 1 finds inspiration in a poem about life in California
written especially for the project by Cathy Park Hong. Taking cues from her poetic
imagery, the notion of a high-speed highway drive became the guiding idea for several
key instrumental sections described below. The notion of a car’s aerodynamism, in turn,
also suggested processes to determine the unfolding and elaboration of musical lines, as
well as the influence of the “parametric” forms of the architect Zaha Hadid.
Following the dual influence of the poem and of the parametric æsthetic, I sought to
design patches in OpenMusic to help me create music of continuous metamorphosis that
mimicked air flowing over a car and embodied the continuous flow of freeway driving.
Audio samples of speeding cars provided an aural model for musical textures and analyses
of these sounds provided the harmonic framework for the composition. To this end, I
used OpenMusic in several ways: 1) to derive harmonies and harmonic unfolding from
partial-tracking analyses of car sounds, 2) to create an interpolation of melodic lines, 3)
to draw musical lines, in a graphical sense, and then create variations by rotating these
lines, and 4) to reconstruct a musical line from a cloud of random pitches.
1 Composedin large part during 2014-2015 in Paris, where I was supported by a Georges Ladd Fellowship
from the Music Department of the University of California, Berkeley.
121
Matthew Schumaker
Parametricism and Cathy Park Hong’s Census

I was drawn to the way Zaha Hadid and her
artistic partner Patrick Schumacher use gener-
ative algorithms to create smoothly developing
buildings, which they call parametric forms
(Figure 1 shows an example). Schumacher
puts forth the modus operandi of the paramet-
ric æsthetic as one of “knitting geometries into
continuous textures” and he describes a flow-
ing, dynamic architecture in which “all forms
must be soft” and “all activities communicate
with each other” [2]. In this architecture, the
process of metamorphosis underlines contin-
uous flux and an interpenetration of ideas.
Here I was excited by both the notion of
smoothness and the idea of using generative
algorithms to create form. Hadid’s smooth
and dynamic architecture of algorithmically
aided “soft forms” influenced me to try and Figure 1. A parametric form of architect
develop morphological processes to help shape Zaha Hadid.
a smooth flow of musical lines in my own
composition.
Beyond the influence of parametricism, the Census
principal inspiration for my piece comes from Cathy Park Hong, 2015
Cathy Park Hong’s poem Census. After The sun hollows our bodies.
talking together about shared ideas, Hong Sunglass shacks melt to molten then molded
wrote a text in which the quintessential Cal- back to blackest fetish masks.
And Surfers riot, beached boys
ifornian pastime of driving provides a link in parking lots, punching down Port-a-Potties.
to the dynamism, sunny beauty, and latent They’ll go out roaring in this
violence of life in the Golden State. In this Vaseline light
poem one might see parametric “soft forms” where infinity pools sparkle like geodes.
in imagery of the freeway’s continuous flow Fade to twilight.
and of forms melting in the heat that take My shyness is criminally vulgar.
My hatred is gentle,
on a threatening character. Ultimately the as I drive the late-night freeways,
thin veneer of civilisation in California slips rampart after rampart
away as the poem alludes to the Huntington shut down,
Beach riots of July 2013, during which the Night glows with the velocity of cars
thinning to a stream
spectators of a surfing competition destroyed of white light, and sodium street lights
the main street of a quiet ocean-front town for and tennis courts
no apparent reason. holy as an ashram.
122
Computer-aided composition in the creation of As I ride the late night freeways
Partial-tracking analysis of car sounds

I came to recognise the harmonic possibilities of car sounds as a way of translating the
metaphor of California freeway driving from the poem into the music. I was drawn,
in particular, to the super-charged and aggressive engine sounds of Formula One race
cars. An important component here was the underlying sense of tonality audible in these
extreme motor sounds.
Using the partial-tracking functionalities from OpenMusic’s OM-pm2 library, I
distilled harmonic content from recordings of these cars. In particular, I created seven
short edits from a longer recording of the famous Formula One driver Ayrton Senna.2
Each one of these seven audio edits had a duration of between 3-6 seconds and carried a
distinctive trajectory in the frequency domain resulting from the driver pushing the car to
its limit through aggressive acceleration, braking, and gear-shifting. The Doppler effect
taking place as the car approaches and passes the microphone also coloured these sounds
in interesting ways. These seven sequences were further broken down, analysed, and then
reconstructed into musical sequences that could be used in the composition. Collections
of frequencies from these analyses became the basis for harmonic fields deployed in many
sections of the piece. In the first seven minutes of the piece, analyses of the seven
sequences are also used in massed glissando string passages that seek to evoke the original
automobile sound sequences.
Figure 2. Extracting frequency contents from the car sounds using the chord-sequence analysis.
Two approaches were used for each of the sound sequences: one to create static
overarching pitch fields; the second to create chord sequences that suggest the harmonic
transition within each sound. The sound files were all processed using the chord-seq-
analysis function from OM-pm2 (see Figure 2). Due to the noisy character of these
sounds, a “standard” spectral analysis would necessarily report a large amount of chaotic
2 DaitoManabe’s exciting project Sound of Honda - Ayrton Senna 1989 - (2013), influenced my choice to
use Ayrton Senna driving recordings. See http://www.daito.ws/en/work/sound_of_honda_senna.html
123
Matthew Schumaker
frequency content. The chord-seq-analysis function contends with this by grouping

frequencies within segments and allowing one to specify a maximum number of partials
(or pitches) to be chosen based on the highest amplitudes, i.e. based on their prominence.
The resulting SDIF files were then converted into music notation using the sdif-›chord-
seq function and displayed in a chord-seq where microtonal deviations were rounded to
the nearest equal-tempered pitch. From this process, several static harmonic pitch fields
were derived that were used in many sections throughout the work (Figure 3).
Figure 3. A pitch field extracted from a race car sound.
The chord sequences obtained from the partial-tracking analyses were also used to
recreate the frequency transitions and the Doppler effect of the car sounds in orchestral
string textures. The extracted chords were exported to a notation program, divided into
different lines of the divisi string section, and given durations roughly proportional to
their duration as components in the overarching sound. Glissandi were often employed
to connect these chords in order, giving the impression of the continuous frequency tran-
sitions in the aggressive acceleration or deceleration of the original car sound. Additional
sculpting of dynamics and the insertion of occasional rhythmic accents and tremolos
contribute to the orchestral emulation (see Figure 4).
Figure 4. Example string texture informed by race car sound analyses.
The music in the opening section of the piece derives its harmony from chords found
in the analyses of the seven race car sound edits. Here the chords unearthed in the
partial-tracking analyses are freely combined to form a new ordering (Figure 5). This
new sequence, in turn, plays an important structural role in the piece, appearing in
different guises as a sequence of musical pillars that signal the beginning, middle, and
ending of the work.
124
Figure 5. “Pillar” harmonic sequence freely derived from the seven race car sound analyses.
The aero line

If the dynamic thrust and pull of acceleration and braking could be suggested in glis-
sando string writing, then the wind instruments might suggest an imaginary flow of
air molecules moving in smooth aerodynamic curves up and down over the surface of
a car. The idea was to draw smooth, imagined curves and then convert them into
continuous musical lines by matching the height at any given point in the drawing with a
correspondingly high pitch from a pitch field originally derived from the race car partial
tracking analyses. In order to do this, I used the drawing features of the OpenMusic’s
break-point function (bpf) editor to fashion melodic lines. These lines were smoothed
into soft curves using the om-spline function, then resampled for the desired length of
the line using om-sample, and assigned to pitches constrained to a given harmonic field
(Figure 6).
Figure 6. Drawing a line using a bpf.
125
Matthew Schumaker
A predetermined pitch field, derived from the initial analyses of the car sounds, is
loaded into the patch as a MIDI file that was processed to yield a list of sorted pitches.
The om-scale and om-round objects are then used in series to map the samples from
the bpf into values between the lowest and highest pitches of this sorted pitch list. A
chord-seq (visible at the bottom left of the figure) provides a preliminary display of the
transcription of the bpf into a musical line. After this, the notes-change object of the
Profile library refines the pitches in this preliminary chord-seq by finding the closest
pitches from the chosen pitch field. Each sampled pitch is then given a 32nd -note duration
in a voice object to produce the final musical line.
The process above created a significant musical statement, referred to as the “aero
line” (Figure 7). From here, the musical texture is thickened through imitation of the
line in another voice with a certain delay. The two lines are then combined and displayed
in a poly object where they can be previewed and exported as MusicXML files into a
notation program. Once in the notation program, these lines were distributed between
orchestral wind instruments to make the sequence.
Figure 7. The “aero line”.
After establishing the aero line, I chose to elaborate it through a series of variations.
One straightforward way to do this was to make circular rotations of the list of points in
the original bpf line. To do this, I routed the values from the original line into a rotate
function, which was also given a random seed number from 0 to 100. The resulting
random circular permutation of the list, as before, was sampled against the given pitch
field using the notes-change function. An imitative line again was generated with a delay
for the start of the second voice. This process generated new lines that were, nonetheless,
still linked to the contours and proportions of the original aero line. Seven variations
were made through trial and error along with other variations created from a retrograde
of the aero line.
Assembling the model

As mentioned earlier, I combined the textures of the aero line in the winds with string
glissandi. These massed glissandi work their way through the harmonic flow of the seven
chord sequences, each one suggesting the intensity of acceleration, braking, and Doppler
effect captured in the short race car sound edits from which they were derived. The fast
aero lines in the winds trade off with the string glissandi in quick succession, emulating an
aerodynamic flow of air particles over the virtual car. Finally, the vibrant and forceful
racing engine noises are also represented throughout this section with tremolo figures
played on wood blocks and log drums: percussion instruments used to suggest the noise
of the car’s pistons. Figure 8 shows an excerpt from the corresponding part of the score.
126
Figure 8. Excerpt of As I ride the late night freeways, mm. 39-42.
127
Matthew Schumaker
Line morphing via bpf interpolation

Aerodynamism, envisioned as a smooth and flowing transformation of air, might also
be conceived as a musical process that “morphs” one musical line into another. In this
view, air molecules are imagined as molding to the shape of the car in a progressive
and continuous fashion as the vehicle rushes through them. Such a “streamlining”
process could also reflect Hadid’s and Schumacher’s notions of “soft forms”, where the
key elements are “interdependent” and “communicate with each other” [2]. With these
ideas in mind, I designed a process to transition smoothly from one distinctive melodic
fragment to another (see Figure 9).
Figure 9. Line morphing using bpf interpolation.
This morphing process takes a source melodic fragment and a target melodic fragment
as inputs, loaded into the patch as separate MIDI files. The getpitch sub-patch then
collects the pitches from each fragment into separate lists, converted into bpf objects that
become the inputs to the “morphing” process. The bpf-interpol function morphs between
the two lines, creating a series of interpolated curves between the source bpf and the
target bpf. Using this function, the composer has some control over the smooth transition
of the overall interpolation in two ways: first, by setting the number of interpolated curves
and, second, by using an interpolation exponential factor which distorts the resulting
interpolation process toward the starting or ending lines as desired.
Pitches are assigned to these lines in two different ways. The MIDI pitches in the
intervening curves resulting from the interpolation were in some cases used “as-is” with
a simple rounding to the nearest equal-tempered pitch. In other cases, where tighter
control was desired, a Lisp function (closest-pitch-match) was used to ensure that all
resulting pitches were derived from either the target or source lines.
The interpolation sequences are then notated musically in voice objects. While the
pitches of the interpolations are automatically derived from the process, the rhythms of
128
the interpolated lines are specified by the composer. In some cases, a uniform rhythm
in the resulting lines was desired in order to draw attention to the smooth unfolding
of the interpolation in the pitch domain. In these cases, one rhythm list was specified
and applied to all the curves. In other cases where the rhythmic profiles of the first line
and the second line differed significantly, a list of rhythm lists, one for each curve, was
supplied. The idea of smooth transitioning from one rhythmic profile to the next was a
leading concern in intuitively drafting these lists. As a final step, the individual voice
objects representing the musical sequences of the interpolation are joined together using
OpenMusic’s concat function. Figure 10 shows the result of an example interpolation
process.
Figure 10. Example sequence of line morphing using bpf interpolation.
Approaching a target line

A process used later in the work involves the gradual algorithmic reconstruction of the
aero line from a random cloud of notes. A shorter sequence of random-sounding pitches
gradually becomes longer and increasingly takes on the profile of an altered version of the
aero line, until a full statement of the line is heard. Here again, notions of “streamlining”
and of Hadid’s and Schumacher’s parametricism are suggested in the sense of “knitting
together” a texture, and in emphasising smoothness and continuity in the resulting form.
Similar to the previous process, the imaginary reference point here is a chaotic flow of
air molecules that is given a temporary shape as the car passes through them.
Figure 11 shows the corresponding patch. To begin, the line is loaded as a MIDI
file. In a couple of sub-patches, the rhythms and pitches of this line are extracted and
stored as separate lists. As in the previous process, new rhythms are input manually as
129
Matthew Schumaker
lists. These rhythm lists grow gradually from a short sequence of 5 notes to one of 21
notes. The last rhythm list is taken from the original MIDI file and corresponds to the
full length of the aero line.
Figure 11. Approaching target line patch.
Having defined the list of attack times for this growth process, what remain to be
determined are the corresponding pitches. A custom Lisp function, choose-set-pitch-or-
random-pitch, was made to carry this out. This function takes four lists as inputs: two
pitch lists and two lists controlling probabilities. Each of the four lists, in turn, has as
many elements as there are attacks in the rhythm lists. For each rhythm list, or “motif”,
one list of pitches is created that retains the pitch ordering from the original aero line.
A second pitch list is generated that contains a random sampling and ordering of pitches
taken from the whole of the line. The first of the probability lists is made from a master
bpf object that shapes the probability that a given attack will come from the first pitch
list, the one that preserves the recognisable ordering of pitches in the aero line. For every
attack in the rhythm lists, the bpf returns an increasingly high number up to 100. In
contrast, the second probability list always generates a random number between 0 and
100 for each attack in the master rhythm list. For each note in the sequence then the
Lisp function compares a value in the bpf probability list with the random probability
list. If the value of the random probability list is the higher number, then the pitch from
the corresponding random pitch permutation list is chosen; otherwise, the pitch from the
list preserving the aero line ordering is chosen. Since the values of the bpf probability
list become increasingly higher, as the process continues it becomes more and more likely
that pitches preserving the aero line ordering will be chosen. The perceptual result of
the finished line corresponds nicely to the desired goal—the aero line appears to emerge
almost imperceptibly from a chaotic mass of pitches.
130
This process was used in the piece primarily in an extended section written for divisi
strings. The resulting motifs are intuitively overlaid and imitated to create a web of
polyphonic sound divided amongst the different string voices. In addition, the form of
this section is elaborated and extended by concatenating sequences of different processes.
For instance, this process of creating the aero line from random pitches is quickly followed
by one of the morphing processes described earlier, now enabling the string lines to
transition smoothly from the newly formed aero line to a single, pulsing high pitch.
Figure 12 (a-b) presents corresponding sections of the score.
Future work: polyphony and repetition

In this work OpenMusic provided a set of tools that supported the parametric inspi-
ration of smooth, morphing transitions and of the drawn “aerodynamic” musical lines.
Over the course of composing, I realised that I wanted to develop further these processes
in order to generate polyphonic interpolations from many simultaneous source lines, each
leading to its own target. For this purpose, a polyphonic version of the morphing patch
was created that represents a significant expansion of the original one (Figure 13). This
version takes as inputs two polyphonic MIDI files, for the source and target line fragments,
and describes smooth transitions between each voice in the source and its corresponding
line in the target. Here the composer provides a list of rhythm lists for each of the
voices, and each sublist corresponds to a motif in a particular voice. The patch also
incorporates other features from my work with rhythm in OpenMusic. In particular,
the user determines how events unfold with two separate lists by controlling both voice
assignment and the time between onsets of the motifs. By explicitly addressing voice
assignment, the composer can sculpt sequences to create surprise through the choice of
the instrument playing next: the what of the musical sequence. Similarly, control of the
onset of each fragment across the voices enables a spectrum of approaches for dynamically
shaping the density of a texture and for potentially imbuing music with timing surprises:
the when of expected events. The patch was used for sketching several parts of this work
and holds promise for use in future compositions.
The focus on the time domain in the polyphonic version of this patch also signals
a more general area for further development: to control algorithmically the smooth
“morphing” not only of pitch sequences but also of rhythmic sequences. In the present
composition, when transformations were required that moved between two different line
fragments with quite different rhythmic profiles, the rhythmic transformation was done
intuitively. Certainly, interesting work has already been undertaken in this area in the
OpenMusic environment, for instance in [1].
Finally, one wonders about the essential role and importance of repetition in these
interpolative sequences. It appears that the overall effect of the interpolation hinges
on the repetitive sequence of motivic figures, each time incorporating slight changes
that increasingly transform the source into the target. Would it be possible to suggest
a similar morphing transformation that did not depend so much on repetition—a more
fluid and more gestural transformation? To this end, it might be interesting to investigate
transitions that were designed not from complete statements of the fragments, but rather
by creating a longer gesture comprising shorter cuts of each step of the interpolation
process from the source motif into the target motif. Perhaps work in this direction could
lead to interesting new means for the morphological unfolding of musical lines.
131
Matthew Schumaker
[...]
[...]
Figure 12a. As I ride the late night freeways, mm. 330-333, 356-358, and 362-365 (showing
only the process in the string parts).
132
Figure 12b. As I ride the late night freeways, mm. 376-385 (showing only the string parts).
133
Matthew Schumaker
Figure 13. Polyphonic interpolation patch.
References
[1] Eric Daubresse, Philippe Hurel. “Creating Polyphonic Sequences in the Form of
Rhythmic and Melodic Canons”. In Jean Bresson, Carlos Agon, Gérard Assayag (eds.)
The OM Composer’s Book 2. Editions Delatour France/IRCAM-Centre Pompidou,
2008.
[2] Patrik Schumacher. “The Parametricist Epoch: Let the Style Wars Begin”. AJ—The
Architects’ Journal, 231(16), 2010.
134
Materials and techniques in
D’improvviso da immobile
s’illumina for bass clarinet, two
orchestras, piano, and percussion
Federico Bonacossa
This paper describes how the computer was used to inform large-scale aspects as well
as to generate much of the pitch and rhythmic materials of D’improvviso da immobile
s’illumina (2013), a concerto for bass clarinet, two orchestras, piano, and percussion. In
particular it will discuss how OpenMusic was used to suggest possible transformations
of melodic ideas composed intuitively and to create a rhythmicon based on the frequency
information gathered from the analysis of a bass clarinet sample and the various com-
promises that had to be made along the way.
While composing D’improvviso da immobile s’illumina I attempted to reconcile two
seemingly opposing approaches to composition: a purely intuitive one, representing the
fascinating ability of our subconscious to naturally distill and combine elements from
the most diverse experiences, and a more “scientific” one, in which the material was
generated either by algorithms or some type of analysis with the help of a computer. The
two approaches influenced one another; the computer was used to generate variations of
the material created intuitively, and intuition guided a number of changes to the original
data collected.
Form
The form of the concerto is based on the shape of the waveform of the short sound
sample shown in Figure 1. The sample is a recording of a low B[ (58.2 Hz) played on
bass clarinet. In order to use the waveform as an outline for the form, I interpreted
the timeline of the sample so that each second would correspond to approximately one
minute of the piece. The envelope of the waveform provided an outline for dynamic
shape and density and consequently suggested a general plan for the use of instrumental
forces. I chose to divide the work into five main sections, corresponding more or less to
the various distinct stages of the envelope:
1. Introduction – bass clarinet solo.
2. Orchestral gradual crescendo.
3. Bass clarinet cadenza.
4. Short dialogue between bass clarinet, bassoons, and cellos.
5. Orchestral recapitulation.
135
Federico Bonacossa
Figure 1. Formal outline drawn from a bass clarinet sound sample.
In addition to defining the main sections, I determined a general harmonic plan by

choosing the central pitches or harmonic fields of each section. The pitches indicated on
top of the waveform in Figure 1 do not represent tonalities but primary pitches. In the
sections labelled “low B[ spectrum” the harmony is extremely dense (although B[ is the
“root”), while in the other sections the texture is rather sparse.
The sections indicated in the outline were clearly defined, but the transition from one
to the other is usually very gradual. This gradual transformation, or “interpolation”, of
sonorities provides an effective way of infusing a sense of forward motion into otherwise
static materials, enabling the music to transition seamlessly between different harmonic
fields and pitch centres. Interpolation in mathematics refers to the estimation of a value
or values between two known values. In this case the values represent pitches, but they
can represent other parameters as well. Similarly, interpolation in the concerto allows
the music gradually to transform the way that the waveform of the clarinet sample does:
several distinct components can be isolated in the waveform, but they all are part of a
single, coherent, sound.
Pitch material
The concerto’s pitch material can be reduced to two main pitch-collections. The first
consists of various set-theoretical manipulations of a basic hexachord. The second was
derived from the spectral analysis of the original sound sample from Figure 1. For the
sake of clarity, I will refer to the two pitch-collections as “Collection 1” and “Collection 2”.
The first step in the compositional process was to write the introduction for solo bass
clarinet. Rather than establishing a clear set of pitches or rhythms to use, I defined a
few general parameters:
Pitch centres: I chose D as the initial central pitch because it was one of the loudest
partials in the spectral analysis (shown in Figure 2) from which I planned to derive
Collection 2, and could act as a pivot point between the two collections.
Germinal motives: I wanted the introduction to include several distinctive motives

that could be further elaborated later in the piece (Figure 3). These include minor
seconds (A), minor thirds (B), trills (C), octave leaps (D), and repeated notes (E).
136
Materials and techniques in D’improvviso da immobile s’illumina
Figure 2. Spectral analysis of the bass clarinet sample used in the concerto (first 13 pitches).
Pitches are rounded to the nearest quarter-tone. Note the dynamic levels.
Figure 3. D’improvviso da immobile s’illumina: bass clarinet introduction (transposed score).
Collection 1 was derived a posteriori from the five pitches used in the introduction.
The pitches were arranged to form “Hexachord 1” shown in Figure 4 by adding an
additional pitch (C5) to fill in the gap between A and D.
Figure 4. Pitch Collection 1.
The prime form of hexachord 1 is the symmetrical set {0,1,3,6,8,9} (Forte number
6-Z29)[2].1 This set is inversionally symmetrical; its inversion corresponds to a transpo-
sition of the original set (T3 ). Hexachord 1 corresponds to T6 of the prime form. After
some experimentation with different transpositions of the set, I decided to combine two
transpositions, the original one (T6 ) and T9 , to generate a larger eight-note set that
forms Collection 1 (last measure in Figure 4).
1 Notethat in the Forte catalogue 6-Z29 is sometimes listed as {0,2,3,6,7,9}, but {0,1,3,6,8,9} is equivalent
and actually more compact. See for instance http://solomonsmusic.net/pcsets.htm.
137
Federico Bonacossa
The MathTools package in OpenMusic includes several useful objects for working
with sets. I created several simple additional tools (such as the abstraction midic-›set)
that I used in combination with the library to accomplish specific tasks. Figure 5
shows a patch that can quickly calculate prime forms and inversions. Figure 6 shows
a patch used to combine sets to form larger ones (in this case Collection 1). In order
to generate additional material I also used OpenMusic to calculate and notate subsets
and transpositions of the initial hexachord. This then served as a pool of material from
which I could draw when necessary (Figure 7).
Figure 5. Using the objects from the MathTools package to calculate the prime form and
inversion of a set. The n-cercle objects provide a graphic representation of the sets.
Figure 6. Transpositions of the hexachord and set combinations (generation of Collection 1).
138
Figure 7. Patch used to calculate and notate transpositions of sets.
The patch shown in Figure 7 allows the user to enter a set and calculate all chromatic
transpositions. This is a simple task, but the advantage here is the ability to notate the
results in a usable layout. The abstraction transp-chrom outputs directly to a voice object
and subdivides each transposition into measures of the appropriate length, which makes
it much easier to read (the user can also enter a beat value). This can then be exported
as a MusicXML file into notation software (or be printed directly in OpenMusic).
Figure 8 shows the same approach applied to subsets. In the patch the user can
specify the length of the subsets to notate. The function sub-power gives a list of all
the subsets of the indicated length. The abstraction notate-divid-chrom calculates all
chromatic transpositions of each individual set and notates them in the specified metre.
There are some slight but important differences between transp-chrom and notate-
div-chrom. In transp-chrom the internal loop simply adds each value generated from the
arithm-ser to the list of pitches. In notate-div-chrom each individual element of the list
is transposed chromatically before moving on to the next.
139
Federico Bonacossa
Figure 8. Patch used to calculate chromatic transpositions of subsets.
The octachord Collection 1 has several features that I find compelling and that, not
coincidentally, characterise the pitch material of several of my earlier compositions, for
example Rivir for solo viola, shown in Figure 9.
Figure 9. Excerpt from Rivir for solo viola.
The collection begins with a minor second, typical of the

Phrygian and Locrian modes. The first and second notes,
D and E[, also have a corresponding pitch a perfect fifth
above them. The two pairs of fifths (Figure 10), especially
when combined harmonically, provide an attractive blend of Figure 10. Structural in-
tervals in Collection 1.
stability and tension and form the skeleton of the collection.
140
Figure 11 shows two recurring motives in the concerto

both derived from Collection 1. Collection 1 contains two instances of motive 2 a major
third apart. This mirror tetrachord is a subset of the octatonic scale and is found in
a number of works. It can be heard for example in John Chowning’s electronic work
Sabelithe (1971) and in Sofia Gubaidulina’s Concerto for Viola and Orchestra (1996).
Figure 11. Some recurring sets derived from Collection 1 (see Figure 5).
In my recent compositions, I have often experimented with the expansion and con-
traction of sets. For this purpose I created two simple patches that enable me to try
out different transformations. The two patches are based on two different approaches:
expansion or contraction by scaling, and expansion or contraction by addition or subtrac-
tion of a constant value. The patch shown in Figure 12 is based on the first approach. It
allows one to scale the intervals by a factor and then see and hear the result. The scaling
factor can be a non-integer (Figure 12 uses a factor 0.5), in which case the results might
need to be rounded (om-round). The resulting set has the same contour as the original
one, starting at a specified pitch (in this case D4).
Figure 12. OpenMusic patch for expanding and contracting sets by interval scaling. A: input
a sequence of notes, B: scaling factor, C: starting pitch.
141
Federico Bonacossa
The patch shown in Figure 13 is based on the second approach. It calculates intervallic
expansions by adding or subtracting a specified number of half steps to each interval. In
contrast Figure 14 shows a short passage from the concerto consisting of an expansion
of the first tetrachord from Figure 11 (1) calculated using the interval scaling approach.
Figure 13. Half-step-expander. A: starting set, B: number of half-steps to add to each interval,
C: starting pitch of the resulting set.
Figure 14. Motivic expansion (intervals scaled by a factor of 3). D’improvviso da immobile
s’illumina, excerpt from m. 143, Bass clarinet.
The second important pitch collection (Collection 2) is based on a spectral analysis

of the clarinet sample from Figure 1. Because of the large amount of data contained in
a sound of this length (16 seconds) I extracted a small sample, slightly longer than 200
milliseconds, from the loudest point of the waveform. I then used Spear to analyse the
sound and gather information on the frequency components.
For the purpose of the concerto, I envisioned a large chord played by the whole
orchestra. I did not want to use the data collected from the spectral analysis melodically
as I felt that it would have a greater effect if all pitches were to be heard at the same
time. I used OpenMusic to read the SDIF file produced by Spear and convert the
information into a usable format (patch shown in Figure 15). The function as-›om lets
the user filter the data contained in the SDIF file by defining various parameters, such
as the minimum partial length, or the number and range of partials to extract.
142
Figure 15. Left: Spectral analysis of the sound from Figure 1 in Spear. Right: OpenMusic
patch used to filter spectral data.
Although this process can be done “manually” by reading the information directly
from the SDIF file, OpenMusic provides a flexible way to experiment with different
settings, and, most important, hear the results immediately. In other words, it allows
the composer to make decisions based on the actual sound.
In the patch shown, I specified the pitch range according to the instruments I planned
to use in the concerto (A). As most of the frequencies in the analysis are non-tempered
pitches, I rounded the data to the nearest quarter-tone (B) and then eliminated any
duplicates that might result (C). The resulting 36-note series is shown in Figure 16 (the
first 13 pitches were already shown in Figure 2).
Figure 16. Pitch collection derived from spectral data (Collection 2).
143
Federico Bonacossa
Frequency to duration (rhythmicon)

My plan was to use the data collected from the spectral analysis to create a rhythmicon
where each note of the chord would be assigned a duration that was derived from its
frequency. Each pitch would then be repeated over and over creating a dense texture
and complex polyrhythms. This concept was inspired by Henry Cowell’s ideas described
in his book New Musical Resources (1958) [1].
Cowell lays out in detail his approach to rhythm based on principles “already familiar
in the field of harmony and counterpoint—that is, in the relation of tones to one another”
(p. 46). He writes:
A parallel can be drawn between the ratio of rhythmical beats and the ratio of
musical tones by virtue of the common mathematical basis of both musical time
and musical tone. The two times, in this view, might be said to be “in harmony”,
the simplest possible. [1] (pp. 50–51)
Cowell proposes the application of the “principles of relating time to musical tone”
not only to durations, but also to metre, dynamics, form, and tempo. In the table shown
in Figure 17 he charts the rhythmic relationships inherent to a major triad that can be
derived from the mathematical relationship between the fundamental and all the other
tones. The first column indicates the tone’s location in the harmonic series. Columns two
and three indicate the interval of the tone in relation to the fundamental and the name
of the tone assuming C as the starting pitch. The last column shows the mathematical
ratios of the vibrations using 16 Hz as the fundamental frequency.
Figure 17. Rhythmic relationships based on the overtone series, from [1], p. 47.
Cowell stresses the similarities between “consonant” rhythmic relationships and con-
sonant intervals, according to how many repetitions of a tone were needed before it
would realign with the fundamental: “the smaller the number of units that must be
passed over before that coincidence is re-established, the more consonant is the interval.”
He also points out that the “the higher one goes in the overtone series, the greater the
number of units that must be passed over, and the further is the effect from one of simple
consonance.” The reason why the tones result in harmony instead of a “chaos of sounds”,
he writes, “is that at regular intervals the vibrations coincide”.
Cowell utilises the ratios of the intervals in the harmonic series to create a “rhythmic
harmony” so that, for example, a melody with a pulse of three half notes in the space of
two would correspond to the interval of a pure perfect fifth, five quarters in the space of
four to a major third and so on. He then gives various examples of their use, many of
144
them quite complex, admitting that most musicians would probably find them difficult
to perform.
Cowell also makes the important point that complex rhythmic relationships can be
easily performed by a player piano and even suggests the creation of an instrument that
would assign specific rhythms to each pitch (p. 64). His dream became reality through
support by Charles Ives, who commissioned the Russian inventor Leon Theremin to build
a rhythmicon, a machine that would convert pitch into rhythm and vice-versa. Theremin
built two models in 1932, effectively creating the first rhythm machine, but neither was
ever reliable enough to be used in performance [6]. In 1960 Theremin built another
rhythmicon while at the Moscow Conservatory or possibly the USSR Sound Recording
Institute. The instrument could perform complex polyrhythms such as 7 against 9 or 5
against 13. The model to this day is still in working condition [7].2
In New Musical Resources, Cowell always utilises ratios based on the overtone series
even though he realises that they are in most cases significantly different from the ratios
found in equal temperament. The reason for his choice is easy to understand if we assign
durations to two pitches a major third apart using just intonation and equal temperament
and compare the results. For instance, taking A4 as the fundamental, if we calculate the
duration of C]5 based on its frequency in the tempered scale (≈554.3653 Hz) and relate
it to the fundamental (440 Hz), we end up with an irrational ratio. (Even rounding the
top frequency to 554 Hz this ratio is still 277/220.) If, however, we use a justly tuned
C]5 (550 Hz),3 we obtain the much simpler ratio 5/4. In terms of durations, this means
that for every four repetitions of the fundamental there will be five repetitions of the
third (see Figure 18).
Figure 18. Durations derived from the ratio of the frequencies of an A4 and a justly tuned
C]5. The duration of the C] is essentially a quintuplet spanning four bars.
James Tenney uses a similar process in Spectral CANON for CONLON Nancarrow in
which, not coincidentally, he tunes the player piano to the first 24 partials of the harmonic
series [3].4 In this piece Tenney calculated durations for the piece by defining an arbitrary
initial duration k and calculating successive durations with ratios from the overtone series,
starting with the ninth partial (9:8 ratio). The details of the algorithm used by Tenney
are discussed at length for instance by De Paiva et al. in [4]. An implementation of the
algorithm is shown in Figure 19.
2 For a video demonstration of the Rhythmicon see https://youtu.be/HkodVcuPVAo.

3 The fifth partial of A4 is C]7 which has a frequency of 2200 Hz. In this case we need to divide it by
four to lower it two octaves to arrive at C]5, hence the ratio 5/4.
4 Thereare many other examples, especially by North American composers (see [5]), and even an “online
rhythmicon” created by Nick Didkovsky.
145
Federico Bonacossa
Figure 19. Algorithm used to calculate durations in James Tenney’s Spectral CANON for
CONLON Nancarrow, adapted from [4].
The approach I used to calculate durations in

the concerto differs from both Cowell’s and Tenney’s
in that the frequencies are assigned an absolute
duration based on their period, and not on their ratio
to a fundamental.5 Each period was then multiplied
by a factor of 1000 to bring it into a playable
range. In addition, I calculated durations after the
frequencies were rounded, therefore they are based
on the frequency actually being performed. This
method naturally created some complications when
the durations were notated because of the complex
rhythms they often generated.
Figure 20 shows a simple patch called mc-›ms
used to calculate the period corresponding to each
midicent using the formula t = 1000/f . Mc-›ms
allows the user to specify the maximum subdivision
in milliseconds. A value of 50 indicates that the
Figure 20. Abstraction mc-›ms
maximum subdivision will be one twentieth of a beat
used to assign durations to the
(corresponding to one second at tempo 60). I chose frequencies in Collection 2.
this maximum subdivision because I wanted accurate
5 Aspreviously mentioned Tenney chose an arbitrary initial duration, although as de Paiva observes, 4
seconds seems to be an ideal length in terms of how it allows the relationships to be perceived [4].
146
results, although I knew that this would present some practical problems once the values
were converted to standard notation. The main patch, shown in Figure 21, also calculates
how many repetitions of the note would be necessary to fill 5 minutes (B). Figure 22 shows
a list of all midicent values with their calculated durations in milliseconds. Each result
was then stored as a voice (see the set of “instance” boxes at the bottom of the figure).
These voices were then consolidated into a single score in a poly object (Figure 23) and
exported to Finale.
Figure 21. Patch to assign a duration to a midicent value based on its frequency and determine
the number of repetitions needed.
Figure 22. List of pitches and durations used in the rhythmicon sections. Each column
represents the chord tone number followed by, in parenthesis, its pitch in midicents and duration
in milliseconds.
147
Federico Bonacossa
Figure 23. Individual voices combined into a score (partial view).
While Cowell advocated notating each part of a polyrhythm in its own metre to
eliminate the need for performers to “negate” the natural accent of the metre ([1], p. 70),
doing so in the concerto would have resulted in more than 20 different simultaneous
metres. I instead chose to notate all parts in the same metre so that the conductor
would at least be able to provide regular reference points for the performers.
The resulting score (Figure 24) was used as a reference to establish durations in
the actual piece. Because standard music notation is ill-suited for this sort of process,
and because the relatively high precision level I chose led to results that were often
unnecessarily complicated and impractical, I often simplified the notation to make it
easier to read and perform. Indeed, D’improvviso da immobile s’illumina was written for
a student orchestra so it was important for me to keep the notation relatively simple.
As an example let’s take the pitch A]4. Its frequency (466.1638 Hz) corresponds to
a period of 1000 / 466.1638 = 2.1451688 ms, yielding 2150 ms when multiplied by 1000
and rounded to the nearest multiple of 50 (1/20 of the beat). If we input this value into
the voice object we obtain the result shown in Figure 25 (A). The resulting notation
is not consistent and is also unnecessarily difficult to read. This problem could have
been avoided by rounding to a different value, for example 250, but I decided to use the
148
Figure 24. First measure of the raw score used to create the rhythmicon in D’improvviso da
immobile s’illumina.
149
Federico Bonacossa
same beat division for all pitches and make modifications as necessary if the results were
impractical. This particular pitch (A]4) was assigned to clarinet 1. I decided to choose
a simpler rhythmic notation that was fairly close in length to the original duration. The
duration I chose (2333 ms instead of 2150 ms) corresponds to 7/12 of a measure, two
beats and a third, which allowed me to make the notation easier to read while remaining
relatively close to the original length, as visible in Figure 25 (A)—the original—and
(B)—the new duration.
Figure 25. Rhythmic approximations.
In my view this process is similar to rounding the pitches from the spectral analysis.
Naturally if we were to align both sets of durations the difference would become quite
noticeable within a few measures, but the result is still very effective. In addition to
simplifying the notation, I also periodically interrupted the pattern to ensure that the
performer could breathe regularly (but always for an full duration unit).
Additional operations
I ultimately decided to reduce the size of the chord to 24 pitches which allowed me to
split or reinforce voices when needed. I diverged further from the original results by
offsetting some of the entrances to avoid having the voices align too soon, especially as
the rhythmicon unfolded. Lastly, to make the parts more interesting I placed regularly
recurring accents in each voice; as a result, most voices have a macro as well as a micro
pulse (Figures 26 and 27).
> >
&Œ Œ œJ ˙ ˙ œœ œ œ ‰Œ Ó Ó Œ Œ œJ
J
f
3 3 3 3
Figure 26. D’improvviso da immobile s’illumina mm. 91–94, Clarinet 1.
b œ >œ œ œ œ b œ œ œ >œ œ b œ œ œ œ œ b >œ œ œ œ œ b œ œ >œ œ œ b œ œ œ œ >œ b œ œ œ œ œ

&
5 5 5 5 5 5 5
Figure 27. D’improvviso da immobile s’illumina mm. 101–107, Violin 2.
Once I collected the raw material for the rhythmicon sections, I separated the chord
into several smaller sonorities and plotted the entrances carefully, leaving the entrance
of the fundamental tone last (Figure 28).
150
Figure 28. Pitches used in the rhythmicon section in order.
I then segmented the score into smaller sections and inserted between them material
based on Collection 1 so that the rhythmicon would only gradually become prominent.
Measures 59–70 of the concerto (see mm. 59–62 in Figure 29) are an example of how
I delayed the crescendo of the rhythmicon in this way. In m. 71 the rhythmicon is
interrupted by the percussion, which recalls mm. 4–7 of the concerto. The rhythmicon
becomes prominent at m. 91, continuing on for nearly 50 measures, its texture steadily
thickening until the fundamental tone of the chord is finally heard in the tubas in m. 117
(Figure 30), signaling the arrival of the climactic point. The low B[1 is played by
both tubas and is also doubled by double bass 2, while being reinforced one octave
higher by double bass 1 and two octaves higher by bassoon 2 and horn 1. By suddenly
becoming the loudest pitch, the fundamental tone provides a sense of meaning to the
otherwise chaotic texture, revealing a large-scale hierarchy that was hitherto disguised.
The instruments occasionally align with one another, but are otherwise asynchronous.
But when the rhythmicon comes back after the long cadenza in m. 176 all voices are
finally aligned, quickly dissipating again into an asynchronous texture. The return of the
chord structurally defines the beginning of the final section.
Conclusion
This article has demonstrated how OpenMusic may be used as a practical tool for
composers to try out different ideas and generate material while still leaving plenty of
room for intuitive choices. Often compositions have a way of taking on a life of their
own and naturally deviate in various ways from the original plan. This unpredictability
is one of the most fascinating and important aspects of a composer’s work.
151
Federico Bonacossa
Figure 29. D’improvviso da immobile s’illumina mm. 57–62.
152
Figure 30. D’improvviso da immobile s’illumina mm. 114–118.
153
Federico Bonacossa
References
[1] Henry Cowell. New Musical Resources. Cambridge: Cambridge University Press,
1958.
[2] Allen Forte. The Structure of Atonal Music. New Haven: Yale University Press, 1973.
[3] Larry Polansky. “The Early Works of James Tenney”. In Peter Garland (ed.)
Soundings 13: The Music of James Tenney, chapter: “XI: Spectral CANON for
CONLON Nancarrow”. Santa Fe, NM: Soundings Press, 1983.
[4] Charles De Paiva Santana, Jean Bresson, Moreno Andreatta. “Modeling and
Simulation: The Spectral CANON for CONLON Nancarrow by James Tenney”. In
Proceedings of the Sound and Music Computing Conference. Stockholm, 2013.
[5] Robert Wannamaker. “Rhythmicon Relationships, Farey Sequences, and James
Tenney’s Spectral CANON for CONLON Nancarrow (1974)”. Music Theory
Spectrum, 34(1), 2012.
Online
[6] Andrew Baron, Mike Buffington. “1932 Rhythmicon”.

http://rcatheremin.com/32rhythmicon.php.
[7] Andrew Baron, Mike Buffington. “1960s Rhythmicon”.
http://rcatheremin.com/60rhythmicon.php.
154
Musique instrumentale concrète:
Timbral transcription in What the
Blind See and Without Words
Aaron Einbond
Transcription is an increasingly influential compositional model in the 21st century.

Bridging techniques of musique concrète and musique concrète instrumentale, my work
since 2007 has focused on using timbral descriptors to transcribe audio recordings for
live instrumental ensemble and electronics. The sources and results vary, including
transformation of noise-rich playing techniques, transcription of improvised material
produced by performer-collaborators, and fusion of instrumental textures with ambient
field recordings. However the technical implementation employs a shared toolkit: sample
databases are recorded, analysed, and organised into an audio mosaic with the CataRT
package for corpus-based concatenative synthesis. Then OpenMusic is used to produce
a corresponding instrumental transcription to be incorporated into the finished score.
This chapter presents the approach in two works for ensemble and electronics, What the
Blind See (2009) and Without Words (2012), as well as complementary real-time tech-
nologies including close miking and live audio mosaicking. In the process transcription
is considered as a renewed expressive resource for the extended lexicon of electronically
augmented instrumental sound.
Reproduction
Mimesis in music may date to the beginning of music itself, and its reach extends both
historically and geographically, from Plato’s Republic to Tuvan throat singing ([13],
p. 58). It is also central to common-practice music, whether the imitation of imitative
counterpoint, the programme of programme music, or the affect of the Baroque doctrine
of the affections. In the 20th century mimesis has been reframed by “technological
reproducibility”, as in Walter Benjamin’s landmark essay. In technology’s wake, Pierre
Schaeffer treats reproduction by juxtaposing “abstract” and “concrete” listening modes;
it is with their synthesis that he defines “reduced listening” ([2], p. 37). Composers
who combine instrumental and electronic technologies of reproduction include François-
Bernard Mâche and his “phonography”, Clarence Barlow’s “synthrumentation”, Gérard
Grisey’s “instrumental synthesis”, Trevor Wishart’s sonic “transformation”, and Peter
Ablinger’s “phonorealism” [15, 5]. Surveying these perspectives as well as those of the
early 21st century, Nicolas Donin writes: “‘composition’ [...] now includes as well the
navigation between different ‘reproducibilities’ through operations of translation such as
155
Aaron Einbond
transcription, transcoding, or transformations” [5]. Transcription has been a particularly

influential trend in what Donin terms “instrumental resynthesis”.
In the purely instrumental domain, Helmut Lachenmann offers a historical counter-
point with his musique concrète instrumentale:
The idea of “instrumental musique concrète”—i.e. sound as a message conveyed
from its own mechanical origin, and so sound as experience of energy [...] signifies
an extensive defamiliarisation of instrumental technique. [16]
Or as Rainer Nonnenmann clarifies, “instead of relating sounds to extra-musical

causes, listeners were now called upon to relate sounds back to the genuinely intra-
musical preconditions of their concrete instrumental production” [14]. Now, over 40
years after Lachenmann’s initial explorations of musique concrète instrumentale, one
could argue that the sound world of Lachenmann’s instrumentarium has itself become
familiar, and is no longer available to be “freshly illuminated” [16]. Yet judging by
its ongoing popularity to a younger generation of composers, it is far from in danger
of being abandoned. How can the composer of instrumental music employ these now
familiar techniques and still hope for them to speak vividly?
The solution is to turn musique concrète instrumentale on its head: instead of focusing
the listener on the mechanical origins of the sounds, to free the instrumental sources
to suggest timbral details of other concrete origins. Instrumental sounds, stripped of
their historical connotation and reduced to their timbral essentials by Lachenmann
and the generation that followed, are now available for renewed connotation. Rather
than musique concrète of instrumental sound, it is instrumental music that is perceived
concretely: musique instrumentale concrète.
Behind the playful turn of phrase lies a wealth of evocative potentials, and associated
technical questions, that have occupied my compositional work since Beside Oneself for
viola and electronics in 2007. How can sonic expression be renewed without falling
back on historically laden expressivity?1 Technology has proven an indispensible means
toward this goal, where concrete reduced listening is effected through processes including
close miking, amplification, sampling, and audio mosaicking. This chapter will focus in
particular in how OpenMusic is used in these processes for a range of compositional
applications.
Transcription
A potential definition of musical transcription is the projection of material from one
musical space to another—a space of frequencies or pitches, timings or rhythms, or other
acoustic parameters. Especially when working with noise-based instrumental sounds, a
suggestive metaphor is timbre space, a model of timbre as a multidimensional percept
[21, 8]. Such timbral dimensions have been associated with audio features, or descriptors:
“characteristics extracted from the source sounds, or higher level descriptors attributed to
them” [17]. The notion of the descriptor is the perfect technological correlate of reduced
listening: like Schaeffer’s phenomenological époché [2], the computer takes no account
1 Thiscontrast is neatly denoted by two distinct German translations of “expression”: Ausdruck/

Expression.
156
Musique instrumentale concrète...
of the sound’s physical origins as it performs its analysis. Of course this does not apply
to the human listener, who must set the parameters of the algorithm beforehand and
interpret the results after, in both cases based on his or her own listening experience.
So rather than transparent measures of perception, timbral descriptors are contextually
sensitive materials upon which the composer necessarily exercises a subjective influence.
Decisions include what samples to use, how to segment them, what descriptors to analyse,
among many others. Far from automatic, timbral transcription can be a rich expressive
resource.
In What the Blind See for viola, bass clarinet, harp, piano, percussion, and electronics
(2009), sources for transcription include instrumental materials, physical gestures per-
formed with an electronic controller, and field recordings of rain, ice, and snow. The
process is termed Corpus-Based Transcription [7] as it takes advantage of corpus-based
concatenative synthesis (CBCS) with the CataRT package for Max, and transcription
with OpenMusic. While the details have been described elsewhere [7, 5], an example
can be outlined as follows: a large database of instrumental samples (the corpus, in this
case made up of samples of the five performers of the ensemble) is compared to a field
recording (the target, in this case an arctic glacier) using CataRT and a purpose-built
Max patch for analysis. After segmenting both corpus and target into short grains, and
matching those segments with the most similar descriptor values, the glacier recording
is reproduced with an audio mosaic of instrumental samples.
This mosaic is then used as a model for an instrumental score. The timings, filenames,
and associated data are stored in an SDIF file, which can be loaded by the OpenMusic
SDIFfile object. The data are extracted, organised, and rhythmically quantified in
OpenMusic (Figure 1), then the contents of the resulting poly object are exported
as a MusicXML file. This is imported into Finale and edited, with the aid of the
descriptors and metadata stored in the SDIF file, by annotating playing techniques,
dynamics, and articulations, as well as adjusting or simplifying rhythms and techniques
to facilitate playability. The resulting transcription is incorporated into the work as the
compositional process continues (Figure 2), and finally the ensemble performs the score
live, reinterpreting the transcribed field recording in concert.
In the finished work, the original field recording of a glacier is not performed, but
its shadow remains through the fusion of the ensemble into a colourful imprint. Even
without an explicit concrete referent, the individual timbral details and identities of each
instrument are subsumed into an “instrumental ambient”2 soundscape. At the same time
an association with other field recordings of rain and snow that are directly revealed later
in the work may retrospectively conjure a secondary denotation of the texture’s origin.
In other examples transcription sources are made more or less explicit, opening up a
powerful expressive resource with a decisive effect on perception.
One of the keys to this multi-valence is the simulation of reduced listening through
close miking, sampling, and amplification. In What the Blind See, contact microphones
are placed on the piano frame, piano soundboard, bass drum skin, and tam-tam, and
miniature lavalier microphones are used on the bridge of the viola, near the embouchure
of the bass clarinet, and beneath the soundboard of the harp. These microphones permit
the amplification of sounds distorted out-of-scale to their mode of production, flattening
2 Bernhard Günther, personal communication, 25 July 2010.
157
Aaron Einbond
their concrete instrumental sources to the surface of a loudspeaker. So unfamiliar

playing techniques are “misheard”—from the perspective of Lachenmann—as they cannot
be identified with their productive energies. But on the contrary, through Schaeffer’s
reduced listening, they are freed to represent other sonic experiences that may be coaxed
out through transcription.
Figure 1. OpenMusic patch to transcribe an audio mosaic of a field recording.
Figure 2. What the Blind See mm. 176-79.

c Edition Gravis Verlag, Brühl, Germany. Printed
with kind permission.
158
The same microphones are used for recording the pre-compositional sample databases
as well as amplifying the instruments in concert, assuring the closest possible reproduc-
tion of the audio mosaic live. More important, the same musicians are called upon to
record the database as will premiere the work, in this case Ensemble L’Instant Donné,
and the imprint of their personalities and instruments becomes a collaborative part of
the compositional process. This process is not only “radically idiomatic”3 but radically
personalised. Like the audio descriptors themselves, the performers and instruments too
become sites of individual expressive variation, privileging the specific, concrete, and
personal over general, abstract, or universal.
While using OpenMusic to transcribe samples of a performer is not new,4 a signif-
icant difference here is the representation of noise-rich audio in its full detail without
recourse to the time-frequency grid of a spectrogram. To reproduce this sonic timbral
palette in symbolic music notation is a significant challenge. During sampling sessions,
performers’ playing techniques are documented with verbal explanations, photographs,
and video. When the sessions are edited into sample databases, the filename of each
sample is annotated by its playing technique in as much detail as possible, which is then
used to edit the score in Finale. Nonetheless the noise-rich timbres push the limits of
music notation: for many works the score is supplemented by sending the performers
a selection of their own audio samples. The goal is to trigger their sonic and physical
memories, so the symbolic notation serves as a mnemonic for personal oral history going
back to the sampling session.
Notably, for many playing techniques, pitch may not be a salient feature of either
the sound or its notational representation. Regardless, CataRT estimates the pitch
of each grain using the yin∼ object in Max. However the yin∼ algorithm may not
give a reliable estimate, or independently, the user may decide not to weight pitch
in constructing the audio mosaic. When the mosaic is imported to OpenMusic, the
estimated pitch is used in the poly display merely as a placeholder for each grain. As
the score is edited into its final form this pitch information may be retained, adjusted,
or eliminated entirely according to the playing technique needed. Comparing Figures
1 and 2, the viola pitches are replaced by x-shaped note heads (pizzicato on dampened
strings) and harp pitches by rectangular note heads (fingernails sur la table), while those
of the breathy low bass clarinet and marimba with Superball mallets are retained with
adjustments and elaborations.
At the heart of the OpenMusic patch omquantify quantises the durations of grains
in the audio mosaic, expressed in milliseconds, into notated rhythms (in Figure 1,
omquantify is included inside the make-voice sub-patch). Significantly, this use of detailed
rhythmic notation does not have a metrical function in itself, despite its “complex”
appearance. As such it could be analogised to an elaborated form of graphic notation
where the metrical grid serves a reference purpose only. Precedents include scores of
György Ligeti, Gérard Grisey, and Tristan Murail.5 But while the notated metre is not
intended to be heard directly, the quantisation parameters are chosen finely enough to
3 RichardBarrett and Daryl Buckley, Dark Matter programme text (2003):

http://richardbarrettmusic.com/DARKMATTERinterview.html
4 See for instance [20].
5 For example: György Ligeti, Études pour piano, premier (Schott, 1985), p. 20, p. 28.
159
Aaron Einbond
facilitate the performers’ interpretation of the “expressive microtiming” [9] underlying

the transcription target: whether the “feel” of a recorded improvisation, or the “groove”
of the interconnected actors (animal, vegetal, mineral) in a recorded soundscape.
Gesture
A physical gesture rather than a field recording may be used as a target: in another
example from What the Blind See, a Wacom tablet was used to “improvise” a short ges-
ture mapped directly to the Max lcd object that CataRT employs as a two-dimensional
representation of the sample corpus. As in the previous example, the exported SDIF
file—including sample timings, textual metadata, and descriptor values—was imported
and transcribed in OpenMusic (Figure 3) and a MusicXML file was exported to Finale
for further editing (Figure 4). The gesture was drawn to connect the approximate spectral
region of a preceding passage, based on an audio mosaic of rain, to other material that fol-
lows, with the goal of a smooth timbral transition. The interlocking timbral associations
that result—including nails and fingers on the harp sur la table, plectrum glissando along
the low piano strings, and wire brushes on the vibraphone—would have been difficult to
conceive with pencil and paper alone. Indeed the smooth trajectory in timbral parameters
does not necessarily match a symbolic pitch trajectory: although short segments comprise
rising scales or arpeggios, they do not connect the instrumental registers of the ensemble
in the most predictable way. Hearkening back to Lachenmann, the re-contextualisation
of extended instrumental techniques in a timbre-driven composite gesture stimulates the
composer’s—and listener’s—imagination unconventionally.
Figure 3. OpenMusic poly transcribed from a gesture improvised with a Wacom tablet.
160
Figure 4. What the Blind See mm. 157-159.

c Edition Gravis Verlag, Brühl, Germany.
Printed with kind permission.
Transformation
A third example is taken from the opening of What the Blind See, where transcription
is applied to instrumental sample targets themselves instead of “extra-musical” sources.
By taking a short instrumental figure from one instrument and transcribing it in varied
form for another, the technique extends a classical form of musical imitation: motivic
transformation. In this case, the audio recording of one instrument, the viola, is taken as
the target for an audio mosaic using the samples of another, either the harp or vibraphone
(Figure 5). This is closer to Lachenmann’s musique concrète instrumentale than musique
instrumentale concrète, as there is no “extra-musical” referent. However by transforming
an acoustic target, rather than symbolic music notation, associations between materials
including rhythms and playing techniques may suggest themselves beyond the composers’
notational habits, conditioned and distorted by personal and historical practice.
Even more than in the previous examples, this one introduces details of the perform-
ers’ collaborative interpretations on several levels. The viola part is derived from the
earlier work Beside Onself, performed and recorded from the notated score by the violist
of the ensemble. So the performer’s interpretation of the written material, complete with
shades of timbral variation and expressive microtiming, becomes the target of the audio
161
Aaron Einbond
mosaic of harp or vibraphone samples. The transcription then takes into account these
variations, as well as the variations of attack time and responsiveness of the viola playing
techniques. The resulting transcriptions differ in significant detail from the original
notated viola part, but preserve its rhythmic and timbral flavour. This transcription takes
advantage of a further layer of rhythmic choice: the OpenMusic patch is used to filter the
audio mosaic output by CataRT according to a range of successive loudness thresholds,
producing transcriptions of different rhythmic densities. The final harp and vibraphone
parts in Figure 5 are drawn freely from these different rhythmic layers, permitting greater
control over the result.
Figure 5. What the Blind See mm. 11-21, showing the viola part transcribed for harp and
vibraphone.
c Edition Gravis Verlag, Brühl, Germany. Printed with kind permission.
Of course all three parts, the viola target and harp and vibraphone mosaics, are
subject to similar variations in timbre and microtiming when reinterpreted in live per-
formance. Performed simultaneously in the finished work, the result is a homophonic
doubling of the motivic material that fuses in terms of timbre and density. The live
electronics (indicated by the pedal marking for cue 3 in Figure 5) respond through a
similar process, complementing the viola with a real-time audio mosaic drawn from the
harp and vibraphone sample databases. In effect this is another version of the notated
harp and vibraphone parts, re-synthesised in parallel with the performance by the live
players, complementing with a successive layer of heterophony.
In a related passage later in the work, a dense ensemble texture is transcribed from
solo instrumental material. In this case, rather than responding to a notated score, the
bass clarinetist improvised on a repeated low B[ tongued as fast as possible: the natural
162
Figure 6. Arnold Schoenberg, Fünf Orchesterstücke Op. 16, No. 3. (reduction with C[ pedal
tone omitted for clarity).
variations in speed and response of the instrument produced a subtly shifting rhythm.
A mosaic was then transcribed for the rest of the ensemble using sample banks combined
from all four instruments excluding bass clarinet. In the finished score the bass clarinet
accompanies with a similar figure, ideally producing a timbral fusion between the soloist
and the ensemble texture.
Instead of the multiple-stranded heterophony of the opening viola solo, this passage
unfolds a virtual timbral polyphony. The mosaic made with CataRT is conceptually
monophonic, matching one sample at a time of the corpus to the target. However, when
the notes of this line are distributed to the instruments of the ensemble each note is
sustained or let ring longer than its monophonic value. The effect could be likened to the
verticalisation of a melodic line into a sustained harmony, for example in Schoenberg’s
Fünf Orchesterstücke Op. 16, No. 3 (Figure 6). The result suggests a “timbral pedal”
accompanying the soloist (Figure 7).
The rhythmic details of the bass clarinet part were derived from a different process.
An audio recording of an improvisation with CataRT, using the mouse as controller, was
segmented in OpenMusic using the transient-detection object from the OM-SuperVP
library (Figure 8). By applying incremental transient detection threshold values, rhyth-
mic patterns of decreasing density are produced. By using OM-SuperVP instead of
AudioSculpt6 the parameters can be flexibly adjusted directly within the OpenMusic
patch, and successive analyses can be applied iteratively. The use of a list of threshold
values to produce a rhythmic matrix resembles the varying loudness thresholds used to
compose Figure 5, however now the results are used formally: each level of the matrix is
composed out successively, to produce a carefully-controlled decrease in rhythmic density
across a long-term section of the work. The bass clarinet rhythm in Figure 7 is drawn
from the fourth line in Figure 8, partway through the process.
6 The OM-SuperVP library provides the means to run most of the AudioSculpt sound analysis and
processing features as part of OpenMusic visual programs. This has the advantage of connecting them
directly to the compositional workflow and iterative procedures.
163
Aaron Einbond
Figure 7. What the Blind See mm. 235-238 “virtual polyphony”.

c Edition Gravis Verlag,
Brühl, Germany. Printed with kind permission.
This rhythmic matrix is hierarchical,7 as each transient detected at a given threshold

is also detected at a lower threshold. However the notated rhythms output by omquan-
tify are not strictly hierarchical, as at different levels the same rhythmic position in
milliseconds may be approximated to different notated values depending on its context.
These differing approximations were retained as-is, motivated both by readability for
the performer, and providing an extra level of rhythmic micro-variation for potential
expressive use. Such a rhythmic matrix, which has functioned in many of my recent
works, could be heard as an extension of common-practice metre. It is intriguingly
parallel to the hierarchical rhythmic structure defined by Lerdahl and Jackendoff in their
Generative Theory of Tonal Music (1983) [12].8 A compositional realisation can freely
draw upon the matrix, crossing rhythmic levels for expressive purposes and underlining
7 In the sense of Lerdahl and Jackendoff ([12], p. 13); see below.

8 The rhythmic structure shown in Figure 8 satisfies “Metrical Well-Formedness Rules” 1 and 2, which
Lerdahl and Jackendoff state are “defining conditions for metrical structures and are universal”, but
not rules 3 and 4, which “define the metrical regularities possible within a given musical idiom”. This
makes the structure comparable to a phonological “stress grid” (Fred Lerdahl, personal communication,
4 September 2015).
164
Figure 8. OpenMusic patch to apply transient-detection to a sound file at successive threshold

values (left) with sub-patch transient_detection (right).
metrically strong attacks-points for emphasis. Of course unlike common-practice music,

this rhythmic matrix is superposed on a contrasting regular metrical notation, but as
mentioned above the metre serves a mere coordinating function.
Imaginary Gardens
In Without Words for soprano, 11 instruments, and electronics (2012), the sample data-
bases and field recordings that have informed my work since What the Blind See are
augmented by a database of vocal texts. The database was drawn from fragments,
translations, and paraphrases of writers including Marianne Moore, Wallace Stevens,
Matsuo Bashō, Douglas Huebler, Kenneth Goldsmith, and many other prose and poetic
texts—chosen because of their connection to transcription, mimesis, metaphor, and
place. They were recorded in collaboration with soprano Amanda DeBoer Bartlett who
improvised with the texts as well as with various vocal “preparations”—objects to be
sung through, including whistle, kazoo, coffee mug, cardboard tube, and spring drum.9
In analogy to the examples above, here audio recordings of the texts are privileged over
their symbolic printed versions.
This vocal database is treated as a mobile resource, ranging from isolated phonemes
to full words, phrases, or occasional sentences. Accordingly semantic references may be
entirely lost or more or less intelligible depending on vocal technique and whether or not
a preparation is present. This continuum of semanticity is exploited in the transcription
process, with smaller units available for audio mosaics suggesting other references, and
longer units conveying intact linguistic denotations of their own. An example of the
9 For a detailed account see [6].
165
Aaron Einbond
former is the opening mosaic, performed by the live soprano, based on a target field
recording of frogs. The work opens with the field recording alone, then gradually cross-
fades it with a recorded audio mosaic of soprano samples, and eventually integrates the
live soprano herself in a seamless continuum. The tissue of whispered phonemes (square
note heads in Figure 9), principally fricatives, sometimes sung through a police whistle
(upper staff), imitates the high spectral range of the frogs, with breathy sung high D5
(diamond note heads) pointing to a prominent frequency of their call. The same field
recording receives a contrasting transcription for the full instrumental ensemble a few
minutes later in the work, by a similar technique to Figures 1 and 2.
Figure 9. Without Words mm. 65-69 soprano part.

c Edition Gravis Verlag, Brühl, Germany.
Printed with kind permission.
In the vocal transcription shown in Figure 9, most of the phonemes are far too short
for their source texts to be identifiable, while a few open out into words or phrases
(“more in”, “okay yeah”), hinting at the semantic origins of the utterances. Instead the
connotation of the underlying field recording is more salient. Its compositional sense is
later revealed in a soprano citation of Marianne Moore: “imaginary gardens with real
toads in them”.
Over the course of the 18-minute work, these textual fragments are hesitantly aug-
mented, like folding fans that expand and contract. Eventually some vocal samples are
exposed in their entirety, like the Marianne Moore text above. For another text, a haiku
by Matsuo Bashō translated by Robert Hass, the simplest possible transcription process
is applied: the original soprano sample is re-notated as faithfully as possible with the
aid of AudioSculpt and OpenMusic. A chord-sequence analysis is exported as an
SDIF file, imported to OpenMusic, and the most salient monophonic pitch is chosen
from each segment. Nevertheless the “accuracy” of the result is belied by the subjective
decisions throughout the process, essential to capture the fine shades of rhythm and pitch
(Figure 10).
Figure 10. Without Words mm. 266-69, soprano part.
When the same soprano who created the original sample reinterprets the transcrip-
tion, she reproduces the historical/autobiographical moment of the sampling session—
the sample becomes a field recording. Yet as the soprano relearns to perform “herself”
166
she underlines the tension between performance, transcription, and referent. As Peter
Ablinger says of his Weiss/Weisslich 36, “the same is not the same. There is a difference.
At least the difference between just being here and: listening”.10 Perhaps the same
difference lies behind Basho’s riddle-like haiku.
Traces
Textual fragments are taken as targets for audio mosaics as well as corpora. Brief excerpts
of Wallace Stevens’s own recorded reading of Credences of Summer are cited, placing
historical recordings of the poet in dialogue with the live soprano. One recording was
taken as the basis of an audio mosaic using a database of bass flute samples. The
patch in Figure 11 adds another feature to the CBCS workflow: textual metadata
output by CataRT are recorded in the SDIF file in name-value tables. Then using
OpenMusic objects getnvtlist and find-in-nvtlist the values can be read and used to
inform the manual editing of the score in Finale, as seen in Figures 11-12. For example,
evaluating the mapcar box labelled “SoundFile names” (Figure 11) yields the filename
“bfl-closed_o_harmonics-C3” for the second sample (in both the poly object in Figure 11
and the score in Figure 12). The pitch calculated by yin∼ is discarded and replaced with
fingered pitch C3 documented in the filename. To indicate the flutist’s closed embouchure
(lips sealed around the mouthpiece), a black circle is added over a hollow rectangular
note head. Finally, the flutist is directed to shape her mouth into the phoneme [o]. This
process is repeated for each sample to build the bass flute part.
Reminiscent of the technique of Figure 8, this audio mosaic was made in CataRT
with a range of different segmentations based on loudness threshold to produce mosaics of
different densities. Beyond the contextual variations in rhythmic transcription produced
by omquantify, there are also variations in the units chosen by CataRT according to the
segmentation. As CataRT calculates an average of descriptor values over each segment,
segments of different lengths corresponding to the same position in the target will have
slightly different average values, and as a result the best match in the database might
be different. So successive rhythms and playing techniques in the final matrix do not
align precisely, even though derived from the same mosaic. This is a case in point of
how slight variations in parameters can produce significantly different musical outcomes.
The durations of the bass flute samples themselves also come into play, for example
the 4-second tongued overtone glissando that concludes the third line of Figure 12.
During transcription a subjective compromise is made between the durations of the
target segments and the durations of the samples segments in the database.
These bass flute lines are reinterpreted in the finished work accompanied by repeti-
tions of the Wallace Stevens target sample itself: “Trace the gold sun about the whitened
sky/Without evasion by a single metaphor”. As the sample loops, the varying bass flute
lines are deployed in a gradual densification, in contrast to a gradual descrescendo by the
voice of Stevens. Instead of a syllable-by-syllable transcription, the bass flute traces a
more subtle mosaic privileging the weight and density of the grainy historical recording,
and emphasising the timbre of the poet reading over the sense of the printed text.
10 See http://ablinger.mur.at/docu1515.html
167
Aaron Einbond
Figure 11. Patch for Without Words with sub-patch get-SoundFiles on the left.
Figure 12. Transcriptions of the same Wallace Stevens sample for bass flute at increasing
loudness segmentation thresholds, subjectively edited in Finale.
168
At the end of the work, the field recording, the Stevens quotation, and their varied
transcriptions return in a collage-like coda. While the recapitulation is perceptible on
a semantic level, it also suggests an abstract musical return. The recapitulation of the
same constellation of timbres, playing techniques, and spectral descriptors defined by a
concrete recording and/or its transcription can be heard in analogy to the tonal return of
common-practice music. In fact, given the statistical interpretation of tonality by scholars
like Carol Krumhansl [10] and David Temperley [19], as well as the spatial tonal models
advanced by Fred Lerdahl [11] and others, a direct connection could be made between the
pitch distribution of a tonal centre and the statistical distribution of descriptor weights
in a soundscape. Tonality as a place: a topic that invites further exploration.
Translation
Since 2012, transcription with CataRT and OpenMusic has advanced in several di-
rections: the bach package for Max has been integrated into the CataRT workflow to
facilitate real-time interaction with musical notation. Bach can also expedite exchange of
data with OpenMusic, which remains useful for high-level computer-aided composition
tasks. CataRT itself is available in a new version incorporating the MuBu package
of Max externals, improving its portability, clarity, and especially permitting access to
larger audio databases by taking advantage of 64-bit Max.
Posing the question of how mimesis can be generalised beyond strict transcription,
I integrated CataRT with a machine-learning algorithm that permits the association
of audio not only by descriptor similarity, but by shared context. Inspired by the
existing program OMax [1] and the PyOracle package for Python [18], the result-
ing tool CatOracle is available in the MuBuForMax-1.8.5 distribution and later.
It combines the powerful Audio Oracle algorithm for musical pattern matching with
CBCS for applications including computer improvisation, high-level control of synthesis,
computer-assisted composition, and musicological analysis. CatOracle extends the
notion of transcription beyond a reproduction coordinated linearly in time, permitting
an “improvisation” based on a target that can backtrack, repeat, or skip. Yet the shared
contextual relationships compel the result to bear a degree of resemblance of the original.
This could be described as behavioural transcription or “style imitation” [3]—or more
generally “translation”,11 a term chosen to suggest displacement of elements along the
time axis, as well as expression through a new improvisational language. Applied to
machine improvisation, it is especially complementary to other sites of spontaneity in
the transcription process: collaborative improvisation during sampling sessions, gestural
control with the mouse or Wacom tablet, and live reinterpretation of expressive micro-
timing.
A first compositional application in Xylography for violoncello and electronics (2015)
begins with an acoustic improvisation by cellist Pierre Morlet, transcribed in detail
in the score and reinterpreted in performance, translated live into multiple strands of
computer improvisation, the courses of which are further guided by ongoing transcription
of the live performer. This layered process of mimetic and creative feedback points to
promising territory at the intersection of composition, transcription, improvisation, and
interpretation.
11 An homage to the “central dogma” of molecular biology: “replication–transcription–translation” [4].
169
Aaron Einbond
References
[1] Gérard Assayag, Georges Bloch, Marc Chemillier, Arshia Cont, Schlomo Dubnov.
“OMax Brothers: A Dynamic Topology of Agents for Improvisation Learning”. In
Proceedings of the ACM Multimedia Workshop on Audio and Music Computing for
Multimedia. Santa Barbara, 2006.
[2] Michel Chion. Guide des Objets Sonores: Pierre Schaeffer et la recherche musicale.
Paris: Institut National de l’Audiovisuel & Editions Buchet-Chastel, 1983-1995.
English translation: Guide To Sound Objects. Pierre Schaeffer and Musical Research
by John Dack and Christine North, 2009.
[3] Arshia Cont, Schlomo Dubnov, Gérard Assayag. “A Framework for Anticipatory
Machine Improvisation and Style Imitation”. In Anticipatory Behavior in Adaptive
Learning Systems. Rome, 2006.
[4] Francis Crick. “Central Dogma of Molecular Biology”. Nature, 227, 1970.
[5] Nicolas Donin. “Sonic Imprints: Instrumental Resynthesis in Contemporary

Composition”. In Gianmario Borio (ed.) Musical Listening in the Age of
Technological Reproducibility. Farnham: Ashgate, 2015.
[6] Aaron Einbond. “Subtractive Synthesis: noise and digital (un)creativity”. In Aaron
Cassidy, Aaron Einbond (eds.) Noise In And As Music. University of Huddersfield
Press, 2013.
[7] Aaron Einbond, Diemo Schwarz, Jean Bresson. “Corpus-based transcription as an

approach to the compositional control of timbre”. In Proceedings of the International
Computer Music Conference. Montreal, 2009.
[8] John M. Grey. “Multidimensional perceptual scaling of musical timbres”. Journal

of the Acoustical Society of America, 61(5), 1977.
[9] Vijay Iyer. “Embodied Mind, Situated Cognition, and Expressive Microtiming in
African-American Music”. Music Perception, 19(3), 2002.
[10] Carol L. Krumhansl. Cognitive Foundations of Musical Pitch. Oxford University

Press, 1990.
[11] Fred Lerdahl. Tonal Pitch Space. Oxford University Press, 2001.
[12] Fred Lerdahl, Ray Jackendoff. Generative Theory of Tonal Music. Cambridge, MA:
MIT Press, 1983.
[13] Theodore Levin. Where Rivers and Mountains Sing. Bloomington: Indiana
University Press, 2010.
[14] Rainer Nonnenmann. “Music with Images—The Development of Helmut Lachen-

mann’s Sound Composition Between Concretion and Transcendence”. Contemporary
Music Review, 24(1), 2005.
170
[15] James O’Callaghan. “Mimetic Instrumental Resynthesis”. Organised Sound, 20(2),

2015.
[16] David Ryan, Helmut Lachenmann. “Composer in Interview: Helmut Lachenmann”.
Tempo, 210, 1999.
[17] Diemo Schwarz, Grégory Beller, Bruno Verbrugghe, Sam Britton. “Real-Time
Corpus-Based Concatenative Synthesis with CataRT”. In Proceedings of the
International Conference on Digital Audio Effects (DAFx). Montreal, 2006.
[18] Greg Surges, Shlomo Dubnov. “Feature Selection and Composition using PyOracle”.
In AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment.
Boston, 2013.
[19] David Temperley. The Cognition of Basic Musical Structures. Cambridge, MA: MIT
Press, 2001.
[20] Tolga Tüzün. “Maquette As Data Structure and Synthesis Agent in Metathesis”.
In Jean Bresson, Carlos Agon, Gérard Assayag (eds.) The OM Composer’s Book 2.
Editions Delatour France/IRCAM-Centre Pompidou, 2008.
[21] David Wessel. “Timbre Space as a Musical Control Structure”. Computer Music
Journal, 3(2), 1979.
Acknowledgements: I thank the musicians of Ensemble L’Instant Donné with whom I developed
the instrumental samples and techniques for What the Blind See and who premiered the work at the
Centquatre in Paris on June 12th, 2009. I thank soprano Amanda DeBoer Bartlett and the musicians
of Ensemble Dal Niente with whom I developed the samples and techniques for Without Words and
who premiered and recorded the work in Chicago on my portrait album released by Carrier Records
in 2012 (http://carrierrecords.com/index.php?album=einbond). I gratefully acknowledge researchers
at IRCAM including Diemo Schwarz and members of the ISMM team for developing and supporting
the CataRT package, Eric Daubresse for production support on What the Blind See, Jean Bresson,
Yan Maresz, Alexis Baskind, Emmanuel Jourdan, Mikhaïl Malt, Jean Lochard, and Cyril Beros. My
participation in two years of the Cursus in Composition and Music Technology was made possible by
the Fulbright Scholarship and Georges Ladd Prix de Paris.
171
Recomposing Beethoven with Music
Neurotechnology
Eduardo Reck Miranda and Anders Vinjar
Musicians have an extraordinary opportunity today to develop new approaches to com-

position that would have been unthinkable a few years ago. Imagine if you could play a
musical instrument with signals detected directly from your brain. Would it be possible
to generate music representing brain activity? What would the music of our brains sound
like? These are some of the questions addressed by research into Music Neurotechnology,1
an emerging field at the crossroads of music, technology and neuroscience.
There have been a great number of interesting initiatives in the last decade to sonify
brainwaves [1], some of which might indeed be useful for creative musical purposes [4].
The burgeoning field of Brain-Computer Music Interfacing (BCMI) is developing powerful
methods to generate music in real time by means of brainwave signals [6], even looking
into harnessing the potential of bio-medically uncertified low-cost equipment for BCMI
applications [3]. However, in this chapter we discuss an approach that goes beyond
sonification of brainwaves and BCMI to explore the impact of Music Neurotechnology to
the field of Computer-Aided Composition (CAC). We introduce algorithms that we have
been developing to compose orchestral music offline with fMRI2 brain scans.
We have an opportunity to take advantage of the fMRI brain scanning method
as we are not concerned with real-time interaction. This method is too cumbersome
for real-time applications, but considerably more powerful and informative than EEG
(electroencephalogram) scanning, which is the method used in sonification and BCMI
research [12].
The compositional approaches introduced below were developed in OpenMusic, orig-
inally to generate materials for two symphonic works by Eduardo R. Miranda, Symphony
of Minds Listening (2013) and Corpus Callosum (2015), as well as Shockwaves (2015),
a concertino for violin and orchestra.3 The discussions in this chapter will be mostly on
the first two works.
We begin the chapter by briefly introducing Miranda’s approach to composing with
the aid of computers, focusing on using algorithmically generated materials. Then we
1 Theterm “Music Neurotechnology”, coined by Miranda and colleagues, appeared in print for the first
time in 2009 [8].
2 fMRI stands for “functional magnetic resonance imaging” and can be used to show which parts of the
brain are involved in a particular mental process.
3 IRCAM’s ATO-MS software was also used to compose Corpus Callosum, and Modalys and
AudioSculpt were used in addition to ATO-MS and OpenMusic to compose Shockwaves.
173
introduce the compositions Symphony of Minds Listening and Corpus Callosum. Next,
we focus on the technical details of collecting and handling fMRI data, followed by
an overview of the OpenMusic patches that we developed for this project, and an
explanation of how ATO-MS software was used to generate orchestrations based on
fMRI information.
On computer-aided musical creativity

In the book Thinking Music [7] Miranda writes about earworms that every now and then
he finds stuck in his mind. Such earworms are often recognisable excerpts from music that
was heard before, but sometimes can not be clearly identified. Could these unrecognisable
earworms be new tunes that the brain creates from scratch? Or distorted versions of
music that one might have heard before? Moreover, earworms are not necessarily “tunes”:
rhythms, timbres, sound textures, sound effects, orchestral passages, and so on, often get
stuck as well. Every so often, they do not match any music that was heard before. They
seem to be distortions of existing music, or evoke only particular aspects, such as rhythm
or harmonic progression.
We hypothesise that the means by which those earworms emerge are manifestations of
some form of musical creativity, and that it is possible to harness such creative processes
by working with computer-generated materials. We are convinced that the process of
designing generative musical algorithms, programming them on a computer, and handling
the computer-generated materials, can flush out from a composer’s mind’s ear earworms
that would not have emerged otherwise.
One might wish to make a distinction here between earworms and new musical ideas,
but we are not entirely sure where the dividing line is. What is deemed important
here, however, is that interaction with computer-generated materials can flush all sorts
of musical ideas out and they often evolve and mingle with new ideas, transformations,
variations, and so forth.
Philosopher Friedrich Nietzsche suggests that great artistic creations can only result
from the articulation of a mythological dichotomy that he refers to as the Apollonian
and the Dionysian [11]. In ancient Greek mythology, Apollo is the god of the sun and
is associated with rational and logical thinking, self-control, and order. Conversely,
Dionysus is the god of wine and is associated with irrationalism, intuition, passion, and
anarchy. These two gods represent two conflicting creative drives, constantly stimulating,
provoking one another. The notion that the Apollonian and the Dionysian tend to counter
each other reminds us of the way in which the brain functions at all levels. Inhibitory
processes pervade brain functions, from the microscopic level of neurons communicating
with one another, to the macroscopic level of interaction between larger networks of
millions of neurons. Indeed, there are parts of the human brain that appear to be
undeniably Apollonian, whereas others are outrageously Dionysian. The Apollonian
brain includes largely the frontal lobe of the cortex and the left hemisphere. Generally,
these areas are in charge of focusing attention to detail, seeing wholes in terms of their
constituents, and making abstractions. It is systematic and logical. Conversely, the
Dionysian brain includes sub-cortical areas, which are much older in the evolutionary
timeline, and the right hemisphere. It is more connected to our emotions. It perceives
the world holistically and pushes us towards broader or more general views.
174
The Apollonian brain is concerned with unilateral meanings, whereas the Dionysian
brain tends to forge connections between allegedly unrelated concepts. This is what our
minds do all the time when composing: the further our Apollonian brains push us to
perceive the world according to its agenda, the stronger the pull of our Dionysian brains
to perceive the world differently. Hence, computer technology is of foremost importance
for our métier, because it allows us to stretch our Apollonian musical side far beyond our
ability to do so by hand, prompting our Dionysian side to counteract accordingly. This
cognitive push and pull seems to be the driving force behind our musical creativity.
Background: Symphony of Minds Listening

Symphony of Minds Listening was a first attempt at putting in practice our Music
Neurotechnology approach to CAC, which still is evolving [9]. It is an experimental
symphonic work in three movements based on fMRI brain scans taken from three different
people while they listened to the second movement of Ludwig van Beethoven’s 7th
Symphony: a ballerina, a philosopher, and a composer (Miranda).
The fMRI brain scanning method measures brain activity by detecting associated
changes in blood flow. The measurements can be presented graphically by colour-coding
the strength of activation across the brain. Figure 1 shows a representation of an fMRI
scan of Miranda’s brain listening to Beethoven’s music during a specific window of time.
In this case, each window lasts for 2 seconds. The figure shows eight planar surfaces,
or slices, from the top to the bottom of the brain. Figure 2 shows an example of a 3D
rendition of such an fMRI scan, devised by Dan Lloyd: it displays different areas of the
brain, represented by different colours (or shades of grey), responding in a coordinate
manner to the music.
Figure 1. A representation of an fMRI snapshot, showing 8 transversal slices of the brain.
175
Figure 2. An artistic 3D rendering of an fMRI scan.
Each scanning session generated sets of fMRI data, each of which we associated
to a measure of the second movement of Beethoven’s 7th symphony. This is shown
schematically in Figure 3 and will be explained in more detail later.
Figure 3. The result of a scanning section is a set of fMRI data for each measure of Beethoven’s
work. (Note: this is only a schematic representation; the brain images do not correspond to the
actual music shown.)
The score of Beethoven’s movement (using a MIDI representation) was cut into
measures and analysed with Artificial Intelligence (AI) software developed at ICCMR4
[2, 10]. This software extracted statistical information about the structure of the music
such as melodic direction, tonal signature, intervals between notes, rhythmic structure,
4 Interdisciplinary Centre for Computer Music Research, Plymouth University, UK.
176
and so on. This information was then used by the AI software to reconstruct the
Beethoven movement, but with fMRI data influencing the process. Not surprisingly,
the fMRI scans differed amongst the three listeners. Therefore, brain activity from three
different minds yielded three different movements for the resulting composition, each of
which displaying varied degrees of resemblance to the original symphony.
In order to test the composition methods (or the “reconstruction” method) thor-
oughly, the composer decided to remain faithful to the original form of Beethoven’s
movement and avoid as much as possible editing the musical reconstructions generated
automatically by the system. The objective was to gain a good understanding of the
outcomes in a practical real-world, large-scale compositional task. However, this decision
heavily constrained his musical imagination during the composition of Symphony of
Minds Listening. One could say that the compositional process as whole was more
Apollonian here than Dionysian. Even though the piece caught the imagination of the
audience and received positive press reviews,5 the composer felt that the æsthetic results
lacked his creative touch and musical experience. Nevertheless, it was a useful experiment
that was followed up with Corpus Callossum and subsequently Shockwaves.
Corpus Callosum
In Corpus Callosum the composer allowed more freedom in his handling of the materials
produced by the computer. The compositional process was more Dionysian here than
Apollonian. And this time he worked with the fMRI data from his own brain only.
The title of the composition refers to the part of the brain that connects the left and
right hemispheres and facilitates communication between them. As already mentioned,
the left hemisphere is largely engaged in processing details. It is often associated with a
more objective, or scientific, knowledge of the world. Conversely, the right hemisphere is
largely engaged in taking a holistic view and is often associated with a more subjective,
or poetic, interpretation of the world [14]. One should bear in mind that this distinction
is controversial because it has been difficult to observe it clearly. In a normal brain
the two hemispheres work tightly together: they are highly interconnected and interact
through the corpus callosum. Nevertheless this notion is inspiring as philosophical and
artistic metaphor.
During the scanning session Miranda recalls that sometimes he lost concentration on
the music and his mind wandered off. We reckon that this is a typical manifestation of
brain asymmetry at work: while one side of his brain was striving to pay attention to
musical detail, the other was making mental associations, producing imageries, eliciting
feelings, and so on.
The orchestra for Corpus Callosum is divided into two groups: one to be placed on the
left side of the stage and the other on the right side, representing respectively the left and
right hemispheres of the brain. The composition develops as an interaction between these
two realms. The instruments on the right (2 flutes, 2 oboes, 2 clarinets, bass clarinet, 2
bassoons, 2 French horns, 2 trumpets, trombone, bass trombone, percussion—2 players,
5 Thesymphony was premiered on 23 February 2013 by Ten Tors Orchestra, conducted by Simon Ible, at
Peninsula Arts Contemporary Music Festival, Plymouth, UK. See a review in Gramophone magazine
and news story in The Times: http://goo.gl/FDM0iX and http://goo.gl/GtacVu
177
Figure 4. Excerpt from the score of Corpus Callosum, showing music representing the right
hemisphere.
Figure 5. Excerpt from the score of Corpus Callosum, showing a string quartet representing
the left hemisphere.
178
2 violins, viola, and violoncello) play segments that were composed emphasising orches-
tration (Figure 4). These passages do not handle rhythm and melody explicitly. Rather,
the instruments play clusters of sustained notes; the focus is on timbre. Conversely,
the instruments on the left (marimba or piano, 2 violins, viola, and violoncello) play
modifications of passages pinched from Beethoven’s score (Figure 5). These segments
are noticeably rhythmic and melodic, and timbre is secondary.
In addition to programming OpenMusic patches to make transformations informed
by fMRI data, in Corpus Callosum we also used computer-aided orchestration software
to generate the orchestrations for the “right side” of the composition. Before we discuss
the orchestrations and transformations, below is brief explanation of how the fMRI data
were handled.
The fMRI data

The time resolution of the Siemens Allegra 3T scanner that we used to collect the brain
data is 2 seconds. That is, it takes 2 seconds for a snapshot comprising 36 image
slices of the brain. Each slice comprises 64 × 64 picture elements, known as voxels, or
volume pixels, totaling approximated 150,000 continuously varying voxels per slice. The
participants were scanned listening to the second movement of Beethoven’s 7th Symphony
twice. The scanning began with 30 seconds without music, then 460 seconds of music,
then 18 seconds without music, and finally more 460 seconds of music again. Thus each
run generated 484 snapshots of 150,000 voxels each.
The raw fMRI data were first pre-processed following standard procedures for func-
tional neuro-imaging using Statistical Parametric Mapping software [15]. The large
number of voxels is exceedingly complex for direct analysis, so data were further processed
with Independent Component Analysis, or ICA [13]. Basically, ICA analysis identifies
groups of voxels that oscillate in unison. These are unified as supervoxels representing
temporally coherent networks of brain activity. The different levels of grey and colours
on the 3D renditions shown in Figures 2 and 3 correspond to ICA components.
In total, 25 ICA components were calculated from the fMRI data. To rank these
components in order of significance, the activity of each component during the first
scanning pass through the listening was compared to that same component during the
second pass. If the two segments of a component time series were correlated, (with
p < 0.05),6 then we hypothesised that the activity might be musically driven, as the
musical stimulus would have been identical at the corresponding time points in the two
passes. The strengths of the 25 ICA components were ranked as follows: 25, 15, 14, 8,
5, 10, 11, 18, 6, 2, 4, 1, 17, 16, 13, 20, 21, 3, 22, 24, 12, 7, 9, 23 and 19.
The varying components were resampled to match the timing of the Beethoven score
measure by measure. The movement comprises 278 measures; therefore, each ICA
component comprises a time series of 278 values, normalised from 0 (meaning lowest
fMRI intensity) to 9 (highest fMRI intensity). As an example, Table 1 shows the values
of the first 5 strongest ICA components (that is, 25, 15, 14, 8 and 5, with p < 0.002) for
the first 10 measures.
6 In psychology, the p-value is the probability of obtaining the same result by chance.
179
Beethoven measure ICA 25 ICA 15 ICA 14 ICA 8 ICA 5

1 7 5 5 5 2
2 5 5 8 5 8
3 7 3 5 5 6
4 5 8 3 5 2
5 5 7 4 4 4
6 6 6 4 5 3
7 7 8 5 6 3
8 4 6 3 4 3
9 6 6 4 5 4
10 5 7 5 5 3
Table 1. The values of the strongest 5 ICA components for the first 10 measures of the
Beethoven movement yielded by the fMRI from Miranda.
In order to enable easy access to the ICA components during the composition work-
flow, the resulting set of ICA components were stored in a class-array object in Open-
Music (Figure 6). This facilitated accessing individual components, and specific lookups
based on time (Figures 7 and 8). Having the plots of the ICA components visually
available in the class-array also provided a valuable overview of the data during the
compositional process.
Figure 6. ICA components stored in a class-array object in OpenMusic.
180
Figure 7. Selecting value from ICA component at a specific measure.
Figure 8. Selecting components from total set of ICA components. This enables the composer
to select ICA values from a total of 25 options.
Right side: generative orchestration

The materials to compose for the group of instruments representing the right side of
the brain were created with the ATO-MS orchestration software [5].7 We hacked the
Matlab code of ATO-MS in order to enable it to take into account brain data in
generating orchestrations. The process of orchestration for Corpus Callosum is illustrated
in Figure 9.
7 ATO-MS is unreleased software for computer-aided orchestration developed at IRCAM, preceding the
current Orchids system.
181
Figure 9. The process of generating orchestrations.
The system takes a given audio file, in this case a section from the recording of the
Beethoven movement, and analyses its spectrum using a Fast Fourier Transform (FFT)
with a window lasting two seconds, which roughly corresponds to one measure of the
musical score. This analysis enabled us to extract a set of 25 frequency values (in Hz)
for the most prominent partials of the spectrum at every two seconds of audio.
The partials of a sound spectrum can be described by frequency and amplitude
values; however, we discarded the amplitudes produced by the FFT analysis. Instead,
we replaced the amplitudes by the intensity of the 25 ICA components associated to the
respective measures of the Beethoven movement. This resulted in what we refer to as
the fMRI-modulated spectrum.
The ATO-MS audio database contains spectral analyses of recordings from the
instruments of a full symphonic orchestra. It contains information for the entire range
of each instrument, including different dynamics, articulations, and playing techniques.
Given a list of instruments, for which one wants the system to produce an orchestration,
the system searches the database for combinations of sounds whose blended spectra best
approximate the target fMRI-modulated spectrum. The system produces a number of
suggestions for the composer to work with.
As an example, let us examine how the first measure of the excerpt shown in Figure 4
was composed. The audio segment (Figure 10) that was used as a target for the
orchestration of the four measures of the score shown in Figure 4 corresponds to measures
139-147 of the Beethoven movement. The 25 frequencies generated by the FFT analysis of
the first 2 seconds of this sound are shown in Table 2, next to the ICA components of the
fMRI taken at the moment Miranda was in the scanner listening to the corresponding
measure. The values of the ICA components replaced the original amplitudes from
the FFT analysis, producing the fMRI-modulated spectral target for the orchestration
process.
182
Figure 10. The waveform of the audio recording of measures 139-147 of the Beethoven
movement, lasting 8 seconds.
Partial Frequency in Hz Amplitude (0-9)

1 220.35 6 (ICA 25)
2 381.73 5 (ICA 15)
3 592.57 7 (ICA 14)
4 805.02 6 (ICA 8)
5 1063.28 4 (ICA 5)
6 1210.06 5 (ICA 10)
7 1432.21 5 (ICA 11)
8 1648.53 2 (ICA 18)
9 1812.22 5 (ICA 6)
10 2099.66 5 (ICA 2)
11 2411.76 7 (ICA 4)
12 2466.86 3 (ICA 1)
13 2701.88 4 (ICA 17)
14 2900.26 3 (ICA 16)
15 3206.43 3 (ICA 13)
16 3297.83 6 (ICA 20)
17 3586.17 5 (ICA 21)
18 3947.55 6 (ICA 3)
19 3980.26 6 (ICA 22)
20 4241.45 3 (ICA 24)
21 4535.72 6 (ICA 12)
22 4788.87 4 (ICA 7)
23 4920.14 5 (ICA 9)
24 5018.24 2 (ICA 23)
25 5395.23 4 (ICA 19)
Table 2. The fMRI-modulated spectrum corresponding to measure 139 of the Beethoven

movement.
In this case the instruments specified were 2 flutes, 2 oboes, 2 clarinets, bass clarinet,
and 2 bassoons. Among the various suggestions generated by the system, the results
shown in the first measure of Figure 4 were selected. Note that oboe 1 was not used, and
that individual notes have different dynamics, which are necessary to obtain the desired
spectral behaviour.
183
Left side: musical transformations with OpenMusic

The materials to compose for the group of instruments representing the left side of the
brain were created using a number of transformation algorithms. In these algorithms, the
amount of modification is scaled according to an index, referred to as the fMRI_index,
which is the value of the ICA analysis extrapolated from the fMRI scans, as explained
earlier.
The fMRI_index control operates as follow. First, a “raw” (maximal) transformation
is generated, and a difference value d between the input and the transformed music is
calculated. Then this is multiplied by fMRI_index to give a final scaled modifier value,
or SMV. The SMV gives a degree of fMRI-controlled variability in each transformation:
a high fMRI_index value will result in significant transformations to the music, whereas
a low fMRI_index value will result in less significant transformations (see Figure 11).
A typical patch for fMRI-controlled interpolation is shown in Figure 12.
Figure 11. Interpolator patch for two lists of values using a scaled modifier.
In the following section we introduce the four transformation patches which were
primarily used in Corpus Callosum, describing two of them in detail.
184
Figure 12. A typical patch for fMRI-controlled interpolation.
Pitch inversion algorithm, modulated by brain activity

The first algorithm inverts a pitch sequence about the axis E4 (MIDI note 64) by
subtracting each MIDI pitch from 128. For example, the sequence {B4, B4, D5, C5,
B4} (beginning on MIDI note 71), inverts to {A3, A3, F]3, G]3, A3} (beginning on
MIDI note 57), as shown in Figure 13.
Figure 13. An example of a short musical sequence and its inversion.
The basic example above assumes a maximal transformation, corresponding to an

fMRI_index value equal to 1.0. However, varied degrees of transformations are possible
by scaling according to the value of fMRI_index. The difference between the input and
the transformed interim output (i.e. the would-be maximal inversion) is multiplied by
fMRI_index before being summed with the input to create the final transformed output
value, as follows:
NewPitch = InputPitch + ((InputPitch − InterimOutput) × fMRI_index)
185
Let us examine what happens if we set the fMRI_index = 0.6. In this case, we would
expect an output approximately half way between the original pitch and the inversion;
in other words, an almost neutral set of intervals. The result when applied to the pitch
sequence on the left of Figure 13 is {62.6, 62.6, 62, 62, 62.6}, yielding {63, 63, 62, 62,
63} when rounded up to the nearest whole number, and corresponding to D]4, D]4, D4,
D4, D]4 as shown in Figure 14.
Figure 14. Sequence after inversion with fMRI_index = 0.6, giving a nearly neutral set of
pitch intervals.
Pitch-scrambling algorithm
An algorithm to reorder the pitches of the input signal randomly provides an simple
stochastic transformation. Using the musical example in Figure 13 as a starting point,
the result of applying this transformation four times could produce, for instance, the
sequences illustrated in Figure 15.
Figure 15. The result of applying the pitch-scrambling algorithm four times on the same input.
As with the pitch inversion algorithm, the fMRI_index can be used to create a control
signal with which the amount of transformation can be varied. In order to illustrate this,
let us assume an fMRI_index = 0.4. Considering the same input measure as before and
the transformed values from the first pitch scramble shown in Figure 15, the value of d,
between the first scramble and the input sequence is calculated as follows:
d = {(74 − 71), (72 − 71), (71 − 74), (71 − 72), (71 − 71)} = {3, 1, −3, −1, 0}
The scaled modifier values are then calculated by multiplying the difference values
by fMRI_index = 0.4:
SMV = {1.2, 0.4, −1.2, −0.4, 0}
Finally, the SMV values are summed with the values of the original input to give a
transformed set of output values:
New Pitches = {72.2, 71.4, 72.8, 71.6, 71}
As with the previous example, the resulting values are rounded up to the nearest
whole number, giving a transformed set of MIDI values equal to {72, 71, 73, 72, 71},
which is rendered as {C5, B4, C]5, C5, B4}, as shown in Figure 16. Note that the
output is significantly closer, in overall structure, to the unscrambled input than to the
first scrambled transformation.
186
Figure 16. Transformed output created by the pitch scrambling algorithm assuming
fMRI_index = 0.4.
Variations based on stylistic analysis

Methods for analysing musical style, mostly based on statistical methods, were tested and
developed in our quest for interesting compositional materials derived from Beethoven’s
original movement. To this end we applied statistical analysis and compression algo-
rithms using the LZ library.
Statistical analysis and compression algorithms can generate new material based on
probabilistic processes, effectively rendering variations of the same material on each
evaluation of the patch. Both approaches provide parameters to control the degree
of proximity to the original input, allowing the composer to generate and to choose
material along a scale from close (fairly similar) to remote (very dissimilar), controlled
by the fMRI_index parameter (see Figure 17).
Figure 17. Generating variations based on fMRI-modulated style analysis using the LZ library
(the fMRI index corresponds to the variable niter in the patch).
187
Variations generated via sound analysis

Another useful method to process musical results was to generate variations based on
analysis of recordings of the original Beethoven symphony. This procedure was used
significantly in Corpus Callosum as well as Shockwaves.
Sound analysis algorithms, such as FFT-based analysis, partial tracking, and fun-
damental pitch estimation, provide diverse parameters for variation. For example, one
can obtain different results by varying time and frequency resolutions (“sensitivity” in
Figure 18), as well as analysis threshold (that is, adjusting what the analysis algorithm
considers to be salient or important information in the sound). Although not explicitly
shown in the patch in Figure 18, fMRI_index was also used to modulate these variation
methods.
Including these variations as an intermediate step, the composer can select results, in
whole or in part, to work with subsequently. This process provides interesting possibilities
for controlling composition along a linear entropy scale, from totally predictable to totally
unpredictable as a function of fMRI_index.
Figure 18. Patch for stylistic analysis passing through analysis of audio. Results from sound
analyses are used as inputs to the LZ library functions visible at the bottom of the patch.
188
Composing with computer generated materials

As we mentioned earlier, for the composition of each of the three movements of Symphony
of Minds Listening, the composer strived to follow the form of the original Beethoven
movement. The scheme we introduce below was devised originally for this work, but it
was also adopted for Corpus Callosum and Shockwaves, albeit much more freely.
The composer divided Beethoven’s score into 13 sections as follows:
Section Measures Section Measures
1 1 to 25 8 149 to 183
2 26 to 50 9 184 to 212
3 51 to 74 10 213 to 224
4 75 to 100 11 225 to 247
5 101 to 116 12 248 to 253
6 117 to 138 13 254 to 278
7 139 to 148
Then each section was processed separately and the compositions (or recompositions)
were reconstructed according to the same number of measures as in the original. In other
words, the 13 sections provided a template for the new compositions. They informed the
overarching forms of each of the three movements of Symphony of Minds Listening, and
to some extent the forms of the other two orchestral works. More information on how
this process was carried out is available in [10].
At the lowest level of deconstructing the piece, processing the materials on a section-
by-section basis, and re-assembling the sections, the process tended to be Apollonian,
particularly in Symphony of Minds Listening. During the process of reassembling—or
recomposing—the sections, Miranda deliberately avoided interfering with the computer-
generated materials (informed by fMRI transformations) as much as possible. But of
course he produced a wealth of materials to choose from by experimenting with different
transformation algorithms and running each many times. From these, the composer’s
choices of the particular materials that were used in the finished compositions have no
objective explanation; the process was Dionysian.
As already mentioned, the composer felt that the æsthetic quality of the resulting
work was compromised by the lack of further Dionysian interventions on the computer-
generated transformations and on the musical form. A different approach was adopted for
Corpus Callosum and Shockwaves. Again the processing of materials for these two pieces
was organised according to the same 13 sections shown above. However, the recomposi-
tion process saw no barriers with respect to these sections: some of them were manually
shortened, others enlarged, and others still overlapped. Further, another dimension to the
compositional process was added by the inclusion of computer-generated orchestration,
which was also generated on a section-by-section basis. Miranda then meddled with the
computer-generated materials extensively before they made their way into the final score.
These processes were fundamentally Dionysian and were not documented.
We tend to think of software such as OpenMusic as an aid to an Apollonian compo-
sitional processes. However, Miranda’s Dionysian processes were very much supported by
one of the key features of OpenMusic: the possibility to archive multiple variations of
a process, letting the composer store those that provide potentially interesting material
with which to work further down the road.
189
Figures 19-20 below show one of the patches we made for the recomposition of different
sections using a maquette.
Figure 19. Recomposing Beethoven section-by-section using transformed material.
Figure 20. The maquette for the project.
190
Concluding remarks
This chapter introduces the algorithms and compositional methods that we designed—
informed and inspired by neuroscience—and provides a glimpse of the potential of Music
Neurotechnology for creativity, and more generally, for musical research. Our approach
is inspired by the way in which the brain listens to music: sounds are deconstructed as
soon as they enter the ear. Different streams of neuronally coded data travel through
distinct auditory pathways towards cortical structures, such as the auditory cortex and
beyond, where the data are mingled with data from other senses and memories, and
reconstructed into what is perceived as music.
Computing technology is an invaluable tool for modelling creative processes such
as the ones we introduce above. However, the authors suggest that composers should
not always approach computer-aided composition as a substitute for their own creative
hand and musical intuition. In theory Symphony of Minds Listening could be considered
more elegant than the other two pieces because the composer followed our methods
“by the book”. However, from the three pieces mentioned in this chapter, the one that
the composer considers most æsthetically successful is the one in which he allowed the
most flexibility with respect to amending the outcomes from the transformations and
prescribed musical form: Shockwaves.
The beauty of a tool such as OpenMusic is that it allows composers to model creative
processes and explore them in flexible ways. In addition to being a powerful programming
tool for developing sophisticated models “of” musical creativity, OpenMusic provides
useful support for exploring the potential of Music Neurotechnology to develop models
“for” musical creativity.
References
[1] Gerold Baier, Thomas Hermann, Ulrich Stephani. “Event-based sonification of EEG
rhythms in real time”. Clinical Neurophysiology, 118(6), 2007.
[2] Marcelo Gimenes, Eduardo R. Miranda. “An Ontomemetic Approach to Musical

Intelligence”. In Eduardo R. Miranda (ed.) A-Life for Music: Music and Computer
Models of Living Systems. Middleton, WI: A-R Editions, 2011.
[3] Mick Grierson, Chris Kieffer. “Better Brain Interfacing for the Masses: Progress in
Event-Related Potential Detection using Commercial Brain Computer Interfaces”.
In CHI’11 Extended Abstracts on Human Factors in Computing Systems. Vancouver,
2011.
[4] Thilo Hinterberger. “Orchestral Sonification of Brain Signals and its Application
to Brain-Computer Interfaces and Performing Arts”. In Proceedings of the 2nd
International Workshop on Interactive Sonification. York, 2007.
[5] Yan Maresz. “On Computer-Assisted Orchestration”. Contemporary Music Review,

32(1), 2013.
191
[6] Eduardo R. Miranda. “Brain-Computer Music Interfacing: Interdisciplinary

Research at the Crossroads of Music, Science and Biomedical Engineering”.
In Eduardo R. Miranda, Julien Castet (eds.) Guide to Brain-Computer Music
Interfacing. London: Springer, 2011.
[7] Eduardo R. Miranda. Thinking Music: The inner workings of a composer’s mind.
University of Plymouth Press, 2014.
[8] Eduardo R. Miranda, Larry Bull, François Gueguen, Ivan S. Uroukov. “Computer
Music Meets Unconventional Computing: Towards Sound Synthesis with In Vitro
Neuronal Networks”. Computer Music Journal, 33(1), 2009.
[9] Eduardo R. Miranda, Julien Castet (eds.) Guide to Brain-Computer Music
Interfacing. London: Springer, 2014.
[10] Eduardo R. Miranda, Dan Lloyd, Zoran Josipovich, Duncan Williams. “Creative
Music Neurotechnology with Symphony of Minds Listening”. In Eduardo R.
Miranda, Julien Castet (eds.) Guide to Brain-Computer Music Interfacing. London:
Springer, 2011.
[11] Friedrich Nietzsche. The Birth of Tragedy. Oxford University Press, 2000.
[12] Ramaswamy Palaniappan. “Electroencephalogram-based Brain-Computer Interface:
An Introduction”. In Eduardo R. Miranda, Julien Castet (eds.) Guide to Brain-
Computer Music Interfacing. London: Springer, 2011.
[13] James V. Stone. Independent Component Analysis: A Tutorial Introduction.

Cambridge, MA: MIT Press, 2004.
[14] Arthur W. Toga, Paul M. Thompson. “Mapping brain asymmetry”. Nature Reviews
Neuroscience, 4(1), 2003.
Online
[15] The FIL Methods Group at UCL. “SMP8 Manual”. Institute of Neurology,
University College London. http://www.fil.ion.ucl.ac.uk/spm/.
Acknowledgements: The authors would like to thank the following colleagues for their contribution in
a way of another at various stages of this project. Zoran Josipovic (Cognitive Neurophysiology Lab, New
York University, USA) and Dan Lloyd (Trinity College, Hartford, USA) for their valuable assistance
with the fMRI scans and analysis. Duncan Williams and Aurelien Antoine (Plymouth University’s
Interdisciplinary Centre for Computer Music Research) for their help with the design of generative
algorithms. Philippe Esling (IRCAM) for granting us access to the ATO-MS code and helping us to
adapt the software to generate the fMRI-based orchestrations. Miranda is grateful to Ten Tors Orchestra
and its artistic director Simon Ible for performing the three compositions discussed in this chapter in
public concerts and festivals in Plymouth. Finally the authors are grateful for institutional and financial
support from Plymouth University’s School of Humanities and Performing Arts and EPSRC, in the UK,
and BEK and NOTAM in Norway.
192
Composing for the resonance:
Finding new relationships between
architecture and musical composition
Ambrose Field
This chapter documents my search for a new compositional method that could be in-
formed by the acoustic response of the venue in which a piece is to be performed.
Although composers have written pieces for particular spaces for centuries, this is tra-
ditionally a process informed by aural memory. I had two main aims: to make a tight
bond between score and acoustic result, and to be able to design pieces where the acoustic
contribution of a venue would be a known entity at composition time—rather than an
after-effect of a performance. Both of these factors create new musical possibilities as
they permit the space itself to become an integrated part of the composition. This body
of work would not have been possible without OpenMusic, and the role of the system
is explained with example patches demonstrating sound analysis, harmonic selection,
rhythmic generation, and texture manipulation.
I have always been attracted to the sound of spaces. The sparkling sheen of the Vienna
Konzerthaus; the short, muted slap-back effect you hear when singing outdoors in dense
forests; the huge, billowing acoustics of large Gothic European cathedrals; or the effects
of the weather on sound transmission in humid Asian cities in monsoon season. I wish
to access these acoustic effects directly as part of my music, as the accepted disjunction
between space and piece is a little odd—it is all sound after all that the audience will
hear, so why compose only half of the experience? The separation of reverberation in
the minds of composers, listeners, and performers from the musical material on the stave
itself is largely a Western phenomenon inherited from the late 18th century. It simply
does not exist in Eastern cultures where music historically was not performed in concert
halls. Performing in a large building, shopping mall, concert hall, or parking lot is an
entirely different acoustic experience from performing the same music in a small room
or domestic setting. So why do we expect one piece to be flexible enough to cope with
such a wide array of performance venues? My goal in researching this idea was to come
up with a compositional method that could help make pieces more specific to the venues
in which they are performed, regardless of what types of venue they might be. In order
to do this, it is necessary to take account of the acoustic at the pre-compositional stage
in the process of making a work.
193
Ambrose Field
Composers for centuries have written music for particular spaces. Just as it is hard to
imagine the early polyphony of Perotin being performed today in Notre Dame without
consideration for the acoustic, it is equally hard to imagine that Perotin would have
written his music without the acoustic effects of Notre Dame Cathedral in mind—for
this was a building with which he was intimately familiar. The link between performance
venue and composition space in this example from early music is significantly close: it is
not until the 19th century that this becomes fractured, to accommodate new marketing
propositions of touring orchestras and performers. From the 19th century on, the space
in which a musical work would be performed ceased to matter so much to composers—
particularly in central Europe, largely due to the construction of concert halls as public
entertainment venues. The space instead needed to serve the music, and concert halls
are today still judged on their clarity as much as their added resonance.1
This bond between acoustic resonance and composition has previously been negoti-
ated within the aural memory of the composer: few composers have precisely measured
the spaces in which their work is to be performed. So, I was prepared to sacrifice the idea
that a piece can be performed anywhere regardless of the acoustic; instead, I adopted an
approach where the composition becomes tuned to a particular space for these pieces.
This has allowed a much more detailed investigation of this idea.
OpenMusic offers an amazing opportunity: the combination of sound analysis with
the ability to feed that analysis back into a compositional strategy, gives rise to new
possibilities. It would be easy to see this method as a kind of parametric version of Alvin
Lucier’s I am sitting in a room: where an audible room analysis is fed-back into the piece,
as discussed in [1]. What I would like to accomplish poses a somewhat harder problem
than establishing a feedback loop: to try and gauge, or assess before performance, how
a room might respond to particular types of musical construction that at this point in
the composers’ working processes, have no realisation in sound. A further goal is to
understand how the response of the room could actually be composed for, enabling the
room itself to take an active part in the performance.
The role of OpenMusic

I am using OpenMusic on this project as part secretary and part scientific assistant.
My music does not follow rules or algorithms in its execution on all levels: I like to
work intuitively, selecting from and continuously refining what is essentially a palette of
different sounds (or the possibilities for sounds). I decided to craft a library of routines
that enable me to explore what I would otherwise do both with greater depth and within
a shorter time frame. The quantification of “what I would otherwise do” is perhaps the
difficult aspect, as it involves a sense of deep searching to quantify even the simplest
processes. The act of identifying a particular process does not mean though that the
music has to become process-led: OpenMusic is complex enough to accommodate
subtle nuances, flows, distributions, and uncertainties. I typically start composing by
imagining what a work might sound like and sketching that out, rather than with a more
abstract idea of process or structure. As a result, I have learned that not all parts of my
1 The effect of this relationship on the perception of sound localisation in contemporary music is discussed
in [3].
194
Composing for the resonance...
compositional approach can be applied to OpenMusic, and I have come to understand

that the first step is to decide which parts of the creative problem are best suited to it.
The examples in this chapter are based on a work for solo flute called Quantaform Series,
written in 2013,2 and a new work for 10-part choir Architexture 2.
Quantaform Series I-XX for solo flute and

acoustic resonances
Quantaform Series is a collection of highly unified, self-contained miniatures that can
be played in any order the performer requires. Each piece demands a different acous-
tic: from large resonant spaces to small, highly damped acoustics (or outdoors). The
flautist can move between acoustics, or perform only the movements from the series
that are suitable for a particular space. Quantaform Series, unlike my later work, is
not based on an analysis of the impulse response of the venue. Instead, it is based
on pre-existing acoustic information. The initial aim was to make a commercial studio
recording; so to provide accurate data to inform the compositional process, I consulted
the time/frequency response graphs of a well-known studio reverberation unit which was
available in the studio. This provided me with RT-60 data in a number of bands without
having to take impulse response measurements.3
Generating rhythm from acoustic data

in Quantaform Series
The work seeks a perceivable unity between the rhythmic flow of small-scale cells, sec-
tions, and larger formal units. Rhythm generation starts with the formulation of a
very short rhythmic fragment, which is itself calculated from the reverberation times
returned by the previous stage in the compositional workflow. Rather than function as
a cell for future “development”, I see this material very much as an “object”. It has
a certain outline—a described behaviour, which can be “instantiated” in a variety of
different forms. OpenMusic permits the use of a patch independently of any structural
associations—it is all just data—which lends itself ideally to this approach.
Figure 1 shows an OpenMusic patch generating rhythmic fragments starting from
the acoustic characteristics of a room (based on the reverberation times, RT-60, of
different frequency bands). The purpose of this patch is to generate a stack of possibilities,
which all occupy a predetermined length of time, yet have a number of different internal
behaviours and identities. These possibilities, or fragments, are related to the acoustic
data in a very direct way: the reverb times become durations from which rhythmic
material is built. First, an initial fragment is computed to set-up a graph containing
one point per rhythmic pulse (Figure 2 shows the contents of the make_first_fragment
sub-patch). The user can then edit this graph to time-stretch or compress the rhythm.
2 Quantaform Series performed by Jos Zwaanenburg, flute. Sargasso Records, London, SCD28071, 2013.
3 The RT-60 is a measurement of the reverberation time of a room. It indicates how long reverberation
takes to decay by 60 dB [2].
195
Ambrose Field
Figure 1. Generating a stack of rhythmic fragments based on specific reverb times. The sub-
patch make_first_fragment generates an initial rhythm. As the purpose of the omloop is to
generate further variations on this rhythm, I lock the graph of the fragment when I am happy
with the aural result.
A technique I have adopted whilst working in OpenMusic is the use of polyphonic

musical objects, such as the multi-seq object, as large-scale data storage containers to
hold multiple possibilities of a small phrase or musical unit. As one might ask a vocalist
or instrumentalist in a studio to lay-down multiple takes of a line in order to get closer to
the “perfect” recording, a key part of my work with OpenMusic lies in the stacking of
multiple “takes”, or subtle variations of the same compositional idea, from which choices
can be made. I will aurally select from these possibilities or plot trajectories between
them within another patch. I call this container a “poly-stack”—it is used only as a
resource rather than for playback.
196
Figure 2. Sub-patch make_first_fragment from Figure 1.
Within the inner omloop of the patch, a rhythm is created by the m-rhythm function
from the OMTimePack library, for reverb time using simple 0-order Markov probability
functions (see Figure 3). In The OM Composer’s Book 1 [6], Paul Nauert notes that
the simpler probability tables are in fact the ones which most often offer the greatest
compositional potential. In experimenting with the generative process for this piece, I
tried constraints-based methods using the OMRC library and although they would have
been well suited to generating a layered substructure, m-rhythm provided faster solutions
of equal quality.
Figure 3. Sub-patch inside the omloop from Figure 1 to generate rhythmic fragments.
197
Ambrose Field
The same patch was re-used in a variety of contexts in making this piece, from
forming local-level rhythm to providing high-level structure. It generates slightly different
rhythms each time, as each fragment is assembled from a random process, albeit one
subject to design through the use of a probability table. However, each rhythm output is
related to the other materials generated in that it is a product of the same process. The
rhythms generated are then folded into the piece using a series of structural time markers
derived from the initial rhythm generated (from Figure 2), creating another unified
musical bond between the contents of the fragments and their use in-time within the piece.
As a result of this folding process, and the rescaling of sections and structures, the audible
relationship between acoustic proportion and notation becomes progressively abstracted.
Thus Quantaform Series does not have an absolute one-to-one correspondence between
reverb time and notation: one has absolutely influenced the other, but by way of choice
and modification.
Checking the model, and helping reach an

informed performance
Being able to model and predict the acoustic of where a piece is to be performed at a pre-
compositional stage has a number of implications for the performers and the audience.
I also wanted to find a way to help performers approaching my pieces imagine the
consequences of performing in an appropriate acoustic environment. Of course, this
could be done through technological sound processing means, such as using a portable
computer, microphone, headphones, and an appropriate impulse response. However,
latency issues, the availability of technology, and the impact technology makes on practice
and rehearsal situations all need careful negotiation. I needed a simpler solution, where
performers could visualise how harmonies would become prolonged or blurred according
to their interaction with the acoustic. Writing a composition with the expectation that
it will be performed in a certain reverberant environment also has practical implications.
First, specifying the relationships between the score and the audible result of a piece
within the score itself is not without consequence for the performer. Any sensitive
musical performance involves finding the right tempi for a particular acoustic, avoiding
un-necessary blurring or dryness as appropriate. In Quantaform Series, the performer
must take care to accurately keep in time, and the responsibility of where rubato can
take place therefore is shifted to the composer. This occurs together with an increased
compositional responsibility for allocating what extents of acoustic blurring and dryness
are acceptable and appropriate to the material. The composer, however, gains another
compositional parameter.
Figure 4 is an excerpt of the score from Quantaform 16. This movement is designed
for a reverberation time of 0.5 to 0.8 seconds. By notating in 2/8, the performer has a
clear view of how real time interacts with performance time. The performer is also asked
to shape his or her attack profile for note onsets (indicated through a series of graphical
symbols above the staff) and vary the pitch-to-noise ratio of the air stream.
It is not simply tempo, as a raw and disconnected parameter, that performers ma-
nipulate when reacting to an acoustic. We also modify our articulation, accents, note
onsets, and timbres. In vocal music, once speed has been set, greater clarity can be
achieved through paying particular attention to consonant placement and note onsets.
So speed is only part of the equation. For example, in the recording of Quantaform 10,
198
Figure 4. Score extract: Quantaform 16 for solo flute and reverberant acoustics.
whilst keeping precisely to time, Jos Zwaanenburg provokes additional reverberation to

highlight key attack points by increasing air pressure on specific note onsets. As this
was a new technique, and one that may form part of my ongoing compositional practice,
I felt the need to try and validate what I was hearing in a little more depth. Again,
OpenMusic proved to be a worthy tool in this respect, due to the agnostic relationship
with different types of data that can be displayed and with which one can work.
The patch in Figure 5 computes a simulation of the reverberation effect produced by
a room by modifying the durations of an input chord sequence. The output is formatted
as a multi-seq split into three staves corresponding to different time scales. Figures 6
and 7 show respectively an input chord sequence and the resulting output.
From the input sequence (Figure 6), and without further analysis, it would be rea-
sonable to make the following assumptions:
• Because pitch content of the original phrase is not highly clustered, the effect of
the music on the reverberation might be relatively consistent across the texture.
• Some of the faster passages may need to be performed at a slightly slower tempo
for maximum clarity.
Instead, the patch reveals a more nuanced picture, showing that my initial assump-
tions about how a performance might need to be produced are, in fact, not totally correct.
From running this analysis, I gain the following additional information (see Figure 7):
• The acoustic has a specific resonance around the pitch F], and this pitch will
dominate what the audience hears. F] has effectively become the most important
harmonic axis.
• Some of the shorter duration material from the opening still remains relatively
transparent in this particular acoustic, as it does not become significantly blurred
by the reverb. This is good from a compositional perspective, as lower-register
flute does not have the penetrating effect that pitches in upper registers can have.
If these low pitches were becoming significantly blurred, for example, it would
perhaps indicate that this section needed rewriting to take better account of the
performance acoustic.
199
Ambrose Field
Figure 5. Patch analysing how an impulse response affects harmony. (The input and output
multi-seq are shown in Figures 6 and 7.)
Figure 6. First seven measures of Quantaform 16 as a chord sequence (chord-seq).
Figure 7. First seven measures of Quantaform 16 showing the prolongation effect the
reverberation has on the input material. The texture is spread onto three staves for clarity,
with the longest resonances on the lowest stave.
200
Architexture Series: analysing the acoustic to inform

the composition
The Architexture Series (2012-2015) is a set of large-scale vocal pieces for solo-voice
ensemble (i.e. with one voice to a part). For this compositional commission, I sought to
develop the ideas present in the Quantaform Series, but I wanted to be able to work in any
space and not just spaces for which I had access to known reverberation data. I visited
the venues for which the pieces were commissioned with a measurement and analysis
team.4 Our goal was to capture an accurate, three-dimensional map of the venue’s
acoustic. To do this, an impulse response of the performance venue has to be taken to
determine the resonant characteristics of the space. This was accomplished by recording
a sine-sweep within the venue using a spherical microphone array, and processing the
resulting audio file with deconvolution. The resulting impulse response file unusually
does not need to be used for any audio purpose in this project, as the aim is to offer data
to inform the creation of acoustic compositions. As in the Quantaform Series, I wanted
to find a means of extracting data with OpenMusic that could inform my harmonic and
rhythmic choices for a piece in such a way that there would be a perceivable relationship
between the compositional materials and the final sounding result.
The impulse response of the room is stored as four audio files according to first-order
ambisonics [4]: X (front-to-back information), W (omni-directional information), Y (side-
to-side information), and Z (height information). Only the W (omnidirectional) channel
is actually used in my patches. A partial tracking analysis of this audio file is performed
based on standard McAulay-Quatieri techniques [5] using Spear, and imported as an
SDIF file into OpenMusic, as shown in Figure 8 or 10.
Finding the RT-60 from the analysis data is a straightforward task: having set a
-60 dB analysis threshold in Spear, it is simply the length of the SDIF file in milliseconds.
The RT-60 proved helpful in understanding how rests, and long sustains, could be
deployed in the piece. In the Architexture Series, I used this information intuitively
within the pre-compositional process, outside of any particularly deterministic frame-
work. Other information about the acoustic characteristics were more systematically
exploited, such as the most frequently occurring resonances within short- or longer-term
time windows.
As reverberation decays, specific resonances in upper registers disappear; so looking
for the strongest resonances in the whole file generally returns largely low-frequency
information, which is not appropriate to my compositional purpose. The patch in
Figure 8 was used to extract the strongest resonances in short-term windows. In this
patch the function as-›om is used to read the SDIF file, and output a chord sequence
in a given specified pitch range. By setting 5 ms as <delta time>, the resulting data is
windowed accordingly for a short-term time scale.5 In calculate occurrences each pitch
is then counted and given a score of how frequently it occurs. In select n most frequent,
the list of pitches and numbers of occurrences are sorted so that the most frequent come
at the top, and I choose to include only n of those in the final output. Data is held in a
4 From the Department of Electronics at the University of York, led by my colleague Dr. Jude Brereton.
5 The <delta-time> parameter in as-›om specifies a window during which all notes will be gathered and
considered as part of the same chord.
201
Ambrose Field
Figure 8. Extracting the pitch of resonances which occur on a short-term time scale from an
impulse response.
text-file box as calculate occurrences can take a while to compute and the data may be
required in other patches (and then simply copied from this text file). Figure 9 shows a
detail of the calculate occurrences sub-patch.
By extracting information about the short-term resonant profile of a space, valuable
information is generated influencing how an acoustic might instantly respond to, and
help to inform, my choice of harmonic colours appropriately. As these harmonies are not
necessarily based on the same pitch centres as those derived from the longer-term analysis
(see Figure 10), they provide a means of informing the pre-compositional planning for
the moment-to-moment evolution of the piece.
Having ascertained which pitches occur most frequently on a very short-term timescale
in the acoustic, I then also wanted to obtain information that might match a general,
longer-term perception of the space. The patch in Figure 10 analyses the impulse response
to find the resonances with the longest durations, as opposed to the most frequently
202
occurring in a particular time-window as calculated in the previous patch. By writing

music involving these pitches, I immediately knew they would provoke a more spectacular
response from the acoustic, and that this information could be used to inform a sense of
dialogue between the musical score and the acoustic space.
Figure 9. Detail of the calculate occurrences sub-patch in Figure 8.
Figure 10. Patch to find which pitches occur over longer time scales. Note the longer analysis
window of 500 ms, and also the display of only the pitches with the n longest durations.
203
Ambrose Field
Conclusion
OpenMusic has played an evolving part in my compositional practice. It is perhaps at
its most helpful to me in presenting a wide set of possibilities, rather than suggesting
any one specific solution. Whilst many of the methods I have used could be extended
through stricter, algorithmic approaches, my working practice instead foregrounds human
intervention and a questioning of computed output. In short, I need choice.
Singers working on my piece Architexture 1 commented that they felt liberated: the
acoustic clarity that was generated through appropriately spacing out the notes through
time was very welcome, allowing them to concentrate on articulation and diction. Open-
Music has permitted me to develop a new compositional approach: the investigation of
what the most effective compositional strategies might be for particular spaces based on
measured, architectural acoustics. It has played a significant part in making these pieces
a reality, and in helping performers approach the material with clarity.
References
[1] Densil Cabrera. “Acoustical, psychoacoustical and subjective assessment of Alvin
Lucier’s I am sitting in a room”. In Australasian Computer Music Association
Conference. Sydney, 2001.
[2] F. Alton Everest, Ken C. Pohlmann. Master Handbook of Acoustics. New York:
McGraw-Hill, 5th edition, 2009.
[3] Jean-Pascal Jullien, Olivier Warusfel. “Technologies et perception auditive de
l’espace”. In Les Cahiers de I’IRCAM, 5. Espaces. IRCAM-Centre Pompidou, 1994.
[4] David G. Malham, Anthony Myatt. “3-D Sound Spatialization using Ambisonic
Techniques”. Computer Music Journal, 19(4), 1995.
[5] Robert J. McAulay, Thomas F. Quatieri. “Speech analysis/synthesis based on a
sinusoidal representation”. IEEE Transactions on Acoustics, Speech, and Signal
Processing, 34(4), 1986.
[6] Paul Nauert. “Algorithmic Strategies in A collection of Caprices”. In Carlos Agon,

Gérard Assayag, Jean Bresson (eds.) The OM Composer’s Book 1. Editions Delatour
France/IRCAM-Centre Pompidou, 2006.
204
Sketching, synthesis, spatialisation:
The integrated workflow of Cognitive
Consonance
Christopher Trapani
Composed in 2009-2010, Cognitive Consonance is a single piece divided into two: two
plucked-string soloists from two disparate worlds, two different sets of tools for composing
with electronics, and two different approaches to the practicality of writing non-tempered
ensemble music.
This article outlines the role of computer-aided composition in the first part of
Cognitive Consonance, an 11-minute movement entitled “Disorientation”, written for an
unusual solo instrument with extreme limitations: a custom-designed microtonal qanûn.
Three distinct stages of the process can be singled out. First, the pre-compositional
phase, where OpenMusic served as a practical tool for organising microtonal pitch
content. Second, the synthesis stage, where physical models of plucked strings were
created via a customised Modalys patch, alongside the batch retuning of sample libraries
(using the phase vocoder SuperVP). As a final step, the OMPrisma library was used
to generate polished eight-channel sound files combining samples and synthesis, ready to
be exported to the performance patch.
Sketching
The qanûn (kanun in Turkish orthography) is a trapezoidal zither, common throughout
North Africa, Turkey, and the Middle East. It is capable of being retuned while played
thanks to a series of short levers underneath the strings known as mandals, which can
be raised or lowered to change the length of the string.
The layout of these mandals varies by region, with multiple degrees of microtonality:
chromatic semitones in Armenia, tempered quarter-tones in North Africa, and a finer
gradation in Turkey, where most models divide the semitone into six parts. Pictures of
the instrument are included as Figures 1 and 2.
“Disorientation” was composed for a specific instrument, designed by the late French-
Swiss qanûn player Julien Jalâl Eddine Weiss. Rather than equidistant mandals, Julien
has created an intricate system of microtonal tuning with the goal of representing a
maximal number of just intonation ratios. Strings have as many as fourteen mandals,
for a total or fifteen possible microtonal accidentals on a given string (notated with a
personal series of symbols derived from Western, Arabic, Turkish, and Persian prac-
tice), representing the just intonation ratios outlined in Julien’s handwritten schema in
Figure 3.
205
Christopher Trapani
Figure 1. Julien Weiss’s custom qanûn.
Figure 2. A close-up look at the mandals.
206
Sketching, synthesis, spatialisation: The integrated workflow of Cognitive Consonance
Figure 3. Julien Weiss’s just intonation tuning diagram.
Figure 4 shows the eight gradations of pitch available between a C and C], using
the intervals in cents taken from the lower right-hand corner of Julien’s diagram. The
challenge of tracking the combinatorial possibilities of this detailed pitch network fit
naturally with OpenMusic modelling at the pre-compositional stage. It was necessary,
first, to gain a comprehensive idea of the practical applications of this theory—to explore
which pitches were actually available on the instrument—and second, to have a means
of comparing Julien’s approximations to independently derived pitch content to better
understand how closely his microtones were capable of approximating given passages.
Figure 4. Microtonal gradations between C and C].
207
Christopher Trapani
As a first step I constructed a patch (jjweiss-kanun, see Figure 5) capable of filtering

pitch material and providing the closest matches on Julien’s qanûn. Given a list of
midicents, the patch returns the midicent value, string, and number of mandals (from -7
to 7) that correspond to the nearest available microtone.
Figure 5. The closest matches on the qanûn to any incoming pitches.
The results are sometimes surprising: we learn that the closest available pitch to a
tempered middle C is actually a B + 6 mandals, and not the qanûn’s C, which would be
two cents lower (thanks to the instrument’s Pythagorean tuning). Figure 6 gives a look
at the loops and operations inside this patch.
Figure 6. Inside the jjweiss-kanun patch.
208
The next step was to compare the pitches on Julien’s instrument to an ideal harmonic
spectrum. As shown in Figure 7 we can visualise the deviations in cents within a bpf,
discovering that the largest divergence between the instrument and the first 25 pitches of
the harmonic series falls on the 17th partial, which as a C] on Julien’s qanûn is 10 cents
higher than a tempered C]. (Compare this to the difference between a just major third
and a tempered major third, an approximation central to nearly all tonal music, at 14
cents.)
Figure 7. A comparison of the Weiss tuning system to a harmonic series.
This tempered C] on the ensemble instruments (guitar, mandolin, harp, vibraphone)

becomes the focal point of the opening of the piece. A central component of “Disorienta-
tion” involves creating microtonal modes derived from just intervals—a practice similar
to that of ancient scholars of Islamic music, but extended into higher partials.
Figure 8 outlines a symmetrical mode around C], with the 17th through 25th partials
and just inversions—a close fit to Julian’s qanûn, which notably mirrors its ascending
and descending intervals (though occasionally mandals which would not be used in any
traditional music context, such as the A quarter-sharp here, were omitted).
Figure 8. A mode derived from high odd partials.
209
Christopher Trapani
Finally, Figure 9 shows a score excerpt with the first appearance of this material,
underpinned by the low C fundamental in the harp and contrabass.
3
" " 3 . , 0(,-,0! , . *

SOLO until rehearsal mark S
B
0! (,0, * (,0, . *
20
! "*
(in the style of a Taksim improvisation) near bridge
#
+0, / 0,
Qanun $ " 6
,- , ,- ,
+ + % % % +
3
6 " - "
(, , ,
)/
p ppp , pp f
3
mf pp f
Elec. $ "" !
"
"
"
" " ! "

Mand. $" % % " % % % % % " % %
0 , , ', , , 0 ', , / .
,, ,, 2
0/ , (,
" ! + - "" ,
#$ "
Gtr. % % " % % % . + . .
)((
!
3 ppp sfz mf
pp p
/
+0,&, 0/ . &/ . 0/ . &/ . 0/ . &/ . 0/ . &/ .
(put chopstick down)
" * % ! " 0/ &/ . 0, &,
"* * *
3
$" "
sfz mp pppp p pppp pppp
Hp.
)" ! "
pp hit with chopstick
" * % " % % % % % " % %
/
* )/
( VIBRAPHONE
$ "" ! " , 0,
with bow
Perc. % * $ 0/ " /. /. , 0/ /. /. " / 8
con Ped. n
" " 3. . 3. 3. 3.
pp p
3 1
!3 3
(irregular tremolando)
Vln. $" ! ! "! ! ! ! ! . "" % %

mp pp n
)" 1
"1 3. . "! 3. " 33 3 1
3. 1
% % % . 3 " 33 3
Vc.
3
n n mp
col legno battuto col legno battuto
punta d'arco 7 pizz. punta d'arco
ricochet l. v.
)" . ! ",
#+ "
ricochet
Cb. + 6 . % " % % % % % "' + 6 . %
, ,,,,,,,,
pp p , ,,,,,,,,
pppp pp pppp
Figure 9. Score excerpt, page 3 of Cognitive Consonance.
Synthesis
In generating the electronic material for “Disorientation”, physical modelling synthesis
with Modalys also played a large role. The goal was to create a “virtual qanûn”, whose
capabilities complement and exceed the actual instrument, by using vibrato, glissandi,
and by pushing beyond the capabilities of a human performer. The “mono-string” and
“pluck” models seemed to mimic the freely-resonating triple strings of the qanûn. This
synthesis also was controlled using OpenMusic, via the Modalys library, with precise
microtonal pitch information and onsets described within chord-seq objects.
The Modalys sub-patch itself, the result of extensive experimentation (and a bit too
large and labyrinthine to be included here) consists of three embedded patches. At the
lowest level, there is the Modalys synthesis patch that creates, for each midicent value
in the chord-seq, two or three strings (depending on the register, as on the actual qanûn
or indeed the modern piano), which may also be attached to a resonating membrane.
To these synthesised strings a pitch-variance control applies a random variation in cents
within a given threshold, recreating the natural variance in acoustic tuning that gives a
slight “chorus” effect to doubled strings. The distance between the virtual strings can
also be controlled, as well as the height and velocity of the attack, lending a further sense
of tangible physicality to the final synthesis.
210
The intermediate sub-patch, shown in Figure 10, is a loop that creates, for each
midicent pitch in the chord-seq being synthesised, breakpoint functions capable of piloting
Modalys. Onset data is translated into bpfs that describe the up-and-down motion of
a Modalys bi-two-mass object (mimicking the plectrum). Access points (the position
where the plectrum touches the string) and listening points (the position on the string
whose vibrations are calculated) are both programmed to change dynamically around a
given threshold for a bit of automated timbral variety.
Figure 10. Intermediate level of Modalys patch.
The top layer of the Modalys sub-patch, the control panel (see Figure 11), contains
all of the relevant data for the later synthesis. A key object is chord-seq-spread, which
reorganises the chord-seq’s data according to pitch, listing all of the onsets, durations
(irrelevant to the freely-ringing strings here, which are given a fade-out after the final
attack of the length given to the last-onset-plus-dur sub-patch), and velocities associated
with each midicent value in the chord-seq. Each virtual string (or, properly speaking, each
set of two or three strings, as explained above) can thus be re-attacked, with the physical
model of the vibrating string taking resonance into account. After much experimentation,
a range of scaled parameters were linked to velocity, each contributing toward a more
realistic synthesis: the height from which a virtual plectrum attacks the string, the drop
time, the stiffness of the plectrum, and the force of the pluck in Newtons. The upper and
lower limits of each of these synthesis parameters is given in the control panel. There
is also an option to make every note in the chord-seq sound as an octave harmonic, by
calculating strings an octave lower and placing a bi-two-mass object at their midpoint for
211
Christopher Trapani
a harmonic effect. Finally, an attached membrane is implemented to colour the resonance

of the string. It can be turned on and off, panned, mixed, or have its connection point
to the string changed. The outcome of this synthesis process is a list of mono sound files
corresponding to all unique midicent values found in the chord-seq.
Figure 11. Top level of Modalys patch, control panel.
In order to hear in a musically meaningful way the unfamiliar microtonal modes

mentioned in the first section of this article, I made use of Modalys synthesis not only to
produce polished sound files, but throughout the composition of the piece. I constructed
small phrases (as in Figure 12) which use a pre-established contour and durations (not
pictured here) that can be retuned to given modes and synthesised in a short mono file,
providing a sense of the colour that would result from each precise tuning.
Later, in the interlude that bridges the two movements of Cognitive Consonance,
Modalys synthesis is pushed further to create a range of experimental plucked sounds,
some based on actual stringed instruments (an autoharp, Harry Partch’s harmonic canon,
the Ethiopian begena), others with exaggerated timbres such as a wide, wobbling vibrato
or a deliberately heavy resonator connection that muffles the resonating string.
212
Figure 12. A phrase synthesised with Modalys.
Samples and batch processing

Another important tool for creating the electronic part of “Disorientation” involved the
batch-processed transposition of instrumental samples, using the OM-SuperVP library
and a collection of custom modular patches and abstractions. The process resembles
an offline version of concatenative synthesis, specifically “targeted transposition” later
implemented with the package CataRT [1]. In preparation, a sub-patch containing a
sample library must be assembled. Each sample must be associated with a midicent
value, as in Figure 13.
Figure 14 shows a typical chain process and the interaction of the batch processing
tools with an independently generated chord-seq object. First, four sample libraries (in
the patches at the upper-right part of the figure) are combined with the add-multiple-libs
abstraction. A second abstraction called find-closest-sample-in-lib takes each note in the
chord-seq and searches the sample library (or a collection of libraries: add-multiple-libs
chooses at random when multiple matches are proposed) for the closest pitch, outputting
the mono sound files on the right and their midicent values on the left. The next patch
in the chain, transpose-to-target-pitch, calculates the difference between the original
sample’s midicent value and the target pitch value; this difference is then sent as the
transposition value to the SuperVP phase vocoder via the OM-SuperVP library.
The abstraction vel-equals-vol takes the velocity values of the chord-seq and scales
them as gain values (this step can also be omitted from the chain for a completely flat
mix). In the final step, a patch called stereo-randompan-save-edels creates a stereo file
from a list of mono sound files, applying a random pan value to each and inserting a
length of silence that corresponds to the onset value in the chord-seq before each sample.
The mechanisms and vocabulary of the tools therein are the subject of the next section.
213
Christopher Trapani
Figure 13. A sample library sub-patch.
Figure 14. Batch processing of samples according to chord-seq data.
214
Spatialisation
The final sound files for “Disorientation” are in eight channels, in a circular configuration
around the audience. The generation of these eight-channel files was integrated into
OpenMusic as well, via a set of custom abstractions built around the Csound spatial-
isation library OMPrisma. The examples that follow focus on Modalys synthesis,
but OMPrisma was also used extensively to spatialise the batch-processed samples
mentioned above.
OMPrisma comprises a set of tools for
creating multichannel sound files accord-
ing to several different spatialisation algo-
rithms (including VBAP, DBAP, and Am-
bisonics) and types of movement (discrete,
continuous, trajectories) capable of adapt-
ing to customised speaker configurations.
Taking a list of mono sound files as an
input, an OMPrisma class object (named
for the type of spatialisation to be applied)
calculates a Csound score according to a
series of matrix data. Parameters can be
added as keywords: durations, entrance
delays, gain, x- and y-positions, x- and y-
envelopes, among others. Each parameter
can be input as a list or as a single value
that will be applied to all columns of the
matrix. The object can also be opened in
an editor for a look at all of the matrix
values, and the possibility of editing these
by hand; Figure 15 shows an editor view of
a VBAP.discrete object containing values
corresponding to seven sound files. Figure 15. Matrix data inside of an OM-
Figure 16 displays a typical musical ap- Prisma VBAP object.
plication of a discrete OMPrisma process.
The chord-seq contains an algorithmically generated passage which will be used to pilot
Modalys synthesis, where each midicent value will be synthesised as a separate string.
The angle-panning-vbap abstraction then acts as an intermediate step between the list of
sound files and the OMPrisma objects. Its simplified controls calculate a list of positions
along a given arc (described in degrees) so that each sound file is given a unique position
in space for a truly vivid synthesis. These positions can also be randomised by toggling
the value 0 or 1. In Figure 16 a light ring of high strings positioned in the back of the
hall (95 to 265 degrees) can be seen in the eight-channel waveform.
Figure 17 provides a look inside a slightly expanded version of the abstraction which
also applies a random distance value to each of the sound files. Note the bpc object
(“speaker setup”) connected to the vbap-setup function near the end of the synthesis
chain; this box contains speaker position data, and can be customised to any configura-
tion.
215
Christopher Trapani
Figure 16. An integrated process: chord-seq, Modalys synthesis, and spatialisation.
Figure 17. Inside the VBAP discrete spatialisation patch.
216
A second example in Figure 18 provides an illustration of the dynamic trajectory

function of OMPrisma as well as a look at the algorithmic generation of a musical
passage. The starting point for the passage is the interval of a just major ninth from C4
(5994 midicents in Pythagorean tuning) to D5. Starting with two pitches, the sums and
differences of all frequencies in one chord are used to create the subsequent chord, in a
cumulative process that results in seven increasingly dense chords. Breakpoint functions
are sampled to generate onsets in an accelerando pattern (with a random addition of 1-17
milliseconds to the smooth curve for a more jagged rhythmic rendering) and velocities
for a global diminuendo.
Figure 18. From algorithmically generated passage to spatialised eight-channel sound file.
Rather than a discrete position, each synthesised string is here given its own trajec-
tory. This is accomplished by using a VBAP.continuous object along with a set of tools
for manipulating trajectories, as displayed in Figure 19. An initial trajectory, described
in a bpc object, is reflected using the traj-mirror function. Traj-interpol then calculates
a number of trajectories (corresponding to the length of the incoming list of sound files)
between these two extremes. The result is a smooth movement over the length of the
sound file from a single point, front and centre, to a splayed arrangement at the back of
the hall, where each virtual string again has a unique position.
217
Christopher Trapani
Figure 19. Interpolating trajectories in an OMPrisma process.
References
[1] Aaron Einbond, Christopher Trapani, Diemo Schwarz. “Precise Pitch Control in Real
Time Corpus-Based Concatenative Synthesis”. In Proceedings of the International
Computer Music Conference. Ljubljana, 2012.
Acknowledgements: Cognitive Consonance was composed during my second year participating in

the IRCAM Cursus for Composition and Music Technology and is deeply indebted to all researchers,
developers, and musicians at IRCAM who contributed their time and expertise: notably Nicholas Ellis,
Marlon Schumacher, Jean Lochard, Jean Bresson, René Caussé, and Yan Maresz, as well as to Cyril
Béros and especially Eric Daubresse. The work was premiered at the Festival Agora on June 9, 2010 at
the Centquatre in Paris, with Julien Jalâl Eddine Weiss on qanûn, Christelle Séry on hexaphonic electric
guitar, and Ensemble L’Itinéraire conducted by Mark Foster.
218
Germination
Jean-Luc Hervé and Serge Lemouton
Germination was created for IRCAM’s Manifeste Festival in 2013. The work originates
from the architectural characteristics of IRCAM itself and its location in Paris. The
research and production activities of the institute are located underground beneath place
Igor Stravinsky, which is traversed daily by thousands of people who generally are not
aware of their existence. The idea of Germination was to make emerge, on the surface
of the pavement, a work of music performed in the concert hall beneath.
The work is in two parts: an ensemble piece for 13 musicians and electronics performed
in the concert hall is followed on place Igor Stravinsky by an electroacoustic piece diffused
by 50 MP3 players, each connected to a miniature loudspeaker (see Figure 1-b). After
the first part, the audience is invited to emerge from the building and to listen to the
second part of the piece outside. This path, from below the ground up into the air, is
reminiscent of a plant sprouting in two phases: the development of a seed in the ground
followed by the growth of the plant on the surface. A vegetal installation emphasised this
idea of plant germination, and allowed better integrating of the diffusion system on site
(see Figure 1-a). The electroacoustic piece was performed on three successive evenings,
each time in a different version: starting from instrumental sounds the first night to more
abstract and noisy sounds the last.
Figure 1. Left: Germination vegetal installation on place Igor Stravinsky. Right: MP3 player
and its loudspeaker on-site. Photographs
c Astrid Verspieren and Serge Lemouton.
219
Outline of the project
The organic idea of vegetal growth was the compositional model, both for the instrumen-
tal and electroacoustic parts. OpenMusic was used as a computer-aided composition
tool for the entire piece; however in this text we will focus on the creation of the second,
electroacoustic part, which was played on the 50 MP3 players. Every “agent” (MP3
player and speaker) played a different sound file, and the whole installation produced
an enormous polyphony where each part had its own rhythmic, harmonic, and timbral
specificity.1 Generating such a large number of sound files, one for each voice but all
derived from the same model, was a difficult task to perform manually. OpenMusic
allowed easy iterative implementation through “batch” processing, turning out to be a
perfect tool to achieve this goal.
Each sound file or voice is made up of several successive sections, organised according
to a process of transformation of sound material in the domains of pitch, timbre, and
rhythm. Every section is developed in 50 variants, corresponding to the 50 voices. These
variants have slightly different durations, so that the transition from one section to the
other is usually asynchronous among the different voices. However, some of the sections
(e.g. 1, 2, and 5 as shown in Figure 2) start at the exact same time in all tracks: in this
case silences are inserted between the sections in order to synchronise the next starting
point.
Figure 2. The 50 voices are divided into sections of varying lengths. A number of
synchronisation points (bold arrows in this graph) group the beginnings of specific sections.
The electroacoustic part of Germination was composed in several steps (see Figure 3).
Sound files were selected from a database following specific criteria and then processed
using OM-SuperVP in order to multiply and vary the sounds in the corpus. They were
then ordered according to a process of evolution of sonic parameters, and put together
rhythmically in order to generate the 50 variants of each section. Then the sections were
concatenated in an OpenMusic maquette to produce the final 50 tracks.
1 Like
a population of frogs living in a pond, which all belong to the same species and emit the same
sound, but each at a different pitch and rhythm.
220
Germination
Figure 3. Workflow and software used for the electroacoustic part of Germination.
Constitution of the sound corpus
A set of sounds from the corpus is selected according to a given criterion (instrument,
pitch, playing mode, etc.) and enriched by applying the same modifications to all sounds
in the set. For each section of the piece, a different sound transformation is applied.
In the example shown in Figure 4, a section of “sliding” sounds is processed with pitch
and time modifications.2 A set of similar sounds (for instance, sustained string sounds)
is processed in a loop that produces for each sound n variants (10 in this case), selecting
random time-stretch and transposition values within predetermined ranges. The sounds
are processed one-by-one within the loop trsp-ts-loop2, and each time 10 variants are
produced by the loop trsp-ts-loop3. The contents of trsp-ts-loop3 are visible in Figure 5.
The sub-patch make_glis_alea selects random start and end values within a specified
pitch range for the dynamic transposition process (generating artificial glissandi).
Figure 4. Nested loops implementing batch processing of sound files.
2 The treatments are performed using the OM-SuperVP library.
221
Figure 5. Sound treatments using OM-SuperVP. The input sound (<son>) is processed n
times (<nb calculs>) by dynamic transposition and time stretching.
Classification of sounds using CataRT

We have used CataRT software to process and sort our generated sound corpus: not
for its audio segmentation or synthesis features, but for the possibilities afforded by this
software for analysing and classifying such a corpus in a multi-dimensional space. The
files generated by CataRT were loaded in the OpenMusic patch shown in Figure 6.
In this patch, the sub-patch catart-filenames routes specific files containing the list of
descriptor values, the corresponding descriptor names, and sound file names to separate
textfile boxes. The descriptor names are instantiated in the selection list shown at the
left. Depending on the selected item (in this example duration), a given descriptor index
is identified and the corresponding values can be used as a reference to sort the list of
sounds. The output of this patch, stored in the textfile box at the bottom of the figure,
is a list of sound filenames ordered according to the chosen descriptor (with the duration
descriptor, for example, this means from the shortest to the longest one).
222
Germination
Figure 6. Ordering a set of sound files according to descriptor values computed by CataRT.
Sound montage with rhythmic processes

The ordered list of sounds is then concatenated according to a rhythm defined in a
specific patch for each section. In Figure 7, 50 versions of the concatenation process
are executed, corresponding to the 50 tracks of a given section of the piece. Each time,
a random group of successive elements is chosen in the ordered list of sounds (with
starting index between 1 and 41, and ending between 101 and 141). The selected sounds
are collected and positioned in a maquette. With each iteration this maquette is then
mixed down into a sound file using the maquette2sound function.
In this example the sounds are positioned in the maquette (that is, in time) following
a curve of exponential acceleration. This process is contained in the algo_rtm1 sub-patch
and detailed in Figure 8. The first duration varies around 22 seconds with a standard
deviation of 2 seconds, and the last duration around 4 seconds with a standard deviation
of 1 second.
223
Figure 7. 50 iterations of a sound concatenation process.
Figure 8. Rhythmic process to determine of the time-positioning of sound files. Left: random
variation of the value range using a Gaussian distribution. Right: internal patch to compute
the exponential curve.
224
Germination
Creation of the sound tracks

As shown previously, each voice is made of a succession of sections (sound files) of varying
durations. As this duration varies among the voices, the sections are generally not
synchronised. Some specific sections, however, are forced to start at synchronisation
points called pivots.
Figure 9 shows the main patch to generate the sound tracks. The pivots are each
specified by their position (section index) and the exact time (in seconds) where the
corresponding sections must start (there can be up to three pivots in total in the work).
Figure 10 shows the contents of the omloop makemaquette, which generates a list of
maquettes each corresponding to one of the 50 tracks from the sound directory and
this set of synchronisation data.
Figure 9. Main patch to generate the maquettes and sound files corresponding to the 50
tracks of the piece.
With each evaluation of the savesounds omloop, sound files are generated from each
maquette to be loaded on the MP3 players. The full set of maquettes can also be
collected in an other maquette, the maquette of maquettes (see Figure 11).
225
Figure 10. Inside the maquette generation process. The arithmetics computes and adjusts
the onsets of the different sounds (sections) according to the specified pivots.
226
Germination
Figure 11. The maquette of maquettes. Each rectangle represents a track: a maquette
containing its own sequence of sections, as shown in the open window.
Conclusion
Sound processing software is generally designed and adapted to commercial musical
practice. OpenMusic, on the contrary, is not primarily dedicated to sound processing
and therefore presents a number of particularities and constraints in this domain. But
constraints for a composer often present opportunities to imagine other ways forward; to
explore alternatives and finally open new horizons.
In Germination, OpenMusic permitted the development of an algorithmic approach
to sound processing and editing, where the composer defines a prototype actualised
each time in a different way thanks to aleatoric processes. This approach is perfectly
suited to an organic notion of music, taking living species as a model, where the genetic
code defining the species is transcribed in each individual of a population with specific
variations. The same musical sequence can be repeated, yet always modified. This
characteristic allowed us in Germination to compose hyper-heterophonies: where each
voice realises the same sequence differently, reminiscent of the plastic versatility of living
things.
227
Dialogue with OpenMusic in the
process of composing Nothing that is
not there and the Nothing that is
Takéshi Tsuchiya
In Nothing that is not there and the Nothing that is for violoncello and electronics1
OpenMusic was used at several stages of the compositional process: in particular in the
construction of the overall structure, in the pitch structure of the violoncello part, and in
various aspects of sound design for the electroacoustic part. In this paper, I will explain
some important details of these different aspects.
Introduction
Nothing that is not there and the Nothing that is reflects the underlying theme of a
philosophical hypothesis, a kind of “simulation hypothesis” [1]. I also feel that this
theme is close to the Zen philosophy that is part of the traditional culture of Japan.
When I compose a work accompanied by electronic techniques, my first consideration
is the relationship of the instrument and the electroacoustic sound. In Nothing that is not
there and the Nothing that is, this relationship is not only that of a mere accompaniment.
The electronics are internalised as an extension of the violoncello. This extension is not
only timbral: it has various repercussions for the structure as well. But for this, it is
necessary to reconsider traditional musical parameters such as pitch, rhythm, dynamics,
tempo, and extend them to become new structures and transformations of timbre at the
micro-level of sound synthesis and processing.
Since I began to think about and write music with the use of computer-aided com-
position, OpenMusic has become one of the most significant tools among the variety of
methods involved in my compositional process. It is of course applied differently in each
work, but one important common point is structure-building and its coordination with
sketches written by hand.
The role of handwriting in my compositions can be roughly divided into two kinds
of thought processes. One of them is the idea of the existence of a whole structure.
I illustrate this with the calligraphy shown in Figure 1, which had an important role
in inspiring Nothing that is not there and the Nothing that is. At first glance this
1 Written for Ensemble Contemporary α and premiered at the 20th Seoul International Computer Music
Festival in 2013.
229
Takéshi Tsuchiya
calligraphy may look like “avant-garde” drawing, but it actually represents traditional
Kanji characters. These characters are written with Japanese black ink called sumi.
Their meaning is “meditation”.
Figure 1. Calligraphy by Kakku (left) and its radicals (right).
The drawing has a time dimension (the order and speed of the successive paintbrush
strokes), just like a musical performance. It can be seen as an expression/interpretation
the Kanji characters. It also creates a space that is uninterrupted. At the right in
Figure 1 is the “block body” of the calligraphy, with numbers indicating the order of the
strokes. This order is important: it determines the busyu (or “radicals”) that constitute
the internal structure of the Kanji. The radicals have meanings on their own, and bring
up new meanings when they are combined (see Figure 2). This specific structure of Kanji
can be said to be a characteristic of thought in Asia and particularly in Japan.
Figure 2. Detail of busyu (radicals) and combination of ideographic characters.
I was inspired by this calligraphic structure to construct Nothing is not there and the
Nothing that is. However its reflection in the piece is quite different from the technical
approach seen in Impressionistic music: instead I have aimed to represent the concept
of time in Japanese or Asian cultures, as well as the relationship between the whole
structure and its parts.
Ryo Yanagi2 explains Japanese structural beauty using the Western concept of the
“golden section” [2]. For instance, he uses this ratio to describe the relationship of whole
and parts of the structure of the Tea Room. Although I have not used his specific
2 Japanese art critic (1903-1978).
230
...Nothing that is not there and the Nothing that is
methodology, I was sympathetic to this way of streamlining the approach to structural

beauty in traditional Japanese culture. In particular, formalisation using the computer
allowed me to model the internal time dimension from the structure of calligraphy.
A second important role of handwriting in my composition is through the concept
of sketches, first written by hand in graphics and characters and then developed further
in the computer-aided compositional processes. Figure 3 shows one of these sketches,
expressing some aspects of form and compositional materials, including the pitch and
rhythmic structures that will be developed in OpenMusic in the following sections.
Figure 3. Sketch of Section IX of Nothing that is not there and the Nothing that is.
Construction
The overall structure of this work is implemented in the maquette shown in Figure 4.
The whole can be divided into twelve sections (I-XII). These twelve sections are aggre-
gated into five groups (or “scenes”), each colour-coded within the maquette and marked
-
1 .
5 Figure 5 shows the patch generating this maquette.
Figure 4. maquette of the compositional structure.
231
Takéshi Tsuchiya
Figure 5. Patch to construct the maquette.
The groups -
1 5 have the following features:

1 (Section I, XII) [Static] Violoncello part played on the bridge and on the tailpiece.
No clear pitch and very low volume level.

2 (Section II, VII) [Kinetic] Various glissandi over a wide pitch range of the violon-
cello.

3 (Section III, VI, VIII) [Static] Symmetrical with scene .
2 Uses violoncello noise
material.

4 (Section IV, V) [Kinetic] Complete contrast with the other four. Focus on the
violoncello (normal and extended playing techniques). Electronics consist of spatial
processing, remaining in the background.

5 (Section IX, X, XI) [Kinetic] Uses various techniques of interpolation. Comprises a
process staring from the lowest tone and reaching the highest tone of the violoncello.
Figure 6 details the internal constitution of the maquette for Section II. Its contents
are four temporal objects (patches inside the maquette). II-A1 is the principal object
determining the structure of the section. It is connected functionally with the other
objects II-A2, II-A3 and II-B. Figure 7 shows the contents of the patch in II-A1: the
pitch profile is filtered using the posn-match function, and then transmitted to the other
objects.
Such processes using OpenMusic (and particularly the maquette) allow the con-
struction of substructures with non-periodic and unpredictable timings. The conception
of these substructures is “narrative” and does not rely on traditional compositional logic.
232
Figure 6. Internal constitution of the maquette of Section II.
Figure 7. Inside the temporal box patch II-A1 from Figure 6.
233
Takéshi Tsuchiya
Sound design for the electroacoustic part

The electroacoustic part is composed of violoncello sounds processed offline, triggered and
controlled during the performance using Max. At the beginning of Section I (Figure 8),
in the space between the barely audible violoncello part and silence, a sound texture
is created by intertwining the cello and electroacoustic sounds in real time. Then the
music begins again, somehow continuing the tension generated by the silence. The sound
of the violoncello is extended with playing techniques such as col legno battuto and a
fine pianissimo created by bowing the tail-piece of the instrument. An important detail
is the presence of subtle amplification, for sounds difficult to hear in normal acoustics.
This amplification, as well as additional spatial processing, are performed in real time.
The electroacoustic part was mostly created with the OM-SuperVP library using sounds
performed by the violoncello. The main patch to generate the sounds is shown in Figure 9.
Figure 8. Opening of Nothing that is not there and the Nothing that is.
Figure 9. The main patch to generate the electroacoustic part.
234
The process begins with the analysis of a violoncello recording using the transient-
detection tool from the OM-superVP library. This analysis generates approximately
a dozen temporal markers that define segments of the original waveform. The sound is
then processed using four different combinations of time stretching or pitch transposition,
also using OM-SuperVP. Finally the original sound and the four processed sounds are
concatenated into a single sound file.
Figure 10 shows a detailed view of one of these processes, corresponding to one of
the four sub-patches in Figure 9. The OMAlea library is used to determine stretching
factors for supervp-timestretch using pseudorandom number generation (zoom1 ), zoom-
ing up to double the duration of the segments. Other processing on the original sound
is applied using the OM-Sox library (left part of the patch); the two results are then
concatenated (sound-seq) and normalised.
Figure 10. The sound processing sub-patch i-a01_2.
235
Takéshi Tsuchiya
Some sounds of the electroacoustic part of this section were also prepared using the
CSound engine through the OM2CSound library (see Figures 11 and 12). OM2Csound
allows the control of Csound by converting data to Csound scores, but also the visual
programming of Csound orchestras. In this case, most of the control process actually
takes place in the orchestra (Figure 12). The Csound instrument processes fragments
of the recorded sounds (recordings of violoncello played with the bow on the tail-piece
and on the bridge) using finely tuned feedback effects.
Figure 11. Sound processing/synthesis patch using OM2Csound.
236
Figure 12. Csound instrument definition (contents of inst-1 from Figure 11).
237
Takéshi Tsuchiya
Pitch structures for the violoncello part

Section II is the beginning of more active musical material (see Figure 13). In the
flow from the previous section, it begins with a harmonisation of the violoncello sound
combined with the fundamental tone of the filtered D] played by the violoncello (pro-
ducing an almost pure sine tone). After a passage of glissandi, the real-time recording
of the violoncello is played and integrated into a multilayered texture (see EVT 010 in
Figure 13).
Figure 13. The beginning of Section II.
The kind of glissando presented in this passage is one of the most distinctive features
of the piece (see also Section VIII, mm. 66 to 79 in Figure 14).
Figure 14. Section VIII.
238
These passages are not based on the general pitch system of the other parts of the
work. Pitches are determined following the process implemented in Figure 15 where
glissando time and width are computed from a graph drawn with a bpf object. Each
iteration, implemented as an omloop, allows the pitch intervals determined by the bpf
points to be calculated, and the timing of the glissando to be controlled piecewise for each
bpf segment. Note that this loop uses tak.acum3, a module of the Tak-Tools library
(a collection of my own composition tools) adding a special feature for accumulation in
OpenMusic loops.
Figure 15. Patch for generation of glissandi (II-A1_patch in Figure 7). The iterations (omloop
at the right of the figure) generate glissandi from bpf segments.
Sections IX to XI form one group developing pitch material with the goal of reaching
the highest E[ of the violoncello. In Section IX the specific notation of a curve is used
to represent the beams of long grace notes sequences. These notes are played a piacere
(see Figure 16). The patch used to create these structures is shown in Figure 17.
239
Takéshi Tsuchiya
Figure 16. Curved beams for grace notes in Section IX.
Figure 17. Main patch for Section IX. The pitch range of each beam (start and end pitch for
the sequence of grace notes) is determined by the set of chords at the top of the figure.
The main loop generating the pitch sequence is shown in Figure 18 (left). A random-
walk procedure is used (internal loop ran-w, displayed in Figure 19) to generate a sequence
of pitches within each specified range, given the number of elements required (specified
in the preliminary sketches—see Figure 3) and a pair of weights for the probability of
ascending vs. descending intervals (1:1, 3:1, 1:5, 1:1, 7:3, 1:4). In this process the internal
loop l-rem-d (also visible in Figure 19) discards tied notes and repetitions.
240
Figure 18. omloop for generating sequences of grace notes and its main sub-patch.
Figure 19. Internal loops from Figure 18. Left: random walk. Right: removing repeated notes.
241
Takéshi Tsuchiya
Eventually the melodic lines from Section IX continue to develop into figures including
ascending accelerandi (in Section X), then toward the final stage of a stable pitch profile
in Section XI (see Figure 20). This process is implemented using the interpolation tools
from the Profile library (see Figures 21 and 22).
Figure 20. From Section X to XI.
Conclusion
I have described some relationships of OpenMusic and compositional process in my work
Nothing that is not there and the Nothing that is for violoncello and electronics. With
the work’s background in the culture of Japan, as well as my own identity, it has aspects
related to Japanese tradition as a frame of thinking. As a result, while Kanji characters
or calligraphy are not directly represented, they do appear implicitly, for example, in the
relationship between the whole and parts of the structure of the work. One could say that
these strong ties are a unique cultural and artistic property, and become fundamental
propositions for musical creation.
Since I became acquainted with the field of computer-aided composition in the early
1990s, OpenMusic has become an invaluable tool for the control of temporal structures
in the field of sound synthesis, instrumental writing, and form. The concept of time in
composition, and its creative meaning, is a major concern for every composer. In this
respect the maquette offers a shift away from stereotypes, and allows for the creation
of new temporal frameworks—establishing a relationship between the internal structure
and the evolution of the piece in time. In addition, a close coordination with sound
generation and processing engines such as CSound and SuperVP, as well as the output
to Max, are now indispensable for the construction of my work.
242
Figure 21. Pitch interpolation in Section XI.
Figure 22. Visualisation of the chords inside the chord-seq from Figure 21. First and last steps
of the interpolation.
References
[1] Keiji Hashimoto. “A Dream of a Butterfly of Zhuangzi: A Meaning and Structure of
Wuhua”. Tetsugaku: The Journal of Hiroshima Philosophical Society, 51, 1999.
[2] Ryo Yanagi. Golden Section – The Ratio of Japan. Tokyo: Bijutsu Shuppan-Sha,
1977.
243
Koch’s Space
Julián Ávila
When we listen to a sound, our brain analyses its characteristics. This process of
extracting sound information in our brain always includes a description of spatial hearing;
that is, for any sound we hear, or brain determines its location so there is no “non-spatial
hearing” [2]. This behaviour could be an anthropological phenomenon with practical
application in life; but if source location is such an important element for our brains, it
seems reasonable that it should also be considered by composers.
Space is one of the most important elements in my music, and it is handled and
composed in the same way I work with other elements related to time, structure, process,
or the vertical and horizontal organisation of sound. For this reason I try to balance both
space and time (which in other disciplines are closely related concepts, for instance in
physics, which could be said to search for a “theory of everything”), a concern similar to
the connection that Karlheinz Stockhausen made between pitch, rhythm, and structure;
or later to spectral techniques.
An important issue is how to relate the two basic elements of time and space in order
to write music. In my music, fractals and self-similarity are highly useful tools to control
the processes that take place in a work. They allow me to apply the same idea in different
compositional strata without losing unity. The main goal is to choose a system applicable
to all the elements included in the composition of a space-time field.
Main idea of Koch’s Space

Koch’s Space (2011) is a work written and conceived from a spatial vision of sound. This
spatiality has informed the characteristics of the electronic and acoustic space, as well as
the structure of the score and sound materials.
The performance space is organised as a topographic filter 1 in which the sound is
spatialised using a light held by the performer. The illuminated area represents the
place that is excited by electronics, and sound filtering responds to different iterations of
a fractal structure, the Koch curve.2
1A topographic filter is a filter that depends on spatial location of the signal. For instance, location
could control cut-off frequency, Q factor, or any other filter parameters.
2 The Koch curve, also known as Koch Island or Koch Snowflake, “is a mathematically defined
coastline with triangular promontories whatever the scale of magnification. [. . . ] The Koch fractal
is the prototype of an extensive family of fractals based on the repetition of a simple geometrical
transformation”[3].
245
Julián Ávila
The score is divided into three pages located on three stands, positioned in a semicircle
(centre, left, and right) around the interpreter. The work is performed in darkness and it
is the interpreter, playing with a head lantern switched on throughout the performance,
who illuminates the staves.
The light is captured by photocells that allow electronic spatialisation using Max,
behaving as if it were the result of the illumination. The sound material is generated in
the form of an increasing and decreasing process of constructing the Koch curve so that
each performer’s movement, from left to right or from right to left, corresponds to an
iteration in this process. In this way the work attempts to “fractalise” not only sound
material and temporal process, but also the concert space and the electronics.
This work is closely related to the topic of spectral diffusion.3 Even though Koch’s
Space is not a “spectral diffusion piece”, as it is more related to topographic filtering and
spatial processes, it begins a compositional practice that moves in this direction and in
which research is currently open.4
Spatial development
There is only one sound source in this piece, the saxophone, along with its spatialisation
through the topographic filter, a “shadow” sound. This filter, parameterised using values
from the Koch curve, works as a shadow of the direct sound of the saxophone, meaning
that the public hears the direct sound coming from the stage—where the saxophonist
is positioned—as well as the filtered amplification coming from a surrounding array of
speakers. The filtered sound is spatialised precisely to the location that the performer
lights up in order to view the score.
The first step in the compositional process was the implementation of Koch’s fractal
shape in its first, second, and third iterations. These curves have been calculated as bpc
objects in OpenMusic using the OMChaos library and exported to an interpolation
table in Max. Figure 1 shows the fractal shape implementation that will be the main
source for all the other patches. The function make-w is used to define the horizontal and
vertical translation and angular shifting of the first iteration in order to build iterations 2
and 3.
Usually a filter graph is defined in dB (in the vertical, y-axis) as a function of frequency
(in the horizontal, x-axis). The filter used in this work is an FFT-based band-pass filter5
in which all the frequencies outside the band are completely attenuated, so there is
no need to graph the level. This makes it more convenient to present how the filter
3 Spectraldiffusion is the main topic of my Ph.D. research at NOVARS, University of Manchester, Spectral
Diffusion and Spectral Energy in Electroacoustic Composition: Sculpting Space in Time Informed by
Applications in Biophysics.
4 SpectralDiffusion and Spectral Energy research employing spectroscopy measurements has been widely
applied in the field of biophysics, using the Fourier Transform as its main analytical tool. Studies of
spectral diffusion in proteins [1] have identified behaviour, which composers can observe and reinvent
in the language of spectromorphology [6] and space-form [7] in electroacoustic composition. This work
with proteins may well be transposed and placed under the umbrella of spectral analysis of aesthetics
and music composition with computers, with a special emphasis on spectral and physical space.
5 The use of an FFT-based filter helps avoid a slope (dB/octave) at the cut-off frequency that exists in
traditional filters, as well as a change of phase in the filtered signal.
246
Koch’s Space
Figure 1. Implementation of Koch’s fractal iterations.
evolves throughout the space: the band-pass graphs show a frequency range on the y-
axis and represent space as an azimuth angle (in degrees) on the x-axis: hence the name
“topographic filter”. In other words, the band-pass filter limits vary depending on the
amplified sound position inside the performance space.
The shape of this topographic filter is derived from Koch’s fractal curves. In order to
keep the band-pass limits well-defined as a function of space, and to avoid holes in the
frequency bands, the original curves are transformed by taking the absolute value of the
x-increment as shown in Figure 2.
Figure 3 shows the original fractal shapes (iterations 1, 2, and 3), and the correspond-
ing transformed shapes for the topographic filter. The high-cut frequencies are scaled
between 200Hz and 500Hz, and the low-cut frequencies are an inversion of the same shape
scaled between 25Hz to 100Hz (see Figure 2). Figure 4 complements Figure 2, showing
the topographic filters that correspond to the first and third iterations.
247
Julián Ávila
Figure 2. Shape filter for iteration 2.
Figure 3. Transformation of filter shapes (iteration 1, 2, and 3).
248
Koch’s Space
Figure 4. First and third topographic filters.
The outcome of the patch from Figure 3 is exported to Max to modulate the limits
of the band-pass filter depending on the position of the performance illumination. The
three iteration curves produce three different filters that are applied to different sections
of the piece.
These filters have to be spread and “located” around the audience in the concert
space in order to achieve a spatial sound structure that changes when the light of the
performer is pointed towards a specific direction in the hall. As a many concert halls have
the configuration of an “Italian” theatre, the spatial distribution of the filtering process
is organised in a similar way (even though other kinds of configurations are possible).
The surrounding matrix of speakers6 cover the left, right, and back areas of the hall,
whilst the stage is reserved for the saxophone. So the direct sound will come from the
front, and the fractal-shaped filters will spread over the left, right and back, as shown in
Figure 5.
Figure 5. Topographic filter spatial distribution for the second iteration.
6 The number of speaker depends on the hall size, but at least four are needed to implement the filter
distribution in the room.
249
Julián Ávila
From space to score
There exist many examples of music that has been composed and afterwards spatialised
or distributed in space. However, in this case, the aim of giving the spatial design
equal importance to other musical element leads the compositional process towards an
organisation of all the elements around the space. The organisation of the score is
therefore conditioned by the spatial form: as the topographic filter is divided into three
“sections”, the fractal curve that shapes the sound material is divided into the same three
parts and distributed in three score pages.
The musical process consists of 12 constructions and deconstructions of the fractal
states, as well as mixtures or deformations between fractal iterations. Each complete
fractal shape is distributed across the three score pages (each one placed on one stand),
and the score of Koch’s Space is read horizontally across the pages, rather than from
the top to the bottom of each page. The first iteration (corresponding to the first stave
of each page) is read from left to right, but the second iteration is read from right to
left, and so on. Reading the score in this way will produce movements of the lighting
during the performance, which are tracked to control the amplified sound filtering and
spatialisation processes. Figure 6 shows the correspondence between the topographic
filter and the disposition of the score.
Figure 6. Relationship between space design and score structure.
250
Koch’s Space
Register and ranges

A characteristic of the Koch curve (and also of Koch’s snowflake7 ) is the relationship
between the limits of the perimeter and area of the fractal shape. When the number of
iterations approaches infinity the length (L) of each segment of the curve tends toward
0, while the perimeter (P ) of the whole shape tends toward infinity:
1 4n
L∞ = lim =0 P∞ = lim =∞
n→∞ 3n n→∞ 3n−1
In other words, whilst the area of this shape is finite, the perimeter is infinite. This
peculiarity is utilised in this work to define pitch range limits and register. The perimeter
of the shape is applied to the pitch range (scaled to fractions of an octave) and the length
of the small segments of the curve is applied to the register or starting note (also scaled
within an octave). Figure 7 shows the first nine numbers that determine the perimeter
and segment length of each of the first nine iterations of Koch’s fractal and their scaling
within an octave range.
Figure 7. Range calculation patch.
Generation of musical material

To generate the material for the saxophone score, the inversion of Koch’s curve has been
mapped to musical content in four different ways:
• Direct mapping.
• Mixing one curve iteration for pitches and another for rhythms.
• Interpolation between two different curve iterations.
• Superposition of curve iterations for use in different layers.
7 Koch’s snowflake is similar to the fractals presented in this chapter, but applied to the edges of an
equilateral triangle [4].
251
Julián Ávila
Direct mapping is the simplest translation between Koch’s fractal and musical ma-
terial, consisting of resizing the x-axis according to time and scaling the y-axis according
to register. This correlation between the fractal curve and music is used in Sections 3,
6, and 9. Figure 8 shows the patch for Section 6. Note that the align-chords function is
used to group nearby points from the bpc into chords. Figure 9 shows the corresponding
score extract.
Figure 8. Patch for Section 6 and example of musical transcription.
Figure 9. Score extract from Section 6.
252
Koch’s Space
Mixing uses different iterations for mapping rhythm and notes in order to obtain more
rhythmical content within an iteration. This rhythmical information is used not only for
repeated notes, but is translated into music using many different strategies. In fact, this
internal subdivision is used to change any parameter within the same note, for instance
timbre, dynamic, etc. The example in Figure 10 uses values from the first iteration inside
the rhythm of the second iteration of the curve, using the x-transfer function. Figure 11
shows the corresponding score extract from Section 4.
Figure 10. Patch for Section 4 (mixing).
253
Julián Ávila
Interpolation allows one to obtain a shape “between” two iterations (see Figure 12).
This method is used to generate additional material from the three iterations used for
this piece; the only requirement is that the two curves be scaled within the same range.
As different iterations have a different number of points, the om-sample function is also
used to resample both bpc objects before connecting them to bpf-interpol.
Figure 12. Patch for Section 8 (interpolation).
254
Koch’s Space
Superposition allows the use of points of two different fractal iterations for the same
section (see for instance the patch and score of Section 7, Figures 13 and 14). In this
case, lower notes are used as fingerings to produce non-ordinary harmonics in contrast
to the same notes produced with normal fingering. Due to the complexity of iteration 3,
the function omquantify is used here to avoid small rhythmical subdivisions, as well as
align-chord to group coincident points in a chord, and merger to mix both shapes into
the final voice object.
Figure 13. Patch for Section 7 (superposition).
255
Julián Ávila
These four methods for mapping material are combined to compose the structure
of the piece. Sections 1 to 9 comprise a process in which the fractal grows little by
little—from the simplest shapes to the most complex. Section 9 contains the shape that
corresponds to iteration 3 of the fractal—the furthest stage of the fractal used in this
piece—and after this point iterations decrease down to the first one again. Table 1 shows
how the musical material is organised to create the macro-structure.
Section Pitch range Duration

Direction Iteration Filter
/staff (midicents) (seconds)
Rhythms: K1
1 → Pitches: K0
7400 25”
Rhythms: K2
2 ← Pitches: K0
7400 25”
3 → 7400–7350 33”
K1
Rhythms: K2
4 ← Pitches: K1
8200–8100 33”
5 → 8200–8000 33”
Interpolation
K1-K2
6 (←→←) 8200–7900 45”

K2
7 → K2 and K3 8450–8000 45”
8 ← 8450–7850 45”
Interpolation
K2-K3
9 → (no filter) 8600–7400 60”

K3
10 ← Same as Section 6 (different musical realisation)
11 → Same as Section 3 (different musical realisation)
12 ← Same as Section 1 (different musical realisation)
Table 1. Overall structure of Koch’s Space. Direction indicates the reading direction (i.e.
spatial movement) of the section. Iteration indicates the curves used to determine the rhythm
and pitch material. Filter is the “topographic filter” used in the section.
256
Koch’s Space
System design
The live electronic system in Koch’s Space works in a similar way to procedural audio—
for instance as used in game engines—in which it is not possible to know in advance
the exact final result of the sound, but the behaviour of the sound is clearly defined.
Max is used to handle live interactions, and Arduino serves as an interface between
the photocells and Max.8 No score follower or event trigger is used; the system reacts to
the performance and leaves the performer absolutely free to take control of the electronic
system with her or his movements. Figure 15 shows a diagram of the different parts of
the system.
Figure 15. Block diagram of the live electronic system.
8 Arduino is an open-source system allowing to create micro-controllers and digital devices.
257
Julián Ávila
On stage
This chapter points out the importance of space and describes a methodology to work
in time-space composition balancing these two basic musical elements. I try to make no
separation between time and space elements, so that almost all the elements in this work
have properties in both fields. In fact this particular treatment of space is not only a
compositional idea: the performance also puts forward relationships between sound and
space, intensified by the visual effect of illumination which takes an important role in the
transmission of the musical idea.
Spatiality in music is not a new invention. There are numerous ancient examples,
such as Ancient Greek theatre; the works of Alessandro Striggio, Giovanni Gabrieli,
Thomas Tallis; the 17th -century cori spezatti [5] in San Marcos of Venice; and 20th -
century examples such as Karlheinz Stockhausen’s Gruppen (1955–57) or Luigi Nono’s
Prometeo (1981–85), to name just a few. But today’s technology allows us to think
differently about spatiality in music, so that new methods can be implemented as a
potential basis for new music. The initial idea of Koch’s Space was developed through
recent compositional tools, and most important, the idea of space has transcended the
score or the compositional process to be present in the performance.
References
[1] Jürgen Baier, Mads Gabrielsen, Silke Oellerich, Hartmut Michel, Marin van Heel,
Richard J. Cogdell, Jürgen Köhler. “Spectral Diffusion and Electron-Phonon
Coupling of the B800 BChl a Molecules in LH2 Complexes from Three Different
Species of Purple Bacteria”. Biophysical Journal, 97(9), 2009.
[2] Jens Blauert. Spatial Hearing: The Psychophysics of Human Sound Localization
(Revised Edition). Cambridge, MA: MIT Press, 1996.
[3] Hans Lauwerier. Fractals: Images of Chaos. London: Penguin Books, 1991.
[4] Benoit Mandelbrot. Fractals: Form, Chance, and Dimension. San Francisco: W. H.
Freeman & Co., 1977.
[5] Davitt Moroney. “Alessandro Striggio’s Mass in Forty and Sixty Parts”. Journal of
the American Musicological Society, 60(1), 2007.
[6] Denis Smalley. “Spectromorphology: explaining sound-shapes”. Organised Sound,
2(2), 1997.
[7] Denis Smalley. “Space-form and the acousmatic image”. Organised Sound, 12(1),
2007.
258
Electronic sound creation in
Balænoptera for bass clarinet,
electronic sounds, and live electronics
Fabio De Sanctis De Benedictis
On March 19th , 2013 a fin whale was beached south of Livorno. As, I suppose, in
recent times many Italians and perhaps Europeans—surely I—may feel confused like that
cetacean, this was the pretext to begin to plan Balænoptera, a quadraphonic composition
for bass clarinet, electronic sounds, and live electronics.1 The present paper is focused
on the use of OpenMusic to generate electronic sounds, or Computer Generated Sounds
as expressed by David Cope [2]. One of the strongest aspects of OpenMusic is its
integration with other IRCAM software; therefore it is in this direction that it has been
used in Balænoptera.2 A concise description will be given of formal and compositional
techniques in the work as they relate to algorithmic composition.
Remarks about the form

The starting sound materials in Balænoptera are whale calls found on the Web, as well
as bass clarinet sounds performed by Carlo Failli: breath sounds, several low B[ tones,
key noises, slap tongues, and four multiphonics that will be identified as numbers 1 to 4
throughout this text. The formal plan started out from the definition of an overall
duration of about 10 minutes. Pitch classes were extracted from the letters of the word
Balænoptera (considering “æ” like “e”) numbered according to the Italian alphabet from
0 upwards, modulo 12:
0 1 2 3 4 5 6 7 8 9 10 11
A B C D E F G H I L M N
O P Q R S T U V Z
1 Thiswork was requested by my friend and colleague Carlo Failli and is dedicated to him. The premiere
took place in 2015 during the sixth edition of the contemporary and electronic music concert series
Suoni Inauditi (Unheard Sounds) at the Istituto Superiore di Studi Musicali “Pietro Mascagni”.
2 The main OpenMusic libraries used in this work are OM-SuperVP, OM-Spat, and OMChroma, as
well as chord-sequence analyses obtained from Audiosculpt. PWGL and the PWCsound library have
been used as well; however in the course of this paper the examples of algorithmic processes, even if
realised with PWGL patches, will be presented in the form of analogous OpenMusic patches.
259
Therefore the word Balænoptera corresponds to the numbers 1, 0, 9, 4, 11, 0, 1, 5, 4,

3, 0. Converted to pitches, these numbers correspond to C], C, A, E, B, C, C], F, E, D],
C. These pitches belong to the pitch-class set whose prime form is {0,1,2,4,5,6,8}, denom-
inated 7-13 according to the classification by Allen Forte.3 The complementary pitches
(necessary to complete the chromatic totality) are D, F], G, G], A], which corresponds
to set 5-13, whose prime form is {0,1,2,4,8}. The intervals between adjacent pitch classes
contained in these two prime forms are (1,1,2,1,1,2) and (1,1,2,4) respectively. These
numbers are a structural reference point in the composition of Balænoptera, in relation
primarily to rhythmic parameters and secondarily to the form.
The total duration of the piece (10 minutes) was divided into four sections of 1, 1, 2,
and 6 minutes.4 The final unit (6) was divided into six subsections with relative durations
1, 1, 2, 1, 1, 2. So the overall form can be summarised as follows:
1. → (0’–1’, mm. 0–15): Introduction.
1. → (1’–2’, mm. 16–30): More melodic part for bass clarinet.
2. → (2’–4’, mm. 31–60): Reprise of material from first section in bass clarinet part,
electronic sounds created by cross-synthesis of whale sounds and bass clarinet
multiphonics.
6. → (4’–10’):
1. → (4’–4’45”, mm. 61–72;): Addition of granulation effect, increase of tension.
1. → (4’45”–5’30”, mm. 73–84a5 ): More dramatic; quarter tones are not used in
bass clarinet part, contrary to previous sections; first introduction of arpeggio
figures in bass clarinet part.
2. → (5’30”–7’, mm. 84b–106): Continuation of preceding tension, but gradually
decreasing; stabilisation of arpeggio figures and arc figures in bass clarinet
part, with corresponding rapid circular movements in electronic sounds.
1. → (7’–7’45”, mm. 107–117): Idem, with additional stretched sounds.
1. → (7’45”–8’30”, mm. 118–129): Continuation of previous processes and syn-
thetic recapitulation of bass clarinet sounds from part 1.
2. → (8’30”–10’, mm. 130–end): Final duet between bass clarinet and whale
sounds; return of quarter tones in bass clarinet part; no live electronics, only
slightly reverberated whale sounds.
The basic metre is 4/4, at metronome marking 60, so the number of beats corresponds
to the real duration of the work or section in seconds. Measure lengths were slightly
adjusted at certain points for musical and expressive reasons: for instance three seconds
were added at the end of measure 72 to create a dramatic pause. The first section to
be composed was the final duet, the conclusive point of the formal path and generative
expressive idea; the other sections were composed sequentially, beginning from the first
to the penultimate one.
3 See [4]. The interested reader can find fruitful information on Pitch-Class Set Theory there as well as
in [8] and [9].
4 (1,1,2,6)
is used here instead of (1,1,2,4) accepting Donatoni’s theory of the unconscious and error [3],
and following my initial feeling about this subdivision even if incoherent with other decisions.
5 84a indicates the first beat of measure 84, 84b the second beat, and so on.
260
Electronic sound creation in Balænoptera
Audio generation in the first section (0’–1’)

As specified in the introduction, OpenMusic has been used mainly to generate a set
of sounds to be mixed together in order to compose the electronic part of Balænoptera.
In this and the following sections the processes of electronic sound generation will be
described.
To compose electronic sound material for this section, clarinet key noise recordings
were mixed and edited together to form a denser sound, then stretched using Au-
diosculpt to reach a duration of exactly one minute. This resulting sound was then
filtered with three surface bandpass filters using the following frequency ranges: 2123–
4116 Hz, 4116–6100 Hz, and 6100–8184 Hz.6 The three filtered sounds were then mixed
together with starting times offset by 0, 5, and 10 seconds.7
Other sounds were obtained using CataRT by importing the four multiphonics,
moving the mouse freely in the descriptor space view proposed by the software, and
recording the audio output. Finally four one-minute-long mono files were obtained, each
composed by different means:
1. Low bass clarinet B[, stretched to 1’, mixed with unfiltered key noise.
2. Low bass clarinet B[, stretched to 1’, mixed with the three filtered key noise sounds
described above.
3. Sound created by improvising with the mouse in the CataRT descriptor space.
4. The previous sound (3), reversed.
These sounds are spatialised on a quadraphonic system using the patch shown in Figure 1.
Onset times of respectively 0, 1, 2 and 4 seconds were applied to the four sounds (+1, +1,
+2, deduced from the incomplete series 112—the first or last three numbers of 112112).
Figure 2 shows the movements of the different sounds over the four speakers.
Audio generation in the second section (1’–2’)

In the first section the bass clarinet mainly uses breath sounds, timbral trills, the low
B[, and key noise. Some of these sounds were also selected for the audio material of the
second section, which becomes a sort of electronic variation of the bass clarinet part in
the previous episode. The sounds selected for the electronic part include: filtered key
noise, the sound of a water droplet, the four multiphonics filtered with the same frequency
bands used for key noise (2132–4116, 4116–6100, 6100–8184 Hz), a breath sound, a slap
tongue, and a whale call stretched to 10 seconds. Besides creating a link to the previous
section, both for the bass clarinet part and the electronic sounds, the aim is to gradually
introduce the whale sounds to the texture.
6 Thesefrequency bands are approximately proportional to the intervals of contiguous pitches in the first
chord extracted from the analysis of multiphonic #4. The threshold of 2123 Hz corresponds to the
highest frequency of the multiphonic.
7 The choice of starting times is based on the number five, which is another important structural element
of the work. Five is also the number of formant filters which are used in other sound transformations
later in the piece.
261
Figure 1. Spatialising the audio files of the first section using OM-Spat.
Figure 2. Sound spatialisation, First Movement. (1) NW-NE-SE-SW-NW, (2) NE-NW-SW-

SE-NE, (3) SW-SE-NE-NW-SW, (4) SE-SW-NW-NE-SE.
262
Four audio files (one for each channel) were created using CataRT by performing
four different movements in the descriptor space (Figure 3).8
Figure 3. Movements in CataRT descriptor space view. (1) NW-NE-SE-SW-NW,

(2) NE-NW-SW-SE-NE, (3) NW-NE-SW-SE-NW, (4) NE-NW-SE-SW-NE.
The recorded audio was then stretched to exactly one minute. Each sound was assigned
to a separate channel, with no further spatialisation or movements. The onset times
were set to 0, 2, 4, and 8 seconds in this section (0, 1, 2, and 4 multiplied by 2). As a
final compositional decision for this section, I decided to begin the electronic part eight
seconds before the instrumental part.
Bass clarinet part in the second section

The bass clarinet part is composed according to criteria of invariance by transposition,
or by inversion followed by transposition, of the pitches extracted from the multiphonics
sounds. Invariants are the common pitches of two different chords, and the theoretical
tools underlying invariance by transposition or inversion are derived from Robert Morris’s
Composition with Pitch-Classes [7] (pp. 36–51). A PWGL patch was created in order to
verify the invariance by transposition of pitch fields, shown in Figure 4 in a version for
OpenMusic.
For the sake of simplicity in Figure 4, the first six multiphonic pitches are en-
tered as MIDI note numbers. On the left side of the patch the first omloop, labelled
“transpositions”, transposes every chord pitch incrementally, by twenty four ascending
quarter tones, until reaching the interval of an octave. The second omloop, labelled
“intersections”, calculates the absolute pitch intersections of every transposition with
the original chord. The results are listed in textfile boxes: here, the first number of
every pair indicates the transposition interval, while the second indicates the number
of invariants for that interval.9 The same process is shown on the right of the patch,
except the initial chord is transposed by descending quarter tones. All of the pairs were
investigated and some were selected to constitute the harmonic progressions of the bass
clarinet part. At the same time, the identical multiphonic sound material was used in
the electronic part to create a connection and coherence between the two parts.
8 Note the similarity to the spatialisation patterns indicated in Figure 2.

9 Morrisuses addition and subtraction matrices for an exact and complete calculation of the number of
invariants; however this process, even if already formalised in other patches, was not used here.
263
Figure 4. Verification of the number of transposition invariants.
Figure 5. The same patch as in Figure 4, now using the function sdif-›chord-seq to extract all
the pitches of the multiphonic chord sequence, sorted and without repetitions.
264
In Figure 5 we can see the same process applied to all the pitches of the multiphonic
chord sequence. Clearly the number of invariants increases as a consequence of the
large number of pitches. A different version of this algorithm can be realised using
the function as-›om as a substitute for sdif-›chord-seq. The former permits one to set
the maximum polyphony and so to control the complexity of the chords and generated
combinations. During the composition of Balænoptera both functions were utilised and
the most satisfactory solutions were selected.10
Audio generation in the third section (2’–4’)

In this section the whale sounds were cross-synthesised [5] with multiphonics using the
OM-SuperVP library (see Figure 6). Six whale calls were selected to be crossed with
the four multiphonics, for a total of 24 combinations.
Figure 6. Patch for cross-synthesis.
10 The use of invariants of chords deduced from audio analysis represents a topic of interest for the
development of compositional tools and a research step toward a possible integration of Spectralism
and Pitch-class Set Theory.
265
On the left of Figure 6 the mean fundamental frequency and the duration of each
sound are extracted. The difference of pitch is used in supervp-transposition to homogen-
ise the pitches, and the ratio of durations is used to set supervp-timestretch to obtain
sounds of the same duration prior to cross-synthesis.
Then the resulting sounds are distributed with attack times derived from a rhythmic
series obtained using the patch shown in Figure 7: the 120-second-long section is divided
proportionally into 1+1+2+1+1+2 parts (112112 series), and then each part is subdi-
vided into 1+1+2+4 (1124 series) as well as 4+2+1+1 (retrograde of 1124), according to
Boulez’s concept of “de-multiplied rhythm” ([1], pp. 145–146). The two rhythmic series
based upon 1124 and 4211 sub-divisions (from the second and third staff of the last poly
object in Figure 7) are shown in Figure 8.
Figure 7. De-multiplied rhythms.
The cross-synthesised audio files were also time-stretched, by factors of two and four,
and arranged in four channels, deployed cyclically from channel one to four until all files
were used. They were ordered according to the attack times of the two rhythmic series of
Figure 8, as well as a third series whose values were determined, in order, by the average
of the respective attack times of the first two series. In other words, the first attack time
of this series is 0, the second attack time is the average of the second attack times in the
two original series, and so on.
The mono mix of the quadraphonic file was analysed and transformed into a chord
sequence, whose rhythm offered the basis for the rhythmic figures of the bass clarinet
part in this section (see Figure 9).
266
Figure 8. Two rhythmic series obtained from the patch in Figure 7.
Figure 9. Extraction of rhythm from a chord sequence.
267
Audio generation in the fourth and fifth sections

(4’45”–5’30” and 5’30”–7’)
In this section the whale sounds were again cross-synthesised with bass clarinet sounds,
as developed in the previous section (specifically the non-stretched ones). These sounds
were filtered by a formant filter, as shown in Figure 10, also applied recursively, as shown
in Figure 11.
Figure 10. Formant filter with random selection of vowels and voice type.
The formant filter in Figure 10 is applied to the sound according to a specified

number of vowels and voice types (soprano, contralto, countertenor, tenor, bass) each
selected randomly with the function database-formants. The random choice is repeated
five times, obtaining five (probably) different filter parameters applied consecutively along
the full length of the sound. The left part of the patch is dedicated to the compilation
of formant filter parameters, in the following order: the start times of each filter, equally
distributed along the duration of the sound; the number of formants (5); and the five
triplets of formant values (central frequency, amplitude, bandwidth) corresponding to a
268
random vowel of a random voice type. SuperVP performs interpolations between the
filter values, resulting in smooth transitions from each state to the following one. In
addition, the functions supervp-timestretch and supervp-transposition apply stretching
and transposition, independently, to the final sound.
In Figure 11 we can see the patch corresponding to the recursive formant filter. In
each sub-patch there is an instance of the filter from Figure 10, with the number of
vowels/voices set to 5. Filtered sounds can be transposed ad libitum using the supervp-
transposition function.
Figure 11. Recursive formant filter (5 poles).
The sounds in the electronic part are constituted by mixing the sounds produced
by the filters in Figure 10 and Figure 11, the latter with transpositions set to 0, -12,
and -24 semitones. The sounds were selected in retrograde order with respect to their
appearance in the previous section. In total, 24 audio files were distributed across a
duration of 45 seconds. The previous recursive rhythm algorithm (Figure 7) was used
once again, dividing the overall duration into 2+1+1+2+1+1 parts (retrograde of series
112112) and subdividing each part into 4+2+1+1 parts (retrograde of series 1124). The
result can be observed in Figure 12.
Figure 12. Rhythm of audio file attack times in fourth section from 4’45” to 5’30”.
269
Other sounds in these sections were synthesised in Csound using the OMChroma
library. Two instruments were created using the STKClarinet 11 opcode (without and
with an additional amplitude envelope) and converted to OMChroma classes.
The synthesis was parameterised using the pitches of chord sequences obtained from
the bass clarinet multiphonic analyses. So natural sounds are put into a dialectic with
the corresponding artificial ones, a sort of mirror game. The patch implementing this
process is shown in Figure 13. This patch also allows a duration multiplier to be set, in
this case to the values five or ten (again using five as structural number). The synthesised
sounds were processed by the formant filter as in Figure 10.
Figure 13. Synthesising with the Csound STKClarinet instrument in OMChroma.
The analysis data from the four multiphonics were used and re-synthesised succes-
sively with the two Csound instruments and the two different duration multipliers. So
a total of 16 different sounds were created, then submitted to the formant filter with
transpositions of -24, -12, 0, +12, and +24 semitones, producing a total of 16 × 5 = 80
different sounds. The transposed versions of the same sounds were mixed in succession,
with a global fade-in/out, according to attack times of 0, 1, 2, 4, and 5 seconds deduced
from the series +1, +1, +2, +1 (a part of 112112). The order of the files in this
mix was derived from the melodic profiles of bass clarinet figures played just before,
and developed by inversion, retrograde, and retrograde-inversion according to Morris’s
P-Space theory.12 Finally the mixed sounds were distributed along a duration of 90
seconds using the same recursive rhythm generation method as before.
11 STK (The Synthesis ToolKit): https://ccrma.stanford.edu/software/stk/download.html

12 Forexample, the melodic profile low-high-medium can be developed by retrograde obtaining medium-
high-low, by inversion obtaining high-low-medium, and by both processes obtaining medium-low-high.
270
In subsequent sections the electronics were developed according to principles and

techniques similar to those already illustrated, adding Paulstretch13 for stretching
audio files created with PWCsound14 and the STKClarinet opcode, as a variation of
the previous operations.
Final duet
The last part (actually the first to be composed, and the initial inspiration for the work)
proposes a duet between the bass clarinet and the whale sounds, as if the instrument swam
side by side with the great mammals, an episode of calm resignation and reconciliation.
The whale sounds are almost unprocessed (except for reverb applied in Max). So in the
end the source material is clearly revealed, and serves to reconfigure and feed back on
the memory of what was heard until this moment (see [6]). The section lasts about one
minute and a half. Ten whale sounds have been selected and distributed along a duration
of 90 seconds, approximately one every nine seconds.
The whale sounds were analysed as chord sequences then translated into musical
notation in OpenMusic (see Figure 14). The chords, approximated to quarter tones
and transposed down by an octave, give a pitch reservoir from which the bass clarinet
Figure 14. Extraction of pitches from a whale sound chord sequence.
13 Paulstretch is a program for extreme audio stretching. In contrast with AudioSculpt, it does not
preserve attack transients when stretching the sounds, but smooths the resulting sounds to obtain
particular textures. See http://hypermammut.sourceforge.net/paulstretch/.
14 PWCsound is a library created by Giorgio Zucco for controlling Csound in the PWGL environment.
See http://pwcsound.jimdo.com/.
271
pitches are derived. The chord sequences obtained by the function as-›om have also
been transformed into rhythmic sequences as in Figure 9 and mixed together to form the
clarinet rhythm following the succession of the corresponding whale sounds.
Figure 15 shows the patch related to this process. This patch gathers in a maquette
the voices resulting from the sub-patches cb01, cb02, etc. with different whale sound
analyses. These patches are similar to the patch in Figure 9, here using as-›om instead
of sdif-›chord-seq. Each voice is assigned an onset time corresponding to the onset of
the corresponding sound in the electronic part. Figure 16 shows the contents of the
maquette.
Figure 15. Extraction of the overall rhythmic polyphony from whale sounds.
Figure 16. The overall rhythmic polyphony inside the maquette.
The maquette allowed me to listen to and verify the overall rhythm, with every
rhythmic sequence transposed to a different pitch. By exporting the individual sequences
in a format suitable to be read by Finale, and rebuilding the whole structure as a
polyphonic score, it was possible to print sheet music to be used as a reference for the
rhythm of the bass clarinet part. Figure 17 shows the beginning of the instrumental part,
transposed, as presented in the final duet.
272
Figure 17. The beginning of the final duet: bass clarinet part, transposed.
Conclusions
OpenMusic and its libraries allowed me to structure the audio material in this com-
position in a way consistent with the form, as well as to produce strong connections
between the instrumental and electronic parts. In particular, it permitted a gradation in
the formal unfolding of the types of electronic sounds used, both in terms of their inner
composition and the process of their implementation. OpenMusic also allowed for the
use of the same material and processes both for the audio part and the instrumental part:
multiphonics, formant frequencies, numeric series, and so on. This careful calibration of
processes and materials ensured a high level of formal control, and the same criteria
were applied to the live electronics. Balaenoptera goes a long way beyond personal and
expressive content to mark a turning point in my composition operations, laying the
foundation for further future developments toward a significant integration of acoustic
and electronic sound.
273
References
[1] Pierre Boulez. Note di apprendistato. Torino: Einaudi, 1968.
[2] David Cope. “Preface”. In Jean Bresson, Carlos Agon, Gérard Assayag (eds.) The
OM Composer’s Book 2. Editions Delatour France/IRCAM-Centre Pompidou, 2008.
[3] Franco Donatoni. Questo. Milano: Adelphi, 1970.
[4] Allen Forte. The Structure of Atonal Music. New Haven: Yale University Press, 1973.
[5] Mario Mary. Audiosculpt Cross-Synthesis Handbook. IRCAM Documentation, Paris,

2nd edition, 1996.
[6] Leonard B. Meyer. Emozione e significato nella musica. Bologna: Il Mulino, 1992.
[7] Robert D. Morris. Composition with Pitch-Classes: A Theory of Compositional
Design. New Haven: Yale University Press, 1987.
[8] Susanna Pasticci. “Teoria degli insiemi e analisi della musica post-tonale”. Bollettino
del GATM, Anno II(1), 1995.
Online
[9] Gary Tucker. “A brief Introduction to Pitch-Class Set Analysis”. Mount Allison
University, 2001.
http://www.mta.ca/faculty/arts-letters/music/pc-set_project/pc-set_new/.
Acknowledgements: I would like to thank Daniele Salvini and Fabrizio Broglia of the Mathematics
Faculty of Pisa University, through whom I had the opportunity to meet Carlos Agon and Moreno
Andreatta. I wish to thank Carlos Agon, Moreno Andreatta, and their colleagues, whose work
has expanded my compositional horizons; and Jean Bresson, whose support, keen suggestions, and
corrections to this paper have brought it greater clarity and lucidity. Last but not least I thank Carlo
Failli for his excellent performance of Balænoptera.
274
Modelling a gesture: Tak-Sı̄m for
string quartet and live electronics
Alireza Farhang
This chapter offers a general outline of the processes of gesture modelling which were
devised for, and applied to, Persian music in the composition of Tak-Sı̄m, written for
The Kronos String Quartet. After a brief elucidation of the background to the work’s
composition, the various strategies that allowed me to integrate elements from Persian
musical culture will be presented, in an attempt to illustrate how conflicts between two
ontologically opposed musical cultures become a source of enrichment. Using analyses
of recordings by master setār player Ahmad Ebādi, I created models of ornamentations,
melodic fluctuations, timbral evolutions, extended techniques, and other parameters of
performance.
One cannot avoid the acquisitions of the past, except by regressing to a truly
primitive state. [...] Far from adding constraints, these acquisitions, or in other
words, our very culture, our mental functioning forms part of our musical material,
just as much as known or imagined sounds, and can be integrated with every degree
of freedom into a new musical discourse.
Tristan Murail [7]
Introduction
As a composer myself, this statement from Tristan Murail is significant to me. The sound
universe of a composer is not born out of a vacuum; rather, it is a reverberation of the
sounds with which that composer has lived. For a composer educated in both the Western
and Persian classical traditions, the process of composition takes on new proportions.
The composer must deal with problems related to an ontological gap between these two,
distant musical cultures. When it comes to rich and ancient cultures, the composer is
torn between two different ways of thinking that are sometimes radically opposed, and
that prevent him from seeking an original and personal language. On the one hand
he/she weaves emotional ties with his or her culture of origin, which places him/her in
the golden cage of eternal and timeless wonders, taking away all freedoms; on the other,
the composer lives in a modern world where discourse is based on a questioning of the
past.
As a composer who is familiar with the rationality of the music of Beethoven, Debussy
and Ferneyhough, the heavenly serenity and spirituality of Persian music, as well as the
noise of war and chaos of modern life, my career path reflects a perpetual challenge:
how to address the gap between the content of a pre-Galilean conscience, and the
275
Alireza Farhang
form of a post-Hegelian discourse.1 In this line of research, notions of expression and

gesture begin to play an increasingly important role in my musical discourse, and have
facilitated an informed reflection upon material which is endowed with strong cultural
connotations. The modelling process is essential; the concept of gesture helps in dealing
with problematic related to the analysis, representation, decomposition, and finally
the composition of elements from the Persian musical culture that resist analysis with
conventional methods.
Tak-Sı̄m 2 for string quartet and electronics was composed with pencil, paper, eraser,
and of course computer. Most operations were performed in the OpenMusic environ-
ment. Using various tools of representation, the potential of Persian music to generate
compositional material are presented. Problems I faced in rendering musical material
malleable by negating its cultural connotations, and the morphology of the gesture as a
model for composition and synthesis, are among the other subjects that are discussed in
this chapter.
The rudiments of expressivity in Persian music

In order better to understand the æsthetic basis of the process of composition of Tak-Sı̄m,
I will present here a brief introduction of Persian music.
The art of melody

Persian music is monodic, i.e. it does not contain superimposed lines. In contrast,
polyphony, which emerged during the Middle Ages, set Western music on a contrasting
historical trajectory. This distinction became all the more pronounced when the Western
tendency for rationalism was applied to polyphonic composition. Rhythm and melody,
within this context, lose their perceptual value and their original base functions. Rather,
they co-exist in architected structures conceived by the composer; their salience ranges
from explicit to subordinate to mere allusion. The need for such acoustic depth is absent
from Persian music, which favours abstract and geometric melodic figures. Closely
related to its counterparts mythology and poetry, it is characterised by an excess of
ornamentation, rich and subtle fluctuations on a grid of micro-intervals, and regular or
complex rhythms. The absence of polyphony, which is the basis for its richness, might
be compared to the absence of perspective in Persian miniatures. The latter, which are
no less connected to mythology and poetry, are a complex art form built upon pure and
simple geometry and traditionally painted in bright colours.
1 Terms used by Daryush Shayegan [8].

2 The reasons for the choice of this title will be discussed later in this text. Tak-Sı̄m was a coproduction
of the Philharmonie de Paris and IRCAM-Centre Pompidou, premiered in 2012 by the Kronos Quartet
(David Harrington, John Sherba, Hank Dutt, and Jeffrey Zeigler) at Cité de la Musique, as part of the
5th Biennale of the String Quartet. The electronics in this piece could not have been realised without
the know-how of the computer music designer, Benoit Meudic, who assisted me throughout this project.
276
Modelling a gesture: Tak-Sı̄m for string quartet and live electronics
The radif
Persian music places great importance upon improvisation, but only according to a strict
set of rules. The radif 3 is a “systematic, sequential organisation of all melodies, motifs
and variations” [2]. This collection of types of melodies, organised according to a special
logic, is the basis of musical material that is memorised and transmitted orally from
master to student; its use implies a structured representation of pre-melodies which are
freely connected by improvised ornamentations. The standard radif is the declension of
12 sequences of which 5 are called āvāz (songs) and 7 are called dastgāh (which literally
means position of the hand).4 The abstract, succinct nature of these melodies gives
considerable freedom to the musician, allowing him or her to improvise with ornaments
that are characteristic of Persian music. In this way the performer may add his or her
own personal touches to a performance whilst still respecting the tradition of the radif.
Modes and micro-intervals

Many theoretical texts have been written describing the modal systems of Persian mu-
sic. The problems associated with the elucidation of a theoretical framework have led
scientists, musicians, and musicologists to propose different approaches. Owing to the
folkloric manner in which the tradition is transmitted, each master typically intuitively
formulates his or her own system based upon the characteristics of the instruments being
used, and his or her personal experience.
The Daryush Talāı̄ model

The model proposed by Daryush Talāı̄5 may be seen as a synthesis of the aforementioned
theoretical paradigms. Talāı̄ crystallised in a simple, succinct manner the general modal
systems which are used in the radif. The value of Talāı̄’s model lies in the fact that the
modes in Persian music are determined through the superimposition of only four dāng
(tetrachord, see Figure 1): shour, dashtı̄, tchargāh and māhour [9].
Figure 1. Intervals of the radif modes (dāng) in midicents.
Intervals and their relationships to one another constitute the basis for musicologi-
cal/theoretical research. In general, they are performed, checked, and analysed on string
instruments, in particular the oud (lute). The strings on the oud are tuned in perfect
3 This term translates literally as “series”.

4 “As
the taxonomy and the order of radif were determined by different personalities, the same melodies
may occasionally have different names and be introduced in different orders” [2].
5 Talāı̄ is a master of Persian music and an outstanding performer of tār and setār.
277
Alireza Farhang
fourths; thus in Persian music the fourth is the key interval which the performer may
play without changing hand-position. The octave, the fifth, and the fourth correspond
to the fixed frets on the instrument.
Two superimposed dāng are known as a māyeh 6 and create a certain modal colour.
According to Talāı̄, a dastgāh or a āvāz may contain several māyeh. The performer must
be able to navigate between various māyeh in an appropriately idiomatic manner.
Circular representation
Circular representation, originating from astronomy, is one means of visually analysing
modal/tonal systems and their relationships to each other. In Iran and Greece, this type
of representation, used for star-charts and music alike, has existed for many centuries
(see Figure 2).
Figure 2. Instances of circular representation used in ancient musicological treaties: internal

divisions of intervals of a fourth (dāng), fifth, or octave, which Abd al-Qāder Gheybı̄ al-Marāghı̄
explains in his 15th century book Maqāsed al-alhān [1].
In OpenMusic the class n-cercle (circular representation) and the c2chord function
(to convert the circular representation to a chord), make it possible to create models of
combinations of modal and tonal transposable systems; furthermore, they allow the user
to analyse non-octavian scales which originate from the superimposition of two ore more
modes. In order to facilitate the representation of intervals, an approximation to 1/8th
tones was used. The superimposition of two dāng of shour and tchārgāh is a means of
constructing the māyeh of homāyun. Once the pitch approximation has been applied,
intervals are organised by 1/8th -tone as shown in Figure 3.
Figure 4 shows an OpenMusic patch containing a circular representation of the
māyeh of homāyun. In this example, the space between each point on the circle cor-
responds to 1/8th of a tone (the octave is therefore divided by 48). We may observe
that the range of the māyeh of homāyun is limited to a minor seventh (in this case from
D to C). Thus, in order to complete the circle (or octave), a tone is added at the end
of the māyeh. The patch superimposes the intervals of two dāng (in this case, shour
and tchārgāh) specifying an interval (also in 1/8th tones) between them. It outputs the
scale-degrees of the māyeh in the form of a chord.
6 Although māyeh is the result of superimposed dāng, it is described as “mode” in Western musical
terminology (i.e., whereby no distinction is made between the concepts of mode and māyeh).
278
Figure 3. Intervals of the radif modes (dāng) in midicents and1/8th tone intervals.
Figure 4. Representation of the structure of the homāyun using the class n-cercle.
279
Alireza Farhang
Each combination of the four modes yields a different acoustical result. On occasion,
the modes overlap to produce new modes with a greater range. In these cases, it
is possible to round out micro-intervallic deviations in order to avoid discrepancies.
The phenomenon of overlapping modes can give rise to flexible-interval modes which
necessitate the use of accidentals. In Figure 5 we see the intervallic structure of āshour-
āvand, a sub-division of the radif which results from superimposing the māhour and
shour modes. In this example, the starting point of the māhour mode is C, and that
of the shour mode is D, a difference of 200 cents. The result of superimposing these
two modes is a new, third mode which spans C to G, in which E is inflected both an
1/8th -tone lower and a 1/4th -tone lower (E[ a 1/4th -tone higher).
Figure 5. The intervallic structure of āshour-āvand.
The patch shown in Figure 6 helps to understand the intervallic structure of the
modes. Also it allows one to make new combinations of modes by superimposing two
dāng while specifying respective starting points separately. The sub-patch scale_maker-2
receives the intervals, the starting-point, a reference-pitch, the degree of approximation
(25 midicents, or 1/8th tone, in this case) and the range of the scale (for instances to
visualise it over several octaves). The combination of modes is calculated and subse-
quently displayed using a circular representation (n-cercle—visible here at the bottom of
the patch).
Timbre, or the spirituality of sound

For a performer of Iranian music, sound quality is not a secondary consideration; rather,
it is at the very core of his or her notion of musical expressivity. Timbre allows the
performer, through the use of abstract figures, to transcend the physical reality of sound
itself; in this process the gestural ordering and evolution of these figures is essential.
Therefore the performer is expected to be acutely conscious of the sonority of his or her
instrument, and to choose from the available sonorities with virtuosity in the performance
of a given melody. The French ethnomusicologist and musician Jean During makes
reference to Henry Moore in order to elucidate this link: “Musical material presents a
striking analogy with the characteristics of the other-worldly spissitudo-spiritualis [...],
just as intangible, but nonetheless consisting of dimensions (high, low) and spacing
(rhythm), musical sound is not judged in terms of frequencies, but rather, in terms
of purity” [4].
During also compares the connection between sound quality and gesture to archi-
tecture. Ornaments, arabesques, perforations in the wall which allow light to enter,
and the use of colour (Persian blue, sky-blue) are material manifestations of the sacred.
In music, “instruments produce ethereal sounds which shimmer like haloes around the
fundamental” [4]. The timbre of the instrument, and the way in which it is controlled
by the musician, is inherently connected to metaphysical thought.
280
Figure 6. OpenMusic patch to combine two dāng, specifying the starting point for each.
281
Alireza Farhang
In contrast to European instruments, whose timbres tend toward simpler spectra,

Iranian instruments favour richer sounds.7 To give one such example: air-sounds are an
important component of the characteristic sound of the ney (a wind instrument). This
instrument, in spite of its simple, almost primitive construction, possesses a rich and
varied palette: each register has its own distinct colour.
During notes that “it is not the whistling of air in the ney, but the warm, solemn,
human breath that brings the reed to life” [4]. In saying so, he implies that the competent
musician must have absolute control over timbre, and be able to nuance it by varying
the way in which he or she breathes into the instrument. Thus, the characteristics of the
performance depend in no small part upon the ability of the musician to control timbre,
both in a technical and musical sense.
In the following section, I will discuss the modelling operations used in the composi-
tion of Tak-Sı̄m. The work’s harmonic evolution is based upon spectral transformations
of segmented samples of a performance on the setār. Each segmented sample constitutes
a “gestural unit”; the totality of these units forms the basis of the compositional material
of the work.
Form
In Iran, the close relationship between poetry and music is considered self-evident. Music
from the Persian radif tradition, which is learned through oral transmission, is from
the first moment of its dissemination associated with poetry. It contains “melodies
which originate from ancient sung-poems, often of sequences which commemorate the
ta’zieh;8 their quality is melancholic, even poignant, and their rhythm is free. In addition,
instrumental demonstrations that highlight the technique have become an element of the
radif in its own right” [3]. Instrumental music is based upon song; in order to learn
the radif by heart, instrumentalists sing the poems which are associated with a given
melody. Melodic/rhythmic figures, dynamics, pauses, the tahrir (a word describing the
particular vibrations in the throat), etc. are all closely associated with poetry, which is
itself considered to be a “perfect musical form”.
As for the notion of time in Persian culture, Jacques Le Goff suggests that Western
culture, and by extension the Western musical tradition has, since the Middle Ages,
become subordinate to the time of the marketplace, “measured, or in other words, clearly
directed and predictable, [...] both eternally re-beginning and perpetually unpredictable
from any natural perspective” [6]. For the Iranian musician, time is perceived in a
fundamentally different manner: it is not chronometric and metres are not based on
regular strong and weak beats.
7 Following the development of polyphony, for combinatorial reasons, symphonic instruments of the 19th
century tend to produce purer, quasi-sinusoidal sounds. This phenomenon has never occurred in non-
European music, or more specifically, in non-symphonic music; the monophonic character of Persian
music and the importance of timbre does not necessitate a harmonic tone. Non-symphonic instruments
(i.e. those used in rituals, battles, therapy, etc.) have a more complex sound and display a higher degree
of inharmonicity. The spectral richness of an instrument like setār or ney allows the instrumentalist to
control the timbre as the expressive parameter of his performance.
8 Ta’azieh is a sacred theatrical genre that primarily commemorates the martyrdom of Imam Hossein
(the third Imam of the Shiites).
282
Tak-Sı̄m
Two anecdotes
Having heard the Arditti Quartet perform my second string quartet, Echo-Chaos (2007),
David Harrington, the Kronos Quartet’s first violinist, professed to have detected sonori-
ties that originated from a distant musical culture. Although Echo-Chaos belonged to
a period during which I avoided using materials coming from Persian musical culture,
the influence was nonetheless present in this piece. After this meeting, I became acutely
aware of the importance of my own personal sonic-universe, of the identity of raw sound
material and its role in works of music.
One year later, following a research trip to New York, a serendipitous event which
reinforced my reflections upon musical identity convinced me to compose a work for
Kronos, which emphasised the synthesis of my reflections upon my own musical identity.
During a composition seminar at Columbia University in 2009, a young Turkish composer
suggested that we listen to an instrumental work of Turkish classical music. Having
played the work in question, he asked those present to identify the instrument that was
featured. The general consensus was that it must be some traditional Turkish instrument.
Some justified this hypothesis on grounds of the instrument’s timbre and intonation, but
no one had realised that it was, in fact, a cello. Thus, I discovered the extent to which
playing technique, intonation, intervallic content, articulation, ways of sustaining a note,
etc. characterise an instrument, its sonic image, and its identity.
Key points of the work

The string quartet is a relatively young instrumental formation but one which has
nonetheless been much used by the leading figures of the classical tradition. Composed
of four instruments but often seen as a single instrument, the string quartet, owing
to its sonic colour and its propensity for virtuosity, is a rather particular entity. The
potential for micro-tonality coupled with the morphological proximity of Western string-
instruments to the Persian setār was a key in my decision to compose this work.
In simple terms, in Tak-Sı̄m I sought to recreate the expressive aspects of Persian
classical music within the context of contemporary Western music. In order to achieve
this without only imitating musical gestures from the Persian tradition, it was necessary
to reflect at great length upon the cultural contexts associated with this sonic reservoir,
and thereupon construct a new edifice whose dimensions and characteristics would fit my
proposed work’s requirements. Once the work was complete it would be necessary for it
to transcend the cultural context in which it was conceived, and to express its aesthetic
identity in an independent, autonomous way.
In the book Le regard mutilé, Daryush Shayegan, Iranian writer and philosopher, says
that in Persian culture “anything that taps into the individual, i.e. ‘me’, gives way to a
collective whole, where the qualities of places and times remain extremely mobile” [8]. In
contrast, modern Western culture is based on the cogito of Descartes, which places the
subject in the centre of knowledge acquisition. In the Middle Ages the musical culture of
the West was already beginning to submit to this worldview. Music, whatever it is, is the
product of the human mind. If the the existence of “I” is certain because he thinks, the
mind is the centre of the perception of the universe and our place in it. Unlike Persian
music, this idea is reflected in Western music where aesthetics is based upon sensorial
factors.
283
Alireza Farhang
However, the musical universe of Tak-Sı̄m, is intimately associated with Iranian music.
In order to render the raw, initial compositional material more neutral in nature, it
was necessary to remove its overtly cultural connotations. It was therefore desirable to
deconstruct this material. To this end, one must possess a sound understanding of the
musical and aesthetic structure of Persian music.
Music for the setār

The setār is a plucked-string instrument which is played with the fingernail of the right-
hand index-finger. The back-and-forth motion of this finger causes the strings to vibrate.
Direct contact between the player’s fingernail and the instrument allows a high-level of
control over timbre and intensity. Up-strokes (rāst) and down-strokes (tchap) create
subtly different sounds. The rı̄z, or tremolo, created by a rapid to-and-fro motion of
the finger upon one or several strings does not imply a mere repetition of a given pitch;
on the contrary, the speed of this motion, its intensity and the timbre it yields are all
controlled by the performer. Techniques such as sul ponticello, ordinario and sul tasto
may also be used for further timbral variation. Changes in pitch are the result of small
glissandi, which may be achieved by varying the tension of the string in use through
vertical movement of the fingers of the left-hand, or by sliding the hand horizontally
along the length of the string, as with a vibrato, or finally by varying the pressure of the
right hand upon the bridge, thus increasing or decreasing the tension of the string.
The use of a pedal-tone, played with a drone string, is another characteristic of the
setār. The drone is typically particularly rich in timbre. Ahmad Ebādi9 developed
a technique which consists of playing on a single string whilst avoiding the drone-
string. Thus, melody and ornament are more developed in his playing style. In this
way, the technique known as tak-sı̄m, or “mono-chord” has become the most important
characteristic of the Ebādi playing-style; it was for this reason that I chose this as the
title of my work.10
Samples and their semantic content

Persian music is a tradition in which each gesture, each melody reveals ideas which go
back many centuries. These ideas persist in the Iranian collective-consciousness; this
phenomenon is divided, sacred, and timeless, and must be respected if the music is
to remain authentic (even if, as an oral-tradition, it is subject to the imperfections of
human memory upon which it depends to survive). The question was, therefore, how does
one go about decontextualising something which is so deeply engrained in the collective
consciousness of a musical tradition? Thus it was essential, as far as possible, to make all
melodic figures abstract in nature, and in doing so to lead the listener beyond a semantic
realm; the more abstract the material, the less apparent the culturally-connoted origins.
9 Ahmad Ebādi (1906-1992) is a famous Master of setār whose unique style, virtuosity, and improvisation
technique make him one of the best setār players of his time.
10 “When Ahmad Ebādi was playing in live radio broadcasts, he asked members of his entourage about
his performance. Some remarked that there was too much noise. Technical means of radio broadcasts
at the time were limited; thus Ebādi, in order to minimise the degradation of sound quality of
his instrument during radio transmissions, began to avoid the drone string and favour the tak-sı̄m
technique.” Houshang Ebtehāg, translated by the author; see https://youtu.be/Zm2_Gf2asVM.
284
The composition of Tak-Sı̄m was undertaken using a recording of the radif, performed
by Ahmad Ebādi. Samples were first segmented into simple musical phrases by ear.
Each phrase, or neume, may be considered as a large gesture, or a combination of
micro-gestures. After transcription of the samples (Figure 7), further dissection into
smaller fragments was necessary in order to isolate individual, single, “pure” gestures
(see Figure 8); this was the basis for the creation of a catalogue of gestures used as a
“reservoir” from which the work’s key material is derived.
Figure 7. Transcription of melodic fragments from setār recordings. Although rhythm is

not accurately represented, gestures are illustrated meticulously by note heads and graphical
symbols.
285
Alireza Farhang
Figure 8. Typology of gestures represented by graphical symbols.
Figure 9 shows a neume, a transcription of a fragment from a melody in mode of

segāh,11 which is used in the modelling process of Section III of Tak-Sı̄m. This fragment
was selected, analysed by ear, and transcribed onto paper. I will describe the processes
applied to this neume, from the moment of its selection to its ultimate representation
by the quartet. As the reader may see, the complexity of this music that is expressed
though timbre, microtones, fluctuation of pitch and dynamics, strokes of the fingernails,
tapping etc. gives rise to particular readings of the fragment. The upward-stemmed
notes represent the primary notes played with the fingernail. The lower voice represents
ornaments that the performer plays by changing pressure in the right hand on the bridge
and/or vertical movement of the left hand finger.
In order to create the work’s harmonic texture, it was necessary that these ges-
tures/samples respond to various forms of analysis, particularly melodic transcription and
analyses of spectral-content. Furthermore, it was necessary to select the most interesting
and useful neumes and disregard the rest.
Figure 9. A neume can be seen as a gesture or a combination of microgestures. This fragment

is an excerpt from the first volume of the CD Setār solo par Ahmad Ebādi, in the mode of segāh.
11 Segāh is one of the modes of radif. The intervallic structure of segāh is similar to shour.
286
Melodic and spectral analyses

In order to observe the behaviour of the energy envelope of a neume we must analyse it
from different points of view. Of course the boundary between timbre variations, pitches
and dynamics are often not well defined; in this sense timbre and melody are interrelated
and variation of one parameter often alters the perception of another.
In the transcription in Figure 9, the pitch range of the neume is limited to a tritone
diminished by a quarter-tone, and it contains micro-glissandi or pitch-bends. The neume
has been transposed using the software AudioSculpt in order to suit the harmonic
structure of the quartet determined beforehand. In Figure 10 we see the dynamic
evolution (amplitude of the waveform) of the neume.
Figure 10. Dynamic profile of the neume segāh.
As mentioned previously, it is important to consider that melody and timbre are

interdependent. Thus the neume was then subject to two analyses, one melodic and one
spectral. Using AudioSculpt, I first analysed the segment’s fundamental frequency.
With harmonic sounds, our perception of melody depends largely upon evolution of the
fundamental frequency, whether real or virtual.12 The output of this analysis is imported
(as a text file) into OpenMusic where it is subjected to treatments and transformations,
then turned into an editable envelope (Figure 11).
Figure 11. The melodic contour is rendered visually using a bpf. Subtle variations in pitch
may therefore be tracked here.
In order to extract spectral information from the sound, it was necessary to deter-
mine the precise moments which were—from a spectral point of view—of most interest.
Once again using AudioSculpt, in this case the chord-sequence analysis and transient
detection functions, I generated a series of markers that identified moments of spectral
contrast. This process requires considerable fine-tuning (in terms of setting the variables
appropriately for the sample in question) in order to avoid an analysis containing digital
artefacts.
12 Infact we detect the pitch or fundamental frequency via the frequency differences between the
components of a harmonic sound rather than via a weak or sometimes absent fundamental frequency.
287
Alireza Farhang
In Figure 12 we see the sonogram or spectral-content and intensity of the aforemen-

tioned neume. A series of “chords” corresponding to the number of transient markers
was then generated by AudioSculpt and the results of this analysis (describing the
frequencies of partials present within each marker interval, as well as onsets between
them) were exported as an SDIF file. Using OpenMusic, this file was then interpreted
and plotted onto a musical staff (see the chord-seq in Figure 13), thus allowing me freely
to manipulate the analysis before any further treatments.
Figure 12. Sonogram of the neume (above) and its intensity (below).
Figure 13. The spectral content of the neume represented as a series of chords.
288
Harmonic texture
Harmonic texture in Tak-Sı̄m is also derived from chord-sequence analyses. Both the
instrumental and electronic content of the piece were based upon a harmonic grid created
from a sequence of chords obtained using this method. As the degree of inharmonicity of
the setār is considerable, the chords obtained were accordingly inharmonic (in a psycho-
acoustic sense). Nonetheless, the inharmonicity alone was not enough to yield a harmonic
progression.13 To this end, it was necessary to “enrich” the spectral-content of the chords
using a simple algorithm which added a series of notes to those already present in the
analyses (see Figure 14). In order to facilitate the composition of the instrumental
material, the aforementioned chords were arpeggiated and printed. Figure 15 illustrates
the chords in their original form, and Figure 16 represents them after application of the
process of harmonic enrichment; the latter was ultimately used for the Section V of the
work.
Figure 14. In this patch a simple algorithm adds a series of notes to those already present in
the analyses.
Beyond manipulating the degree of harmonicity of the chords obtained in the process
just described, other procedures were used to obtain the harmonic texture of the piece:
inversion, change of range or density, interpolation between two or more chords, and
transposition. Figure 17 shows a patch used to invert a chord while applying an algorithm
that increases the spacing between the components (while keeping adjacent pitches less
than or equal to one octave apart). Through this process, although the new chord is
wider, the texture remains close to that of the original chord but increases its range. The
pivot note represents the centre of inversion (here F), and all other components of the
chord are inverted around it. Note that in the inverted chord, the intensity of each note
is identical to that of the corresponding original note.
13 Thedissonance-consonance rule of tonal music is perfectly valid in music whose harmonic structure is
based on timbre. This creates tension-relaxation, and thus a sensation of movement.
289
Alireza Farhang
Figure 15. A harmonic grid created from sequences of original chords.
Figure 16. A harmonic grid with an increased rate of inharmonicity.
290
Figure 17. Inversion of a chord around a pivot note.
291
Alireza Farhang
Harmonic interpolation is another process employed in the piece as a means to

gradually transform harmonic textures. In Figure 18 we see a starting and ending chord
and an interpolation process taking pitches from a “grid” derived from the harmonic
texture of the piece. In the work interpolation often occurs in the form of a long tremolo,
and the whole process of interpolation can be considered to be a single tremolo gesture
(see mm. 12-18 of the score in Figures 19 and 20).
Figure 18. Patch for harmonic interpolation.
292
34
4
12
÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷
. . œ.
œ. œ. œ. œ. œ. œ œ
1

. . œ.
1-8-2
œ. œ. œ. œ. œ. œ œ
3

1-8-1
4
 4 4
œ œ
 1-9-1 
œœœ œœœ
œœ œœ
interaction with dynamics of cello
1-8-3
con legno battuto
# œ# œ
arco ord.
– –. –. –. –. –. –. –. –. –. –. –. –. –. –. –. –. œ# O
s.p. s.t. s.p. s.p. ord.
& (–) ––––––––––––––––

#O#O OO ˜œ
Vln. 1
æ
OO
œ˜œ O
f P p f
arco ord.
s.p.
œ
s.p. ord.
æ Bœ œ #O O
#O #O
O#œ O O O#O#O#O
#O O O
& ( –– ) O O
O O O œ #O
Vln. 2
OO
f F p f
con legno battuto arco ord.

#O O
# O O µµ Oœ O Ÿ~~~~~~~~~~~~~~~~
s.p. s.t. s.p.
O# œ
s.p. ord.
O O
B–––––––––––––––– O # O III– IV
Vla.
–. –. –. –. –. –. –. –. –. –. –. –. –. –. –. –. O
O O
O O # œ O (O)
F p
f
? æO
s.p. ord.
Vlc.
(– )
f p
Figure 19. Harmonic interpolation (1): Tak-Sı̄m mm. 12-14.
293
Alireza Farhang
8 88 6
410
5
15 4 e = 72 q = 60
4
÷ ÷ ÷ ÷ ÷÷÷ ÷
• Samplor
÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷ ÷÷÷÷÷÷÷÷÷÷ ÷ ÷ ÷ ÷ ÷ ÷
œ. œ. œ. œ. œ. œ. .
2
œ. œ. œ. œ. œ. œ. œ.

œ œ.
3

1 4 1
œ
   5-1
œœœ 3-1
œœ
2-1 3-2
. . 3˘ . ord. (h )
# # Oœ ‰
3
r # œ # # Oœ
con legno tratto
œ #œ #O O
tremolo arco
& – ‰ O O O œO –
II III
Ȯ
IV
Vln. 1
æ # O O æ
" J J
p P p P ! F
.
œ #œ œ (q )
ord.
(h )
#O #O O #O #O O
& # œ œ OO OO OO
#O #O
O O O
Vln. 2
" P " P !F
˘
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ . . #œ Ÿ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
œ. # œ œ RÔ ® #
ord. vib.
1/4 ! con legno tratto
B – & Œ # Ȯ # œO B
–O ( O )
Vla.
O
p F P ! F
3
±>
s.p. s.t.
s.t. s.p.
? O Oæ
arco ord.
O
con legno tratto
J
Vlc. O – O
æ æ
ƒ " P "
Figure 20. Harmonic interpolation (2): Tak-Sı̄m mm. 15-18.
294
Sound synthesis
Except for some real-time processes, almost all of the electronic sounds used in Tak-Sı̄m
are made with additive synthesis. In order to build a dialogue between the electronic
sound and the expressive instrumental sound of the string quartet, all the materials
generated by computer must be sculpted carefully. Although sinusoidal waves, as the
most basic component of sound, offer a large palette of possibilities for creating new
colours, the lack of the human musical expression in synthesised sounds is a major concern
for many composers. In Tak-Sı̄m, modelling an expressive gesture generated by a human
performer is an efficient method to make the electronic sound more expressive.
Figure 21 shows the patch designed to control the electronic synthesis in Tak-Sı̄m.
This patch incorporates of five types of control: a harmonic grid (1 and 2), tremolo-like
gestures (3), and melodic gestures (4 and 5). It integrates most of the processes for ma-
terial generation and transformation described in the previous sections, and constitutes
the first level of synthesis control (generation of pitches, amplitudes, and onsets of the
oscillators of the additive synthesiser).
Figure 21. Control-patch for sound synthesis.
In Figure 21 an inverted chord derived from the analysis of the neume segāh provides
material for the synthesis process. Sections 1 and 2 of the patch are similar; however, they
generate two different types of sound. The parameters in the corresponding synthesis
processes are set in order to produce a sound with more vibrato and beating for the first,
and a dry and steady sound for the second. The global dynamic envelope of the sounds
is drawn in a bpf (labelled “wave shape”).
Along with melodic gestures, tremolo, vibrato, and beating are the gestures that
are most often used in the quartet. Section 3 generates a tremolo-like gesture via an
algorithm. A global pitch (freq), dynamic (amp) and tempo can be controlled using three
bpfs. The patch tremolo produces a series of repeated chords, which may be plotted
visually with a chord-seq (see Figure 22). The precise content of this data may then be
manipulated by hand where deemed necessary by the composer.
295
Alireza Farhang
Figure 22. Tremolo-like gesture generated by an algorithm implemented in section 3 of the

patch for synthesis control.
The melodic contour of the neume segāh, imported from AudioSculpt, is edited and
modelled in the section 4 of the patch. It is “coloured” with a group of pitches derived
from the aforementioned harmonic grid. The proximity of timbres between this group of
pitches—represented here as a chord—and the harmonic grid, as well as the morphologic
similarity between instrumental gestures and modelled gestures—all derived from the
neume segāh, make the emergence of the electronic part sound natural.
Finally, section 5 of the patch generates vibrati that are used to extend the neume.
Figure 23 represents different methods for generating these vibrati in OpenMusic. Du-
ration and frequency deviations are among the parameters which allow one to design a
whole new gesture from the data extracted from the previous analyses. Depending upon
the context, beating and vibrato may be considered gestures in and of themselves.
Figure 23. Implementation of vibrato effects in OpenMusic.
All of the control data are then transferred to a synthesis patch that generates five
sounds (see Figures 24 and 25). For each of the five sounds it is possible to define a global
shape using bpf objects (see Figure 25). For instance, a sound can start quickly and
fade out slowly, which is suitable where a short attack with a long release is necessary.
When a sound has a long attack it is more suitable for a harmonic nap or grid. At
the top of the figure, the sub-patch titled sounds defines the actual timbre of the sound
by controlling the components of the sound synthesis process (see Figure 26). In fact,
this patch instantiates synthesis classes from the OMChroma library: the choice of a
particular class determines the contents of the Csound instrument that will be used for
synthesis.
296
Figure 24. Sound synthesis (1): this patch takes the data from the first-level control patch
(Figure 21) and outputs five different sounds (see Figure 25).
Figure 25. Sound synthesis (2): the sub-patch om-composers_synthesis, shown in Figure 24.
In this patch the global shape of the sounds is editable.
297
Alireza Farhang
Figure 26. High-level control of synthesis using the OMChroma library.
Conclusion
As a composer who is cognisant of the centuries-old tradition in which he is situated,
I feel that the act of composing is one of projecting, onto paper, personal and unique
musical thought. However, a composer’s originality may also come from the fact that
he or she is affected by the sensorial and spiritual stimulus of an external event. On
the one hand, as Hegel maintains, “a work of art is superior to anything which exists in
nature” [5]; on the other, Persian musical thought dictates that the highest form of art
is nature itself, and all works of art are merely incomplete imitations of nature.
In the West, composition is purely an intellectual pursuit. What makes it, in Hegel’s
view, superior to nature is the spiritual content the composer instils within his or her
musical world in order that the work might demonstrate originality. Such a notion is
incompatible with the process by which a musician playing in an Eastern tradition sets
out to create music. For the traditional Iranian musician, a supreme work of music exists
only beyond the capacities for human perception. It belongs to an eternal and immutable
centre which is indifferent to conditions in the material world. The role of the performer
is therefore that of intermediary, albeit one who is by necessity equipped with ability
and knowledge, between the metaphysical and the real worlds. It is therefore necessary
to draw a distinction between composer (in the Western sense) and musician (in the
298
Eastern sense); the latter is understood to be a composer, improviser, and performer.

The Western composer is charged with the appropriation of his or her own musical
universe, in order that he or she may derive some new melodic, harmonic, or rhythmic
figuration from it.
The technical challenge of confrontation, between an ancient non-European musical
tradition and an opposite domain of thought—i.e. musical composition, makes Tak-Sı̄m
a key point in my artistic path. Paradoxically the process of composition of the piece
would not have been possible without new technologies.
References
[1] Abd al-Qāder Mohamadtaghi Binesh b. Ghaybı̄ Marāghi. Maqāsed al-alhān. Tehran:
Bongāh-e tardjomeh va nashr-e ketāb, 1978.
[2] Joël Bastenaire. “Panorama des musiques savants et populaires d’Iran”. Écouter
Voir, 130, 2002.
[3] Nelly Caron, Daryush Safvat. Iran. Les tradition musicales. Paris: Editions Buchet-
Chastel, 1966.
[4] Jean During. “The ‘Imaginal’ Dimension and Art of Iran”. The world of music,
19(3-4), 1977.
[5] Georg Wilhelm Friedrich Hegel. Aesthetics: Lectures on Fine Arts. Oxford
University Press, 1831/1975.
[6] Jacques Le Goff. “Au Moyen Age : temps de l’Eglise et temps du marchand”.
Annales. Economies, Sociétés, Civilisations, 15(3), 1960.
[7] Tristan Murail. “Questions de cible”. In Pierre Michel (ed.) Modèles et Artifices.
Presses Universitaires de Strasbourg, 2004.
[8] Daryush Shayegan. Le regard mutilé. Paris: Albin Michel, 1989.
[9] Daryush Talāı̄. A new approach to the theory of Persian art music. Tehran:
Mahour Cultural Foundation, 1993. Translated from Persian Negareshi no be teori-e
mousighi-e irāni.
Acknowledgements: I thank Paul Clift for his translation of this text. Without his generous help this
chapter could not have been published.
299
Electronic dramaturgy and
computer-aided composition
in Re Orso
Marco Stroppa and Jean Bresson
Re Orso (King Bear) is an opera merging acoustic instruments and electronics.1 The
electronics were realised at IRCAM with the assistance of Carlo Laurenzi. The libretto,
written by Catherine Ailloud-Nicolas and Giordano Ferrari, is based on a fable by Arrigo
Boito.
Every moment of the opera is exclusively inspired by and tightly related to the
prescriptions of the libretto and the intimate structure of the drama: there are no vocal,
instrumental or electronic sounds that do not have a deep connection to and a musical
justification in the dramaturgy. In addition, an important compositional objective was
that the electronic material be endowed with a clear dramaturgic role, in order to be
perceived as a character on its own (actually, several characters) with a personality that
develops during the action.2
Preceded by a short exordium, Re Orso is divided in two parts of approximately 45’
(five scenes) and 30’ (three scenes) separated by an intermezzo storico. The ensemble
leaves the pit at the end of this first part and the singers remain alone with the accom-
paniment of electronic sounds. Voice and electronics are therefore essential elements
of the dramaturgy and of the composition. Both have been written and organised
with computer-aided compositional tools. This text explores some of the representative
OpenMusic patches developed for this project.
1 Commissioned by Françoise and Jean-Philippe Billarant for IRCAM, the Ensemble Intercontemporain,
the Opéra Comique in Paris, La Monnaie in Brussels, and a French Commande d’Etat. It was premiered
at the Opéra Comique in May 2012.
See http://www.opera-comique.com/fr/saisons/saison-2011-2012/mai/re-orso
2 For this reason the word “electronics” does not appear in the subtitle of the opera, and is replaced by
its respective roles: Re Orso, Musical Legend for 4 singers, 4 actors, ensemble, invisible voices and
sounds, spatialisation, and acoustic totem.
301
Figure 1. Re Orso. Opéra Comique, Paris (May, 2012). Photos: Elisabeth Carecchio
Compositional environment
Re Orso is organised around a “compositional environment” consisting, at the most basic
level, of 72 “formant chords”3 and 3 scale structures,4 as well as of several rhythmic cells
and dynamic profiles. The music deals with this environment according to a panoply of
rules that will not be analysed here.
This pre-compositional environment is used to generate the basic musical structures
of the piece, which the composer calls Leitgestalt. A Leitgestalt is defined as a cognitive
morphology [11] playing a dramaturgic role; that is, an acoustic and theatrical event
that impinges upon our cognition and hence can be recognised in other parts of the piece
when it comes back, even if it is transformed. For instance, a downward glissando always
means a kind of fall (moral, physical, psychological, etc.), large or small depending on
the extent and details of its realisation.
Re Orso brings a variety of electronic materials into play:
• Imaginary voices (FOF-based synthesis using the OM-Chant library [3]);
• Imaginary sounds (synthesised using the OMChroma library [1]);
• Imaginary instruments (physical modelling synthesis using Modalys);
• Imaginary ensemble of the intermezzo storico (recorded and remixed ensemble);
• Mysterious voices (recorded voices processed with AudioSculpt’s cross-synthesis);
• Mechanic doubles of the two buffo roles:
– Hand-controlled puppet (vocal sounds processed with SuperVP-Trax),
– Computer-controlled, singing Disklavier;
• Real-time processing of the main female role’s voice (Chromax spectral delay [5]);
• Overall coordination and concert setup (Max with Antescofo).
3 Chords algorithmically derived from databases of formant data (see footnote 11) transcribed from sung
vowels (5 or 8 formants) approximated in a chromatic space. In total, Re Orso uses a data base of 44
5-note chords and 28 7-to-10-note chords.
4A “scale structure” is a succession of absolute pitches that encompasses the playing range of the
ensemble. These structures are based on some intervallic and symmetric properties.
302
Electronic dramaturgy and computer-aided composition in Re Orso
The voices (4 singers and 4 actors) and instruments are amplified.5 The sound projec-
tion uses an 8-channel frontal setup (3 loudspeakers placed on the stage, 3 loudspeakers
hung above the stage, plus one hidden below the stage and one oriented downward from
the ceiling) and an acoustic totem, a column of eight loudspeakers that appears from
below the floor in the middle of the stage at the end of the piece.
OpenMusic has been used both for writing instrumental parts and for the generation
of electronic sounds. In particular, the OM-Chant library allowed the connection of
formalised compositional processes to vocal sound synthesis using the FOF technique
and the Chant synthesiser.
Sound models
The sound model is a fundamental concept in Re Orso as well as in most previous works
by Marco Stroppa. It is, with respect to the composition of sound, something similar to
the idea of Leitgestalt, or more precisely, to the concept of Musical Information Organism
developed by the composer in the 1980s for his musical material: a group of sounds
“that consists of several components and properties of varying complexity, maintaining
certain relationships and giving rise to a particular form” [9]. This concept implies that
different sounds corresponding to the same model can be recognised as possessing the
same identity and, therefore, can express the same sonic potential [10]. Technically, this
concept of sound model is implemented in the OMChroma library as a class (called
cr-model), which aims at facilitating the generation of families of sounds [4].
Control of additive synthesis

One on the most used Leitgestalten in the opera is the “phantom” of a bell. A represen-
tative example of it is the series of synthetic bell-stroke sounds that can be heard at the
end of the intermezzo storico. Dramaturgically, these sounds introduce to the audience
the increasing fear of death in the King’s mind. The dark sound of the the great tenor
bell at Winchester Cathedral was taken as a reference and used as a starting point for
the generation process.6 Through a sequence of strokes, the timbre of the synthetic bell
becomes increasingly dark and more ominous.
The OpenMusic patch shown in Figure 2 implements this process. The starting
material is the spectral analysis of a bell-stroke recording, imported as an SDIF file at
the top left of the patch. This file contains a chord-sequence analysis, performed with
AudioSculpt software, based on time-markers hand-positioned in the spectrogram in
order to parse the different states of the spectral evolution of the sound.
The segmentation and spectral analysis data are used to instantiate the cr-model
object labelled “model 1” whose contents are shown at the right. This model provides
a reservoir of time/frequency structures that can be processed in the time and frequency
5 The amplification has three main purposes: first, it allows for a better fusion between the acoustic
sounds and the electronic materials; second, it gives the stage director total freedom of movement
on stage, without risking that the singers are not heard during a tutti. Finally, used as an acoustic
microscope, it can bring to the audience softly whispered sounds that would otherwise be inaudible.
6 Thisis an homage to Jonathan Harvey’s Mortuos plango, vivos voco, which uses the same bell at the
beginning of the piece.
303
Figure 2. Patch synthesising the bell strokes played at the end of the intermezzo storico.
domains (leading to another cr-model labelled “model 2”) and finally converted into a
set of sound synthesis control parameters.
The expand-model function performs the conversion of the successive segments from
model 2 into a list of instances of a given class of “synthesis event” from the OMChroma
library (in this case, the class add-1 is used, which corresponds to a simple additive
synthesis). The function therefore returns one add-1 instance for each segment in the
input cr-model. Along with the cr-model, the second argument of expand-model is a patch
(cr-control-patch), which determines the details of the mapping performed between the
cr-model segments and the sound synthesis parameters.
The editor window of cr-control-patch is also visible in the figure. In this special
patch, the synthesis parameters (or “slots” of the class add-1 ) are represented by the
out-slot arrow boxes at the bottom. The mappings here are of several kinds:
• The number of “components” of each synthesis event (numcols—in this case with
the class add-1, the components corresponds to partials in the additive synthesis
process), as well as their respective amplitudes (amp), are derived from the two
“standard” patch inputs.
• The durations (durs) and frequencies (freq) of the components are determined
according to programmed relations with the data contained in the cr-model segment
(represented by the model-data box at the top).
304
• The amplitude envelopes of the components (aenv) are not dependent on any
external data and are specified statically by the bpf visible at the bottom right of
the patch.
The expand-model function applies this cr-control-patch iteratively, updating the
model-data at each iteration with the values from the current segment, and producing a
new instance of add-1.7 The synthesize function at the bottom of the figure collects this
list of add-1 instances and runs the actual synthesis process. For each state of the spectral
evolution of the original stroke sound (that is, for each segment of the cr-model, and each
iteration within expand-model) a new, increasingly dark bell stroke is synthesised.
Abstract modelling and gestures in the Disklavier part

At the beginning of the 5th scene (Part 1), the dramaturgy calls for both a Troubadour
and his mechanical double (a Yamaha Disklavier—computer-controlled robotic upright
piano) to sing two arias. Making a piano sing may look like a utopian challenge: while
examples of a speaking piano were realised in the past with special hardware added to
the instrument,8 to our knowledge nobody had yet made this instrument sing on stage.
The notion of sound model, previously used to parameterise sound synthesis, was
applied here to generate a score for the Disklavier. The classical approach consisting
of analysing sounds to derive instrumental scores, developed and thoroughly explored
by spectral composers, can be advanced further when combined with symbolic process-
ing in computer-aided composition programs. The process was long, painstaking, and
somewhat tedious but yielded a spectacular result on stage.
To proceed, the composer chose to use the popular aria La Donna è mobile from
Verdi’s Rigoletto. Several performances were compared until one was found (Alfredo
Kraus, 1979) whose spectral analysis was the most interesting for this purpose. As a
Disklavier cannot have the same expressive freedom as a real singer, the first task was
to “straighten” the metre of the original recording by superposing an audible beat to it
as a reference (♩ = 138), and by time-stretching some parts of the audio so that they
adequately fit with the downbeats. In total, this extended version contained 36 measures
at 3/8 and one at 4/8, whereas the original had 34 measures at 3/8. OpenMusic was
then used to generate markers (six per pulse, resulting in a density of 13.8 attacks/sec).
The AudioSculpt chord-sequence analysis of the transformed sound performed using
these markers identified a maximum of 60 partials per time segment (see Figure 3). At
this stage, approximately 45000 notes were generated, which was not only too much for
the Disklavier, but also still sounded too mechanical.
Empirical testing showed that a density of 10 notes per pulse was the upper limit that
a Disklavier could reasonably afford to play without becoming unstable, and eventually,
crashing. Given the amount of data contained in the original analysis, a manual data
reduction would have been extremely inefficient. Programs were therefore written in
Lisp and used in OpenMusic to perform dynamically evolving processing and filtering
over the whole sequence and to confer a more human character upon the instrument (see
Figure 4).
7 Infunctional programming a function like expand-model, taking another function as one of its
arguments, is called a higher-order function.
8 See for instance Peter Ablinger: http://ablinger.mur.at/speaking_piano.html
305
Figure 3. Zoom on 1s of the chord-sequence analysis of La Donna è mobile in AudioSculpt.
Figure 4. Processing a sound model to generate a score (MIDI file) for the Disklavier.
306
The principal processing routines implement the following operations:

• Approximate the data to semitones and eliminate repeated notes.
• Cut out pitches above the range of the piano keyboard.
• Eliminate pitches below G3 so as not to take into account possible analysis errors.
• Underline the difference between f and pp using higher registers for the f passages.
• Intensify the impression of a crescendo by gradually modifying the number of notes
in the piano’s upper octave and a half (i.e. during a crescendo, the higher notes in
this range appear progressively).9
• Add some randomness (i.e. select keys around the original ones) in the higher pitch
range. This allows deviations from the original B-major tonality, but also avoids
too many successive repetitions of the same keys—which can be dangerous for the
Disklavier.
The final MIDI editing was done with Digital Performer (see Figure 5).
Pedal automations are added (una corda or mute), especially in the pp passages. This
not only changes the sound, but also creates a theatrical effect, as the piano on stage has
no lid and viewers can see the entire mechanics shifting closer to the strings.
Figure 5. MIDI file of the beginning of the first song of the Disklavier in Digital Performer.
Notice the spectral difference between f passages (e.g. in the first four 3/4 bars, with more notes
in the high register) and pp passages (e.g. bars 5 to 8, with reduced pitch range), as well as
the pedal controllers: short sustain-pedal strokes (represented by the dark rectangles) in the
f passages, followed in the pp passages by a combination of una corda (wider rectangles) and
shorter sustain-pedal strokes.
9 Itis known that the auditory perception of a high sound dynamic is accompanied not only by a higher
amplitude, but especially by a wider spectrum. In addition, due to the semitonal approximation, the
higher part of the spectrum tends to sound like a chromatic cluster; so filtering out some notes in the
soft passages proves to be helpful musically.
307
Imaginary voices and the control of Chant synthesis

Computer-generated singing voices are a major component of the dramaturgic material
of Re Orso. Their implementation in OpenMusic was probably the greatest challenge
and most innovative research carried out for the opera. They have been realised using
the Chant synthesiser [8] controlled by the library OM-Chant.
The composer had long been attracted by the musical expressivity and dramatic
potential of the early examples of sung voice synthesis that had been developed at IRCAM
in the 1980s using Chant and its control environment Formes [7] (like Jean-Baptiste
Barrière’s Chréode, or the voice of God in Harrison Birtwistle’s The Mask of Orpheus,
realised by his assistant Barry Anderson). Today most of this technological framework
has disappeared with the evolution of computer environments. This project allowed us
partially to reproduce in OpenMusic some of the temporal and spectral processes that
were devised and used at that time.
The Chant synthesiser

The Chant program was initially developed for generating realistic singing voices, al-
though other original sounds based on this voice simulation model could also be synthe-
sised with it. In this program, sounds are produced by a number of parallel FOF gener-
ators10 and filter modules. A FOF generator outputs a periodic train of finely enveloped
sine-wave grains, producing the effect of a vocal formant in the spectral domain.11 The
FOF parameters (frequency, amplitude, attack, release time, etc.) determine and control
the frequency, amplitude, and shape of the formant, while the period of grain generation
determines the fundamental frequency of the output sound.
During the synthesis process, a Chant “patch” (specific configuration of different
available units or “modules”—FOF generators, filters, etc.) runs continuously and its
parameters are modified by external controllers. In the latest version of the synthesiser,
these controls are specified via files encoded in SDIF format, where the values and
evolution of the parameters are set and stored as time-tagged frames. The parameters’
changes and the state of the synthesiser are not necessarily set synchronously at every
point in time: Chant performs an “off-line” rendering and systematically interpolates
between user-specified values of the parameters at a given, global control rate. Between
the specified values, linear interpolations produce smooth continuous transitions.
This is an original approach to sound synthesis, as the control “primitive” is not a
single event (what we might call a “note”) like in most existing systems. In Chant the
succession of states, smoothly connected to each other, generates monophonic sounds
allowing for subtle expressivity and the control of legato details that are needed for
realistic singing-voice synthesis (a continuous paradigm that can be related to the notion
of “phrase”).
10 FOF = Fonction d’Onde Formantique, or Formant Wave Function; see [6].

11 Aformant is an amplitude modulation of the spectrum at a specific frequency and with a specific
shape and bandwidth, characterising the observation of voiced sounds.
308
The OM-Chant library

OM-Chant provides tools to format control structures and create SDIF files adapted
to the control of Chant, as well as utilities to facilitate setting Chant parameters
(database of formant values, implementation of predefined control rules, etc.). As we
mentioned previously, the control can be arbitrarily distributed in time, either very
sparsely (in which case the synthesiser will interpolate between the specified values),
or precisely sampled, describing fine evolutions of parameters via processes of arbitrary
complexity.
OM-Chant inherits from Chant “continuous” control paradigm, but also provides
discrete timed structures (called events). Following the model of the OMChroma library,
a number of synthesis event classes are defined, which correspond to the different available
modules and controls in a Chant synthesis patch: FOF banks (class ch-fof ), fundamental
frequency (class ch-f0 ), filter banks (class ch-flt), etc. An instance of a Chant class is
an array or a matrix of values that determines the evolution of the corresponding module
parameters over a given time interval. Continuous evolutions can be specified for all or
part of the synthesis parameters during this time interval: auxiliary tools are provided
to generate such evolutions (e.g. the vibrato that may occur within the time intervals of
the events), or to control transitions between successive or overlapping events [2].
This consideration of timed events together with the description of their morphologies
and articulations in a phrase combine abstract musical constructs and “continuous”
specifications, drawing an intermediate path where continuous control can be associated
with powerful and expressive time specifications.
Synthesis of vocal sounds in Re Orso

One demand of the libretto was the invention of imaginary voices: sounds that can be
recognised as vocal, but that no human voice could ever produce. Several families of
imaginary voices were created, among them:
• Phonemes comprising the succession vowel/consonant/vowel;
• Messa di voce sounds (crescendo followed by a diminuendo);
• Ethereal voices, containing formants wandering around in the spectrum;
• Humorous voices in the coloratura register.
Vocal sounds generated with OM-Chant appear in various passages of Re Orso. The
patch shown in Figure 6, for instance, generates a sequence of sounds corresponding to the
first time the King hears the Worm’s voice, in the second scene of the opera. Two main
objects (or events) are instantiated in this patch: ch-fof and ch-f0. The combination of
these two objects determines the set of parameters required for the control of a bank of
FOF generators (formant values and fundamental frequency).
On the left, the ch-fof values (formant frequencies, amplitude, bandwidths, etc.) are
determined according to a chosen vowel (e.g. soprano “o”, alto “e”, etc.) thanks to
the vowel-formants utility. A set of “rules” is then applied, such as autobend (which
shifts the first and second formants’ central frequencies as a function of the fundamental
frequency), comp-formants (which adds a resonance in the lower region of the spectrum),
or autoamp (which automatically adjusts the relative amplitudes of the formants). Other
control parameters used are directly derived from the patch inputs, such as <win>,
309
<wdur>, <wout> (which determine the shape of the envelope of the FOF grains), or
<bw-scale> (an adjustable formant bandwidth scaler).
At the right of the figure the ch-f0 object controls the fundamental frequency over
the duration of the sound synthesis process. The base pitch is here modulated (using
param-process) by both vibrato (controlled by frequency and amplitude envelopes) and
jitter (aleatoric perturbation, also controlled by frequency and amplitude values).
The synthesize function at the bottom of the patch collects a list of Chant synthesis
events (in this case, one ch-fof and one ch-f0 ), formats an SDIF control file, and performs
an external call to the Chant synthesiser in order to produce the sound file.
Figure 6. Synthesis of a messa di voce with OM-Chant. Imaginary voices (Part I, Scene 2).
This patch is actually quite generic, and leaves many open parameters (input arrow
boxes in Figure 6) that are set in Figure 7 in order to produce a series of different sound
files. This series of sounds implements a progressive evolution from “realistic” sounds
(formants and fundamental frequency in the range of human voices, natural vibrato
curves, etc.) to more synthetic/imaginary ones, all produced by the same program.
310
Figure 7. Iterative call of the patch from Figure 6 with variable set of control parameters.
The last sounds and set of parameters in Figure 7 are of particular interest. Very low
fundamental frequencies, on the order of a few Hertz, are perceived as trains of impulses
rather than pitches, while the short attacks and growing resonance times of the FOF
grains produce the effect of percussive, almost bell-like sounds.
In FOF synthesis the control of the formants’ bandwidths (and their shape in general)
is important for the perceived result. At the limiting case, an infinitely narrow bandwidth
(0Hz) is a simple partial (sinusoidal signal), and the FOF synthesis can then produce
spectra resembling additive-synthesis sounds. We will get back to these interesting
properties in our last example.
An example of “phrase” generation:

From a cell phone to Queen of the Night
At the very beginning of the opera, the ring of a mobile phone is heard through the
loudspeakers followed by a message asking the audience to turn their mobile phones off,
as is often the case before a performance. However, this time the message is pronounced
live by one of the singers (the Worm) in the language of the country where the piece is
performed. This indicates that the opera has already started, albeit unconventionally.
Just after the announcement the cell phone rings again and gradually turns into an
increasingly eccentric voice (crazy melodic contours, fast tempo, extreme range, and
exaggerated expressive inflections), until it hints at some passages of the Queen of the
Night synthesis, a famous simulation realised by Yves Potard at IRCAM in the early
eighties. It is a humorous start for the piece; let’s examine it more closely.
The first step was to find the right ring. The one selected was a rapid alternation of
a major third (D6/B[5), which had several advantages:
• Sonically simple, yet rhythmically lively.
311
• It contains both the pitches D (Leitgestalt of the King, D=Re=King) and B[ (which
is in the pitch-class range of the Leitgestalt of the Worm, A4-C]5). However, both
pitches are located one octave higher, that is, in the vocal range of the King’s
coerced wife (Oliba, high soprano). A lot of theatrical meanings for a simple ring!
• The range and vocal speed were compatible with those of the Queen of the Night
sound synthesis process.
The patch generating one of the sounds played in this passage is presented in Figure 8.
The omloop named notes-to-chant contains a patch similar to the one in Figure 6.
Instead of synthesising a sound at each iteration, it generates timed Chant events (ch-
fof and ch-f0 ) corresponding to the short tones of the cell phone (at the beginning) or
to slightly longer sung vowels with an expressive vibrato (at the end). The collected
sequence of events is synthesised as a whole phrase (outside the omloop). Automatic
interpolations are therefore performed between the successive Chant parameter changes,
which produce smooth progressions between the distinct sound features, and implement
the gradual embodiment of the initial ring into a human voice. While the short tones
sound quite synthetic and mechanical at first, as soon as a vocal-like vibrato is added
they suddenly acquire more organic and natural characteristics.
Figure 8. Generating a sung sequence (or “phrase”) from a list of Chant events: cell-
phone/queen-of-the-night at the beginning of Re Orso.
312
Integration of sound models and continuous voice

transitions: “Death of the King’s Voice”
Re Orso dies at last. But the composer wanted to extend the death of this character
to encompass the death of his voice as well. How can a voice (as a sound) die? The
passage “Death of the King’s Voice” summarises a number of central notions from the
compositional strategies described in this text and in the opera in general. It is realised
through a single FOF synthesis process lasting one and a half minutes. Figure 9 shows
the main patch to generate this sequence.12
Figure 9. Main OpenMusic patch generating the “Death of the King’s Voice”. The contents
of the framed boxes are detailed in Figures 11–15.
12 Inthis and the following figures the patch has been simplified for the sake of clarity. The presented
values are also not the exact ones used for the sounds played during the opera.
313
The resulting sonogram (Figure 10) reveals a number of interesting details of this
process. The fundamental frequency of the FOF generator starts at a high D (D5, 585
Hz) and plunges into sub-audio values (0.3 Hz, that is, one stroke each 3.3 s). This
fundamental frequency is easy to track (or hear) at the beginning, and then disappears
in the low register. The “plunge” was structurally divided into octaves (D5-D4, D4-
D3, etc.). Within some of these octaves (D4-D3/tenor register, and D2-D1/deep bass
register) a vibrato was applied, conferring a strong vocal connotation to the glissando in
these selected parts of the sound.13 At the same time five vocal formants, that are clearly
visible at the beginning, seem to collapse into independent strokes from approximately
40” after the beginning of the process. The bandwidths of the formants progressively
decrease, from approximately 80-120 Hz down to very small values (less that 1 Hz at the
end), and make the formants increasingly more resonant. The strokes (or duration of the
grains) get longer and longer, from 7 ms to several seconds, in order to let the resonance
be heard. At the end the spectrum looks much like an additive synthesis sound. The
spectral characteristics (frequencies, amplitudes) of this final part are derived from the
bell-sound model described earlier.14
Figure 10. Sonogram of the “Death of the King’s Voice” (1’30”).

The final bell sound is made up of 25 partials which “appear” in the middle of the
synthesis process out of the five formants of the initial vocal sound. 25 FOF generators
therefore actually run throughout the entire process, organised in five groups of five
formants at the beginning (each group of five is centred on the same frequency, but four
of them start with a null amplitude), which progressively spread out to the 25 different
frequency/amplitude values of the target sound spectrum as their bandwidths decrease.
13 However, as classical singers do not often use such long glissandi, and normally do not add vibrato
while performing a glissando, this example already suggests a kind of imaginary voice even for a more
“vocal” moment.
14 Onemight also perceive an indirect reference to the famous passage in Karlheinz Stockhausen’s
Kontakte where a high pitch (E5) starts a downward glissando ending with slow, isolated, reverberated
impulses tuned to E3.
314
The sub-patch at the upper-left of Figure 9 (gen-fql) produces the sequence of formant
data. This part of the process is detailed in Figures 11 and 12.
The sequence starts with formant values selected from a vowel database (Phase 1).
During this initial phase the synthesis process renders smooth transitions between the
vowel “u” of a soprano (labelled “su”), an alto (“au”), a tenor (“tu”), and, finally, a
bass (“bu”). The function duplicate-formants does exactly what it says: it duplicates the
formant data so that the total number of formants equals the target number of additive
partials required at the end of the process.
The sequence ends with the values of the bell-sound spectrum (Phase 3). These values
are generated from spectral analysis data in the west_bell sub-patch visible in Figure 12.
A cr-model object is created from a partial tracking analysis and a sequence of hand-
positioned markers (two separate SDIF files exported from AudioSculpt). Only one of
the segments is selected as a source for frequency and pitch material, and bandwidths
are computed from the frequencies using the autobw function of the OM-Chant library.
In between (Figure 11, Phase 2) the interpolation function generates 12 intermediate
states of the FOF generators. Figure 13 shows the contents of the resulting cr-model
object, considered the “knowledge base” (or “skeleton”) of the synthesis process.
Figure 11. Generation of the sequence of formant data for the “Death of the King’s Voice”
synthesis process. Left: formant values extracted from a vowel database. Right: formant values
derived from the spectral analysis of the Winchester Cathedral bell recording (see Figure 12).
Middle: interpolation of the formant values.
315
Figure 12. Extracting formant values from the analysis of the bell recording. Note that the
original sound has a spectrum similar to C minor, but for the opera, a spectrum similar to A
minor was needed, hence the multiplication of the frequency values (“Tune to A”).
Figure 13. Contents of the cr-model object from Figure 9.
316
Thanks to the combinations of higher-order operators and the library of specialised

tools available, this whole process, lasting almost 1’30”, is actually implemented as a
single “note” in the synthesiser. Once it is triggered, this note develops automatically,
thanks to its inner logic, the knowledge base, and the programmed rules contained in the
patch.
The expand-model box in Figure 9 generates the actual OM-Chant events driving
the sound synthesis process, starting from the cr-model data. As with cr-control-patch in
Figure 2, this time the get_fofs sub-patch (see Figure 14) determines a mapping between
the contents of the cr-model segments and the different parameters of the generated
synthesis events (in this case, the slots of the ch-fof class: freq, amp, bw, wdur, win,
wout, etc.)
Figure 14. Control patch mapping the cr-model data to the parameters of the ch-fof synthesis
events in Figure 9.
The “BW scalers” curve defined in the main patch is used to scale the bandwidth
values of the formants during this mapping process. When the sound is clearly vocal
(at the beginning) this factor is 1.0 (that is, the initial values are used), then it rapidly
decreases from 0.02 to very small values (until 0.01, which is 1/100th of the original
bandwidth) for the last part of the process (remember that narrow formants get closer
to additive partials). Using a multiplication factor rather than absolute values has the
advantage that the wider formants will still have, proportionally, a larger bandwidth and
therefore decrease more rapidly, which is exactly what happens with the resonances of
a bell sound. As mentioned previously, the duration of the FOF grains (wdur) also
increases during the process.
317
The most difficult task in the design of this synthesis process was to find musically
and perceptually convincing values for the successive intermediate states of the changing
parameters. The interesting moments (for example, when the sound begins perceptually
to become a bell) are very sensitive and usually the optimal values for these parameters
are located within small, cross-dependent intervals. We had to proceed through intensive
trial-and-error to find them.
The linear interpolation of the fundamental frequency, for instance, did not work
well in the logarithmic space of pitches (the glissando is too slow at the beginning, and
quickly moves down to “unmusical” values at the end). A linear pitch interpolation
yields better results, but then the process tends to sound quite mechanical. Using linear
frequency interpolation on portions of the glissando within each octave (see “F0 profile”
on Figure 9) seemed to give the more interesting musical results.
The ch-f0 events controlling the evolution of the fundamental frequency of the syn-
thesis process are generated on the right of Figure 9. Here expand-model and the gen_F0
mapping patch only use the time information from the cr-model and take data from the
fundamental frequency curve. The details of gen_F0 are visible in Figure 15. This patch
mostly consists of a choice of whether vibrato is applied to the current segment—see the
selection list (2 3 4) in Figure 9. If the current segment falls within the selection, the
frequency value is turned into a linear ramp to the next frequency in the list, to which
vibrato is applied using param-process.
Figure 15. Mapping between values of the cr-model of Figure 9, the frequency curve, and the
ch-f0 slots. This patch applies (or not) the vibrato function depending on whether the current
segment is among the selected indices allowed to pass from <select-vib>.
318
Finally, the ch-transition function visible in Figure 9, associated with the transition
patch, controls the smooth evolution of the whole process, ensuring transitions between
the “continuous” vibrato curves and static values of the fundamental frequency. The
transition patch performs a number of tests on the successive pairs of ch-f0 events
generated in the previous part of the process (are the two events continuous/bpf objects?
do they overlap? etc.) and makes decisions about the data to be produced as an output
to the synthesis process (see [2] for more details on this process).
Figure 16. Re Orso: death of the king. Photo: Elisabeth Carecchio
Conclusions
Re Orso allowed us to explore several exciting æsthetic and scientific concepts. During
the performances we realised that the electronics could, indeed, play the role of a dramatic
character, certainly invisible to the audience, but likely to tell a “story” and to evolve
emotionally as the opera unfolds.
The Chant and OMChroma paradigms, both present in the compositional process,
proved to be complementary: while OMChroma mainly deals with sound structure
at the (discrete) level of the notes, OM-Chant was naturally suited to address issues
of phrasing, which are crucial for the synthesis of sung voice and for conferring to the
synthetic sounds a high degree of expressivity. With these tools, we were able to produce
sounds of high acoustic quality within a reasonable computation time and, especially, to
give to the whole sonic realm the unity for which the composer was searching.
319
We are aware that we only skimmed over certain aspects of the synthesis system,
and that much more experience is needed fully to exploit its full potential. The constant
guidance of an æsthetic perspective (that is, synthesising a certain sound family because
of the needs of the libretto) helped us to focus on the most salient musical directions.
This was a crucial aspect of the experience.
References
[1] Carlos Agon, Jean Bresson, Marco Stroppa. “OMChroma: Compositional Control
of Sound Synthesis”. Computer Music Journal, 35(2), 2010.
[2] Jean Bresson, Raphaël Foulon, Marco Stroppa. “Reduction as a Transition
Controller for Sound Synthesis Events”. In FARM—Workshop on Functional Art,
Music, Modeling and Design, ICFP’13. Boston, 2013.
[3] Jean Bresson, Marco Stroppa. “The Control of the CHANT Synthesizer in
OpenMusic: Modelling Continuous Aspects in Sound Synthesis”. In Proceedings
of the International Computer Music Conference. Huddersfield, 2011.
[4] Jean Bresson, Marco Stroppa, Carlos Agon. “Generation and Representation of
Data and Events for the Control of Sound Synthesis”. In Proceedings of the Sound
and Music Computing Conference. Lefkada, 2007.
[5] Arshia Cont, Carlo Laurenzi, Marco Stroppa. “Chromax, the other side of the
spectral delay between signal processing and composition”. In Proceedings of the
International Conference on Digital Audio Effects (DAFx). Maynooth, 2013.
[6] Xavier Rodet. “Time-domain Formant-wave Function Synthesis”. Computer Music
Journal, 8(3), 1984.
[7] Xavier Rodet, Pierre Cointe. “FORMES: Composition and Scheduling of Processes”.
[8] Xavier Rodet, Yves Potard, Jean-Baptiste Barrière. “The CHANT Project: From
the Synthesis of the Singing Voice to Synthesis in General”. Computer Music Journal,
8(3), 1984.
[9] Marco Stroppa. “Musical Information Organisms: An approach to composition”.
Contemporary Music Review, 4(1), 1989.
[10] Marco Stroppa. “High-Level Musical Control Paradigms for Digital Signal
Processing”. In Proceedings of the International Conference on Digital Audio Effects
(DAFx). Verona, 2000.
[11] Marco Stroppa. “Auf der Suche nach formalen Polyphonien. Zwischen Musiktheorie
und Neurowissenschaft”. Musik & Ästhetik, Heft 01(1), 2012.
Acknowledgements: The authors would like to thank Thibaut Carpentier, Arshia Cont, Nicholas Ellis,
José Fernandez, Raphaël Foulon, Gilbert Nouno and Xavier Rodet from IRCAM, Marlon Schumacher
from CIRMMT/McGill University, and Jean-Baptiste Barrière for their invaluable help and advice.
320
Rima Flow: Oral tradition and
composition
Alessandro Ratoci
Two genres of improvised musical expression, so distant in time and space as the tradi-
tional ottava rima chant of 19th -century rural Italy and the contemporary “beatboxing”
practice of urban hip-hop music, have been the inspiration for my piece Rima Flow for
tuba and electronics.1
Computer-assisted composition techniques have been intensively employed to derive
all of the symbolic (notational) material and the electronic (concrete) sounds directly
from recorded sources of those oral-tradition repertoires. This provided the opportunity
to experience formalisation as a kind of cultural perspective to discover new possibilities
from a traditional source of material.
Relics of a popular literacy: ottava rima poems

Ottava rima is a stanza form of Italian origin composed of eight lines of 11 syllables
each (thus the name ottava) originated in the late 13th and early 14th centuries. One
of the earliest practitioners was Giovanni Boccaccio, who established ottava rima as the
standard form for epic and narrative poetry. A distinctive characteristic of the form is
the peculiar rhyme scheme, consisting of three open rhymes and a closing couplet: ab ab
ab cc. Below are two examples of ottava rima stanzas:
Dirò d’Orlando in un medesmo tratto

cosa non detta in prosa mai, né in rima:
che per amor venne in furore e matto
d’uom che sì saggio era stimato prima;
se da colei che tal quasi m’ha fatto,
che’l poco ingegno ad or ad or mi lima,
me ne sarà però tanto concesso,
che mi basti a finir quanto ho promesso.
Ariosto, Orlando Furioso, stanza 2 (1532)
1 Written during the IRCAM Cursus in composition and music technologies in 2015.
321
Alessandro Ratoci
Ambition was my idol, which was broken

Before the shrines of Sorrow and of Pleasure;
And the two last have left me many a token
O’er which reflection may be made at leisure:
Now, like Friar Bacon’s brazen head, I’ve spoken,
‘Time is, Time was, Time’s past’, a chymic treasure
Is glittering youth, which I have spent betimes
My heart in passion, and my head on rhymes.
Byron, Don Juan, stanza 217 (1819)
Beside the use of this form for heroic literature, a parallel use of ottava rima is
consistently found in traditional popular chants of rural areas in the centre of Italy,
passed from generation to generation by word of mouth.
These relations between “high” (literary) and “popular” (oral) culture have been part
of the social identity of Italy from the Middle Ages to the advent of marketed mass-media
culture. As of the beginning of the 20th century, oral poetry still had an important role in
the education of moral and religious values for the lower classes, which were still largely
illiterate. The divulgation of spiritual poems, ranging from Tasso’s Gerusalemme Liberata
to the entirety of Dante’s Divina Commedia, was carried out by popular storytellers with
the encouragement of religious authorities. It was still common in the late 1980s and 90s
to find old rural poets who, despite their semi-illiterate condition, were able perfectly to
sing entire poems dating from the Renaissance or even entire chants of Dante’s Divina
Commedia. As an aid to memory, and to catch the listener’s attention, the poems where
often sung over stereotypical melodies of a simple syllabic character. These melodies were
the base for improvised embellishments and ornamentations which were often exaggerated
to demonstrate the storyteller’s virtuosic abilities. These melismas, extracted from some
original recordings by different performers, are the most characteristic source materials
used for the composition of Rima Flow.
From beatboxing to instrumental techniques

Beatboxing is a tradition of vocal percussion which originates in 1980s hip-hop, and
is closely connected with hip-hop culture. It involves the vocal imitation of drum
machines as well as drums and other percussion, and typically also the simultaneous
imitation of basslines, melodies, and vocals, to create an illusion of polyphonic
music.
Dan Stowell and Mark D. Plumbley [3]
Beatboxing is one of the distinctive elements of the hip-hop culture that formed during
the late 1960s among African American youth of the American suburbs, rapidly becoming
a widespread part of identity in the metropolitan areas of the U.S. and Europe. Important
pioneers of the genre have been, among many others Kyle Jones (a.k.a Scratch), Marcel
Theo Hal (a.k.a Biz Markle), Darren Robinson (a.k.a Buffy) and Kenny Muhammad.
There have already been interesting examples of cross-fertilisation between beatbox-
ing style and extended techniques of contemporary instrumental music. One of the first
instruments to be involved in this practice has been the flute, surely for the peculiar
embouchure which leaves the mouth and the lips complete freedom of movement, and
322
probably because of the rich percussive repertoire of effects that where already present
in the Jazz and Rock flute tradition.
It is unknown when beatboxing was first transferred to the flute, though it seems to
have occurred sometime in the early 2000s. Many of the articulatory and extended
techniques involved in beatbox flute performance have been used by composers for
decades. Examples include the “ch” articulations, key-slaps, flutter tonguing, spit
tonguing, and audible breathing in the performances of Jethro Tull’s front man, Ian
Anderson.
Christopher Kuhns [2]
The most relevant exponents of this experience are, among others, Tim Barsky,
Nathan “Flutebox” Lee, and Greg Pattillo. The large and widespread favour that
this genre of performance has encountered from the public through the internet has
encouraged many young instrumentalists to confront themselves with this kind of sound
palette: on the internet is possible to find examples of beatboxing using the most
disparate of wind instruments: saxophones, clarinets, or oboe, even if the presence of
the reed inside the oral cavity of the performer limits the possible percussive techniques.
The large mouthpiece of the tuba is also suitable for these kinds of sound effects and
many of them are already present in the contemporary repertoire of extended techniques.
The idea to include beatboxing elements in Rima Flow came also from the collaboration
with the instrumentalist Jean-Baptiste Renaux, who is both a brilliant classical tuba
player and a virtuoso beatboxer. This collaboration was very inspiring to me and the
final results represented a perfect depiction of the performer’s personality and musical
identity.
A range of heterogeneous sounds was employed to establish a continuum between two
dialectic opposites of sonic identity: the pure sound of the tuba and the singing voice
on the one hand, and harmonic sound and breath noise on the other (see Figure 1).
Ordinary tuba playing techniques have a clear vocal quality, and the very low register
exhibits homologies between the glottal pulses of vocal phonation and the mechanisms of
sound production on brass instruments. At the same time, the large palette of beatbox
vocal percussion, especially if performed inside the mouthpiece of the instrument, can be
employed as an intermediary sound morphology between un-pitched percussive extended
techniques and the unvoiced phonemes of the human voice.
A new form of oral tradition

The decision to use multiple sources of inspiration (ottava rima and hip-hop beatboxing)
instead of dealing with a single coherent corpus of traditional material was suggested by
a number of reasons, some of a more conceptual nature, some more linked to sound
morphology. I have always been fascinated by contrast, hybridisation, and hidden
homologies between concepts that are considered distant or even opposite by common
sense. In this case there are interesting parallels between the two cultures that I have
taken into consideration:
• Singing popular ottava rima poems, commonly know in Tuscany as cantar di poesia,
was often improvised as part of public contests, exactly as in the modern “slam”.
323
Alessandro Ratoci
Figure 1. Sound categories of Rima Flow.
• These improvisation “battles”, or contrasti, were an important moment of social ex-

change, like today in the context of urban culture, where the established performers
and aspiring ones fight with each other using metre and rhyme.
• Subjects that were usually unacceptable in social contexts (sexuality, rebellion
against authority, political satire) were permitted during the improvisation cer-
emonies of ottava rima, suggesting an interesting relationship between language
virtuosity and freedom of speech, which is also part of rap and hip-hop culture.
I absolutely wanted to avoid any nostalgic, conservative approach toward a historical

experience that is part of my personal identity (the ottava rima repertoire was commonly
practiced in my family and represents my very first contact with musical practice) and
favour a creative “fresh look” instead. The sound analysis and consequent formalisation
of the results was important to me as to establish a kind of rational “filter” of this
emotionally charged material and to find a cultural distance that left me enough freedom
to create new music instead of simply preserving a tradition.
The role of computer-assisted composition

OpenMusic was employed at different stages of the elaboration of the piece to solve the
following specific problems:
• Definition of the general form and the sections of the piece, modelled after the
stanza structure of the ottava rima poems.
• Extraction of melodic profiles from audio recordings as a source for melodic material
of both instrumental and electronic parts.
324
• Transformation of the extracted profiles into graphic symbols for semi-aleatoric

notation of instrumental gestures.
• Elaboration of concrete recorded material (micro-editing, dynamic envelope trans-
formation, etc.)
The first step: analysis of the repertoire

Out of the large corpus of oral-tradition poems in ottava rima from anonymous authors,
I have chosen one particular poem of which I possessed multiple recordings by different
storytellers: “Poesia del sogno di un incontro fra un nipote e il nonno morto”. I found
this poem particularly interesting for my work because the abundance of recorded sources
allowed me to confront different examples of melodic improvisation, and also because
of the theme of the narration: during a dream the poet meets the ghost of his defunct
grandfather wishing to have news from the human world and offering, in exchange, a vivid
description of the afterlife. This simple vulgate of Dante’s La Divina Commedia was a
pretext to satirise the post-unity Italian society of that time, but also the celebration of
the power of poetry in its ability to connect different generations. The poem is composed
of 15 stanzas in the ottava rima form with a particularly virtuosic rhyme scheme in which
the last couplet of a stanza is taken as the starting rhyme for the successive one (the full
text is included in the Appendix at the end of this chapter).
The first step was to mark the subdivisions of the sound recordings according to the
three levels of structuration of the poetic text: stanzas, lines, and syllables. Transient
detection was performed using the OM-SuperVP library in order to help with the
syllabic subdivisions. The ability easily to transform transients into rhythmic pulses
was useful to correct the syllabic segmentation by ear, as even the most accurate choice
of parameters will require some manual adjustment of the markers, especially on non-
percussive sound materials like the singing voice (see Figure 2).
The duration in seconds of the successive lines of each stanza, displayed using a
break-point function library (bpf-lib), showed salient similarities in proportion (see
Figure 3). The distribution of these durations is consistent over the stanzas and reflects
the importance given to the different verses in a fixed narrative scheme: an invocative
opening section, a more discursive central part, and a rhetorical ending.
The pattern unfolds as six interlocking lines followed by a climatic couplet. The
three insistent alternating rhymes propel the narrative forward while also encour-
aging meditation and commentary. The couplet, on the other hand, is a stopping
point, a turn or summation [...]
Edward Hirsch [1]
The density of ornamentation also follows this general distribution, with a more
melismatic sequence at the beginning and end, and a more syllabic setting toward the
centre of the stanza. An approximation of the line durations to natural numbers was
used to structure the internal proportions of the different sections of the piece according
to the following numerical series: (10 7 4 7 4 6 5 8) (7 7 4 5 5 5 5 7) (8 6 4 6 5 7 4 7) (7 7 4 6
5 6 4 7) (8 6 4 5 5 6 4 8) (7 6 5 5 4 5 3 8) (7 5 3 5 5 5 4 6) (7 6 4 4 4 5 4 8) (7 6 5 5 5 6 5 8) (8 7
4 6 5 7 4 6) (7 5 4 5 5 6 5 7) (7 5 4 6 5 6 4 7) (7 6 4 6 5 5 4 7) (7 5 4 5 5 6 4 8) (8 5 4 5 5 5 4 7).
The proportions between the large-scale sections of the piece are also given by the
average duration of each of the eight lines of the stanzas (7 6 4 5 5 6 4 7), as if the whole
piece itself were a section of an ottava rima poem.
325
Alessandro Ratoci
Figure 2. Transient detection for syllabic subdivisions.
Figure 3. Proportions of line durations in each stanza.
326
Melodic profile extraction

The OM-SuperVP and OM-pm2 libraries were used to track the pitch profile of the
singing voice by fundamental frequency analysis (Figure 4). The fundamental frequency
traces where stored in bpfs and also exported as SDIF files to AudioSculpt software,
where they were segmented using the syllable onset markers previously extracted. A
series of “MIDI annotations” was then produced by rounding each fundamental frequency
to the nearest tempered pitch value. Two degrees of precision were employed during this
transcription process: 1) the most detailed possible, where any crossing of a tempered
pitch value was reported as a new event, and 2) using syllables markers as averaging
boundaries, where only one pitch was assigned per syllable (see Figure 5).
Figure 4. Fundamental frequency estimation using OM-pm2 and OM-SuperVP.
(a)
(b)
Figure 5. MIDI transcription of a fundamental frequency curve: (a) with maximum precision,
and (b) using syllables.
327
Alessandro Ratoci
Rhythm and quantification

The primary source of information on rhythmic structures was the result of a transient
detection applied to ottava rima singing and various fragments of beatbox improvisation,
quantised using the omquantify function (see Figure 6).2
Figure 6. Rhythm transcription of a beatbox fragment.
Instead of searching for the perfect transcription (in the sense of a resemblance to the
original) I started to use the transcription tools as a first stage of symbolic transformation
of the materials. Figure 7 for instance shows several different transcriptions of the same
musical passage.
The æsthetic (and poetic) role of automatic transcription

The use of different analysis algorithms and different degrees of precision in the previous
compositional processes was not only motivated by a technical exploration of the various
possibilities of the software at hand, but also by a precise idea about the conceptual role
and æsthetic implication of computer-assisted processes of sound analysis. Automated
processes should not, in my opinion, be a substitute for individual aural capacity; in-
stead they could represent, with their unrivaled level of accuracy, an ideal resource for
development of our perceptions and a fertile source of inspiration.
At a certain moment in my development as a composer I started to feel unsatisfied by
my instrumental writing style and felt the urge for a new vocabulary of rhythmic gestures
and melodic motives. I have never had much interest in abstract numerical rules or in
over-complex notation and thus I was searching for a vocabulary which would both bear
2 Thereactive feature of OpenMusic 6.9 (dark-framed boxes in Figure 6) was useful to have instantaneous
visual feedback from the quantisation parameters and their impact on musical notation.
328
Figure 7. Different degrees of accuracy of transcription of ottava rima singing.
some relationship to natural, physical phenomena and represent a development rather

than a radical substitution for traditional human musical gestures. For instance, certain
acoustical phenomena like the irregular vibrato of non-cultivated popular singing could
generate interesting rhythm if analysed with enough temporal accuracy; or delicate vocal
inflexions could transform into sweeping non-tempered melodic arches if stretched over
the frequency axis.
The classic problems of transcriptions found in ethnomusicological research were of a
fundamental importance for my work: the question of whether (and how) to transcribe
the fluidity of music of oral-traditions was resolved for me with a radical choice for
accurate representation. As the generation of new compositions was the purpose of my
operations (and not the trans-cultural preservation of a musical text) I decided to retain
any possible details and filter them through the looking-glasses of an precise musical
representation. All of the fluid inflexions of the improvised chant, the breathing pauses,
the metric variety of ornamentation of ottava rima singing, and the “groove feel” of
beatbox improvisation, were considered as intentional musical features and notated as
accurately as possible.
Last but not least, the use of different analysis parameters, and also the inevitable er-
rors introduced by different algorithms, were used as additional means of transformation
of the original material.
329
Alessandro Ratoci
The profile server

One of the most important practical benefits of the integration of OpenMusic in my
compositional workflow is the possibility of a rational organisation of heterogeneous ma-
terials: numerical data, audio files, fragments of music notation, and also electroacoustic
processing can be organised in an integrated workspace, a kind of musical “spreadsheet”
in which the starting material can proliferate in multiple dimensions.
I organise most of my workflow around a central patch that I call the “profile server”,
whose purpose is:
• To store all of the original sound files, their segmentation, melodic profile curves,
and transcribed notation.
• To access the materials according to the line segmentation of the poetic text, listen-
ing and confronting the original sound files with a “virtual instrument” rendering
of the transcribed fragments.
• To perform different kinds of filtering and the basic operations of inversion and
retrograde on the melodic profile curves.
• To perform different kinds of symbolic transformations on transcribed materials.
• To re-synthesise the transformed profiles into new sound files.
• To generate vector graphics from the melodic profile curves that can be imported
in music notation software (Finale, Sibelius, etc.)
The filtering operations shown in Figure 8 are necessary to control the complexity
of pitch transcription and to produce either a smoother or less stable quality in re-
synthesised audio. The algorithms are taken from the OM-Fil library by Mikhail
Malt. They include the mean-filter function with different values for <window-size>
and <recursion-level> (respectively, the amount of data that is taken into account for
averaging purposes, and the number of times the process is repeated).
The symbolic operations on transcribed notation are also simple, as I wanted to keep a
degree of resemblance to the original melos, and because the most radical transformations
of the material are already done during the transcription process. These operations can
be summarised in the following categories:
• Interval scaling (expansion, compression, and inversion) performed cyclically on

subgroups of notes;
• Pitch retrograde also performed on subgroups;
• Rhythmic retrograde;
• Rhythmic rotation.
330
Figure 8. Profile server: filtering and other operations on fragments of melodic profiles.
Performing the common operations of inversion and retrograde over small subgroups
of notes is a simple and effective way to generate new melodies where the degree of
resemblance to the original material can be precisely controlled. The interval-mult-groups
or retrograde-groups patches in Figure 9 both take a list of chords in their first input and
a list of numbers, corresponding to the segmentation of original melody, in the second.
Figure 10 shows the detail of the interval-mult-group sub-patch. In this example a
multiplication ratio of -1 produces a simple interval inversion using the first element
of each group as the pivot pitch (see Figure 11). In this way an internal hierarchy of
melodic material can be established in which some important notes of the profile are
left recognisable, as they do not vary over successive presentation of the transformed
material.
331
Alessandro Ratoci
Figure 9. Interval scaling and retrograde on subgroups of notes.
Figure 10. Interval multiplication on groups of notes: subpatch interval-mult-group.
332
Figure 11. Inversions calculated on sub-groups of notes.
Resynthesis of profile curves is used in the composition of the fixed-media electronic

part of the piece through the Csound and the OMChroma libraries (see Figure 12).
Among many possibilities, three generators (or “synthesis classes”) have been chosen:
• add-1 : sinusoidal generator employed for simple additive synthesis.

• fof-1 : formant waveform generator, a particular case of granular synthesis.
• buzz-1 : dynamic spectral oscillator capable of producing various noisy spectra
similar to brass sounds.
The melodic profiles control the pitch parameter of the synthesis processes and are
altered by other parameters used to obtain transformed versions of the original melisma.
These parameters are:
• <grain-dur>: the length of granular fragments that are obtained by sampling the
melodic curve at regular intervals. Longer grains naturally produce a more “fat”,
but also “blurred”, sound while shorter grains produce a clearer pitch profile, but
with a somewhat “thinner” timber.
• <stretch>: the melodic profile can be stretched (> 1) or compressed (< 1) in time
before re-synthesis.
• <transp>: the melodic fragments can be freely transposed by a given number of
midicents (mc).
333
Alessandro Ratoci
Figure 12. Re-synthesis of melodic profiles.
Finally, another possible output related to the analysis and elaboration of melodic
contours is the creation of vector graphic objects (SVG files) to be imported in notation
software for semi-aleatoric notation of instrumental gestures (see Figure 13).
Figure 13. Export of melodic profiles as vector graphics.
334
Sound processing in OpenMusic: OM-ReCycle

Rima Flow largely employs fixed-media electronic sounds. Beside the synthesised sounds
produced with the OMChroma library, the rest of the audio materials are re-elaborations
of concrete recordings.
Fragments of popular storytellers singing in ottava rima are included as a sort of
hidden “relic” of this unique material, elaborating it so as to leave the impression of an
ancient voice from the afterlife while making it impossible to distinguish for someone not
accustomed to this peculiar style of singing. The idea of electroacoustic transformation
to mask the stylistic connotations of material is an important concept in Rima Flow. It
is applied to both ottava rima and beatbox extracts to create a sort of “neutral” material
that is part of the dialectic of the piece.
An interesting approach to the deconstruction and successive reconstruction of recor-
ded audio is what could be called “slice/remix”. The process of segmenting audio material
in slices generally corresponding to rhythmic units and then recomposing it was initially
introduced as a method to adapt percussive loops to a different tempo (e.g. in the
ReCycle software) or to perform variations on rhythmic material. The flexibility of
the Lisp-based symbolic elaboration of OpenMusic and the integrated audio functions
inspired me into expanding this approach to conceive a more general set of utilities based
on algorithmic re-composition of the sound, with the purpose:
• To produce variations of the percussive beatbox material of different degrees of
fluidity and irregularity.
• To alter the order of syllables of the ottava rima chant to generate melodic varia-
tions and to mask the intelligibility of the text.
• To mix slices of different sources to produce hybrid sequences.
• To impose a rhythmic quality upon non-rhythmic material by discontinuity (slicing
a continuous sound at discrete intervals).
The first stage is the subdivision of the sound file into a list of segments based on
the embedded markers, as shown in Figure 14. The time-values of the markers are read
two-at-a-time and used as the parameters for the sound-cut function. The new sound
file is then recomposed by successive applications of sound-seq using the reduce 3 function
(see Figure 15).
Symbolic operations carried over the slice list generate different reorderings of the
sound file. This programmable approach enlarges enormously the possibilities that are
offered by regular sampler processors, which usually propose a limited set of operations
such as shuffling or reversing the order of the slices. Any kind of operation can be
executed over the slice order, from simple rotation to complex conditionals based on
the length or spectral content of the slices, or any other process limited only by the
composer’s imagination. Operations based on feature extraction of the slices (length,
amplitude, or spectral descriptors) can be used to impose a clear perceptual ordering
while modulo operations, algorithmic, or weighted-random permutations can give the
most unpredictable (and often interesting) results (see Figure 16).
3 The reduce function takes a list of elements and combines them using an arbitrary binary operation (in
this case joining two sound objects using sound-seq).
335
Alessandro Ratoci
Figure 14. Slicing a sound file. Right: Inside the slicer loop.
Figure 15. Recomposing slices in a new sound file.
336
Figure 16. Construction of hybrid sound material (beatboxing sounds and extracted phonemes
from ottava rima chant) by probabilistic juxtaposition.
Conclusions
More than a powerful instrument to speed up long and repetitive tasks while dealing with
musical composition and audio processing, OpenMusic also opens new and interesting
perspectives of a conceptual nature. By unifying in the same workspace aspects of the
compositional process that are generally considered separate—like the generation of a
score, the synthesis of sounds, and electroacoustic transformations—it inspires the sense
of a higher conception of the work as a whole.
337
Alessandro Ratoci
Transcript of the ottava rima poem “Poesia del sogno di un nipote e del nonno morto”,
as recorded by Francesco Ugolini, Firenze, 2006.
Natura dammi soccorso ai miei bisogni, Ora vi è un dazio che non si può soffrire...
che alla fonte della Musa io mi bagni Il prezo è il macinato di mulino;
Quindici ottave vi farò sui sogni Per riportare a casa il sacco pieno
e un numero all’ambo s’accompagni. Ci vuol tre franchi e ottanta, non di neno.
Di rose, gelsomini e catalogni, Rincarato è le biade, paglia e fieno;
di zanzare scorpion piattole e ragni Rincarato è il dormire e la porzione;
Da un sonno tardo su il letto rivolto Rincarato è le case ed il terreno,
Ecco un Nipote che sogna il Nonno morto. Per tre volte è più cara la pigione
E disse: « Come mai? chi vi ci ha porto? Nonno, ci fu la guerra sopra il Reno,
Mentre un dì voi passaste all’altra vita? La Prussia si batté con Napoleone...
Or siete diritto e voi moriste torto; E Pio Nono ha perduto il temporale.
Ditemi, come la sta questa partita? » Nonno, Roma è venuta capitale,
Disse: « Nipote mio, stai bene accorto: Per quanto sia nelle dimostrazioni...
Sono un’ombra terrena in via smarrita; Nonno, ti ho raccontato il bene e il male,
Il pensier mi guidò, se ti contenti, Qua de’ viventi tutti i paragoni.
Di saper le notizie dei viventi. » Nonno, ’gli è un mondo, sai: chi scende e sale,
Le mura non son più sui fondamenti, L’uomo giusto convien così ragioni,
Com’eran prima, che voi bensapete: Tutto t’ho detto in questa parte esterna
Quelli zecchini d’oro e quelli argenti Dimmi come sis ta la in vita eternal...
Ora son fogli, e c’è poche monete! Tristo è colui che l’umano scherna
Son diradati i frati nei conventi, Principiò a dir cosi, parlò al nipote,
E pochissima stima gode il Prete; l’inferno l’è un orribile caverna,
Non è la Religione come una volta... ombre notturne per le vie remote.
« Seguita, gli dicea, che il Nonno ascolta... » Tristo è colui che l’immortale scherna
La campagna ci dà buona raccolta: L’ingresso è tetro che il terror percuote
Per grazia del Ciel, sono più belle: Gole profonde e tempestosi venti
S’empiono i tini, le bigonvìce e sporta. Ripiene d’urli, strepiti e lamenti.
Ma son tanto più care le gabelle! Più giù c’è l’orlo e il cerchio dei serpenti,
Un miglio intero, sai, fuor d’ogni porta, Di mostri di demoni scatenati;
Se tu vedessi, l’hanno fatte belle: Che riguardano tutti i malviventi
Si pesano i barrocci alla stadera, Ch’ebbero al mondo i poveri strapazzati.
E il nome è intitolato la barriera. Ci sono degli avari qui presenti,
Rispose il Nonno: « Ma Nipote, è vera Ci sono fattori milordi preti e frati;
Quel che mi dici tu ne’ tuoi pensieri? » Di spie di ladri di ruffiani d’ogni veleno
Disse il Nipote: « Un altro n’è in carriera: La terza parte dell’ inferno è pieno
Gli hanno disfatto, sai, tanti poderi. La discendenza tua là troverai,
E vi è un vapore da mattina a sera, Genitori, fratelli e le sorelle,
’gli è quello che trasporta i passeggeri... Nonni, bisnonni e figli se ce n’hai
Tranta vagoni s’accatena attorno, Son costì fra gli Arcangeli e le ancelle.
Da Firenze in due ore va a Livorno. Il Padreterno tu saluterai,
Poi c’è le guardie con trombetta e corno, Quel divino che fece opere belle;
Acciò non attraversi il viandante, Cerca di star con lui di notte e giorno,
E un fil di ferro lungo all’intorno, Ti lascio in terra e in paradiso torno.
Che si chiama il telegrafo volante. Si risveglia il nipote tutto adorno,
In un quarto d’ra, sia di notte o giorno, Di fede di speranza e religione;
Si dà notizie per le città tante, Ogni vizio mondano gli fa scorno,
Nella chiamata, come al referire... » Solo confida alla sua vocazione.
Risponde il Nonno: « Tu mi fai stordire! » I dritti del cantore interpretonno,
Un’altra, Nonno, ve ne voglio dire, I segni dati all’immaginazione;
Di quando voi facevi il contadino, L’indice mi svegliò qui all’improvviso,
E che il grano valeva dodici lire, Per risvegliar pensieri in paradiso.
Il più bello e il più caro uno zecchino:
338
References
[1] Edward Hirsch. A Poet’s Glossary. Boston: Houghton Mifflin Harcourt, 2014.
[2] Christopher Kuhns. “Beatboxing and the Flute: Its History, Repertoire, and
Pedagogical Importance”. Theses, Treatises and Dissertations, Florida State
University, 2014.
[3] Dan Stowell, Mark D. Plumbley. Characteristics of the beatboxing vocal style.
Technical report, Centre for Digital Music, Department of Electronic Engineering,
Queen Mary, University of London, 2008.
Acknowledgements: I would like to thank Mikhaïl Malt and Jean Bresson for their continuous support
during my time in the Cursus 1 at IRCAM, Jérémie Garcia for the development of the graphic export
routines, and Florent Jacquemard for his interest in the question of rhythmic quantification.
339
Ab-Tasten: Atomic sound modelling
with a computer-controlled grand
piano
Marlon Schumacher
We observe a fraction of the process, like hearing the vibration

of a single string in an orchestra of supergiants. We know, but
cannot grasp, that above and below, beyond the limits of per-
ception or imagination, thousands and millions of simultaneous
transformations are at work, interlinked like a musical score by
mathematical counterpoint. It has been described as a symphony
in geometry, but we lack the ears to hear it.
Stanislav Lem, Solaris.
This chapter describes concepts and techniques for the composition of the piece Ab-
Tasten for computer-controlled grand piano and electronics. It will discuss some of
the conceptual implications of sound representations for music creation and introduce a
model for corpus-based atomic decomposition, which served for the composition of both
acoustic and electronic materials. The accurate control of timing and dynamics on the
computer-controlled piano allowed me to compose in a continuum between instrumental
writing and sound modelling, and exploit principles of auditory organisation to create
the illusion of spatial sound synthesis with an acoustic instrument.
Introduction
The piece Ab-Tasten for computer-controlled piano and electronics was commissioned
in 2011 for a live@CIRMMT concert, dedicated to works reflecting the historic, cul-
tural, and technological transformations of the clavier.1 With a history of over four
hundred years of development, its latest embodiment, the modern grand piano, is one of
today’s most versatile instruments. In an organological sense, it stands out as a hybrid
instrument associated with different instrument families, based on its sound producing
medium (vibrating string), excitation mechanism (striking action of hammers), or playing
technique (keyboard interface). Indeed, composers have interpreted the piano in various
ways, e.g. as percussion instrument (Béla Bartók, Szabadban), string instrument (Helmut
1 http://www.cirmmt.org/activities/live-cirmmt/clavisphere/
341
Marlon Schumacher
Lachenmann, Klangschatten – Mein Saitenspiel), resonator (Luciano Berio, Sequenza

X ), and even as a predecessor of the electronic synthesiser (Karlheinz Stockhausen,
Klavierstücke XV–XIX ).
The piano’s harmonic and polyphonic capabilities, together with its timbral richness
(see e.g. [1]), enable it to conjure up sonorities evoking extramusical sounds, which has
inspired generations of composers. Maurice Ravel and Claude Debussy, for instance, used
idiomatic pianistic gestures (arpeggi, glissandi, tremoli) to describe the fluid, amorphous
movements of water, e.g. in Jeu d’eau or Reflets dans l’eau [11]. Other notable exam-
ples are Olivier Messiaen’s transcriptions of birdsongs (Catalogue d’oiseaux), and more
recently, Peter Ablinger’s resynthesis of human speech in Quadraturen [13].
The tradition of recreating the sonorities of concrete sources inspired me to develop a
method that transcends the notions of sound (timbre) and symbolic organisation, and to
extend piano writing to the fine structure of sound itself. While digital sound synthesis
has allowed us to craft virtually any sound material on a medium, the constraints given
by the physics of instruments and human performance have made it more challenging to
apply similar techniques in acoustic composition. Implementing such a concept requires
a sound model that considers the timbral and physical characteristics of the instrument,
as well as a degree of control and accuracy that exceeds the limits of human performance.
The concert took place in the Multimedia Room (MMR) of the Schulich School of
Music of McGill University, which is equipped with a Yamaha DCFX Mark IV Disklavier
(a MIDI-compatible concert grand piano). The precise polyphonic control of pitch,
dynamics, and timing possible with this instrument seemed well-suited to pursuing the
idea of sound composition with a physical instrument. Accordingly, the piece has no
human performer (hence no symbolic score) and can be considered a fixed media piece
for robotic instrument and electronics.
This chapter will describe two compositional concepts developed for the piece, realised
in OpenMusic. The first is a corpus-based, atomic sound model that allows writing
music in a continuum from abstract musical materials to modelling of concrete sound.
The second is an approach for transferring concepts of spatial sound synthesis to an
electroacoustic setting using principles of auditory organisation.
Abstract sound representations

Using computer technologies, any sound can be captured as a series of samples on a
digital medium. Data and structures can be extracted via analysis techniques to build
higher-level sound descriptions, which can then be integrated into symbolic compositional
processes. Such descriptions may serve for the creation of musical materials, but also
inspire compositional thinking and lead to a rich dialectic between symbolic and sonic
processes (see for instance the many examples in [5]). We should remind ourselves, how-
ever, that extracting meaningful information from the modulations of a sound wave is a
non-trivial task. Indeed, a given description of a sound is conditioned by the assumptions
of the underlying model used for analysis. This choice determines which aspects of sound
are considered meaningful and exposed to the foreground versus aspects that are regarded
as less relevant (and rendered implicitly or possibly not at all) [4]. Consequently, the
information retrieved from a sound is subject to different interpretations, each deriving
from the structure imposed by the underlying assumptions.
342
Ab-Tasten: Atomic sound modelling with a computer-controlled grand piano
In musical contexts, representations based on the short-time Fourier transform (STFT)

or wavelet transform are popular examples, using time and frequency/scale as two
dimensions (reminiscent of a musical score). These representations are typically agnostic
with regards to the content of the sound to be modelled, presuming that the salient
aspects of a sound can be modelled via expansions of a single frame of functions. When
the frame of functions fits the structure of the sound well, a meaningful representation
results; inversely, discrepancies can obfuscate or distort the information (e.g. representing
noise or transients via sinusoidal components). In addition to the consequences related
to the fidelity of describing different sound characteristics, every model puts forward
its structural characteristics. An additive model, for example, describes sound as a
superposition of homogenous sinusoidal partials, whereas a granular model describes
sound as a succession of short-duration sound quanta. In an abstract sense, this could
also be seen as supporting a compositional preference for simultaneous (vertical) vs.
sequential (horizontal) organisation. Thus, each model provides a particular viewpoint
defining a framework of possible operations. A further consideration in the context
of instrumental composition and transcription concerns the abstract nature of these
sound representations. Their smallest structural elements (i.e. the frame functions) are
typically based on mathematical objects (sinusoids, wavelets, etc.), which are on the
one hand not a perfect match for acoustic sounds encountered in the physical world
(such as instrumental timbres), and on the other not universal or objective, since they
depend on many analysis parameters (resolution, windowing, etc.) and emphasise specific
characteristics and structures, independently of and possibly far from the nature of the
sound or compositional context.
An alternative category of sound representations, that aim to adapt to different sound
characteristics using heterogeneous sound elements, are dictionary-based models. Widely
used in signal processing applications (e.g. compression, in-painting, denoising) they
decompose sound into a linear combination of elementary waveforms (called “atoms”)
contained in an analysis dictionary. Rather than using a single frame function, a dic-
tionary may contain atoms of variable duration, bandwidth, spectral content, etc. and
is thus capable of associating atoms that ideally match different sound characteristics.
Once a dictionary has been defined, techniques for finding a sparse combination of
atoms (i.e. with the least number of elements) can be used, such as Matching Pursuit
(MP [10]). MP is an iterative algorithm that aims at finding a combination of atoms best
to approximate a sound using a greedy search strategy (i.e. choosing at each iteration
the atom that best matches the residual part of the sound). The temporal structure is
specified via temporal locations in the target sound which correspond to possible positions
of atoms in the search procedure. In simplified terms, the structure of the algorithm can
be described as follows:
• For each temporal location, calculate a correlation coefficient of each atom with
the corresponding segment in the signal (i.e. starting at the temporal location and
lasting for the duration of the atom);
• Select the atom with the highest global coefficient, store it as part of the model
and subtract it from the signal (leaving a “residual”);
• Repeat this process on the residual until a breaking condition is met.
This process results in a model, i.e. a linear combination of atoms over time, and the
remaining residual (the difference between the target sound and the model). Since for
343
Marlon Schumacher
each iteration the globally best matching atom (over all possible temporal locations) is
selected, a model is built from the most significant to least significant element (in terms
of energy), rather than from beginning to end of the target sound (in terms of time).
Consequently, the final selection of atoms as well as their temporal positions in the
model are determined in the matching process. A dictionary-based model can be further
analysed to derive higher-level structures (similar to partials derived from STFTs). For
instance, atoms can be grouped based on their own parameters (e.g. amplitude or
duration) or depending on their context (e.g. proximity of atoms in time or frequency).
See also [20] for an overview of dictionary-based methods for analysis, visualisation, and
transformation of audio signals.
Corpus-based atomic decomposition

The composition of Ab-Tasten was the incentive for my software developments in sound
representations based on Corpus-based Atomic Decomposition (CBAD). Rather than
using abstract signals as in signal processing applications (short-duration waveforms),
in CBAD a dictionary is built from a collection of arbitrary sound files, also referred to
as a corpus. These sound files are typically concrete sound objects themselves (i.e. with
a complex spectral morphology) and constitute the smallest structural elements of the
sound representation. As in the case of dictionaries containing abstract signals, a match-
ing pursuit algorithm is used to find a combination of atoms that best approximates
a target sound. This approach establishes a direct relationship between the sounds in
the dictionary and the sound to be modelled, and can be thought of as representing a
sound as a polyphonic organisation of other sounds. Because in musical contexts we
are not necessarily aiming to find a sparse representation that eventually converges with
the target sound, there are no constraints on the temporal structure, contents of the
dictionary, or cardinality (number of atoms) of the model. Instead, these specifications
become a compositional choice. Atoms can be indexed and tagged, and can thus be
assigned arbitrary meanings, such as a note, instrumental gesture, or abstract symbol—
which allows us to think of the dictionary as a kind of musical vocabulary. Compared
to other sound representations, CBAD has a number of interesting characteristics for
compositional applications:
• It extracts polyphonic, temporal structures;

• It is a non-uniform representation using arbitrary collections of sounds;
• It is an iterative approximation of variable resolution, i.e. it is possible to control
the perceptual similarity of target and model;
• It leaves a residual sound that is complementary to the model (mixing the model
and the residual perfectly reconstructs the target sound);
• It permits creating multiple models from the same target using different dictionaries
(parallel), or consecutive decompositions by using the residual sound as a new target
which can be modelled with a different dictionary and so forth (serial);
• The selection of atoms from the dictionary, as well as their horizontal and vertical
organisation in the model, are determined by the matching pursuit algorithm, which
can be an interesting creative resource.
344
The use of a target sound as a template for sound synthesis based on collections of
concrete sound material relates to a number of techniques, such as audio mosaicing [22],
adaptive micromontage [19], or data-driven concatenative sound synthesis [9].2 Although
these systems offer powerful sound analysis/synthesis techniques, few of them provide
possibilities for linking the models to other compositional processes and symbolic repre-
sentations. One notable example is the work by Einbond et al. using the CataRT system
for feature-based transcription of audio targets in OpenMusic [7]. This representation is
based on a Euclidean vector space of audio descriptors populated with atoms (or units),
which are selected based on their proximity to a target position. One difference between
this approach and CBAD is that its k-nearest-neighbour matching does not iterate over
a residual in an out-of-time context and thus has no model for polyphony or temporality.
Analogies to visual arts

The concept of modelling larger-scale forms as a combination of concrete, smaller-scale
objects can be seen in various manifestations in the visual arts: an early example is
the work of 16th century painter Giuseppe Arcimboldo, who created portraits that were
made entirely of recognisable smaller-scale objects, such as fruits, flowers, fish, etc. These
objects, often symbolising an underlying theme, are combined in such a way that their
visual features (colour, shape, etc.) create the emerging perception of a larger-scale
form. While artists in the pre-digital age carried this work out manually, today there
are computer technologies for approximating image targets. Robert Silver, for instance,
patented a computer algorithm that selects and combines images from a database for
creating “photomosaics” [17]. The artistic interest in these techniques lies not in creating
an exact reproduction (facsimile), but rather in the addition of a semantic layer, by which
the characteristics of both the object to be modelled and the smaller elements used
for the modelling are retained. The emerging appearance of the larger-scale form can
also be considered an intended illusion, exploiting principles of perceptual organisation
described in the theories of Gestalt psychology [8]. Figure 1 shows a number of works by
Arcimboldo, Salvador Dalí, Bernard Pras, and Silver, each using a distinct technique.
Figure 1. Visual artworks by Arcimboldo, Dalí, Pras, Silver. Note the simultaneous perception
of local objects and larger-scale forms.
2 Inanother perspective, the possibility of modelling a sound in several stages with different sound
elements shares similarities with spectral modelling synthesis [16].
345
Marlon Schumacher
The visual analogy lends itself well to illustrating the notion of different modes of
perception as a function of context and resolution: in the case of a target image that
is modelled with only a few smaller-scale images, it will hardly be possible to recognise
the underlying larger-scale form. Instead, the characteristics of the smaller-scale images
themselves and the relationships between them (e.g. relative position, organisation) are
perceived as the content of the image. In the opposite case, with a high number and
density of smaller-scale images (as in a photomosaic), their cumulative features become
perceptually more salient and perception shifts to a more holistic mode, focusing on
macro-scale characteristics. At this resolution, details of the smaller-scale elements will
be barely visible, shifting the mode of perception from recognition of individual objects to
qualities of the “fabrics” of the modelled image (think of individual threads in a carpet).
A well-known example for a similar effect in the musical domain is György Ligeti’s or-
chestral work Atmosphères, in which the dense superposition of instrumental parts shifts
perception from recognition of individual notes and patterns to the textural qualities of
a larger-scale morphology. It is hardly possible to follow the characteristics of individual
instruments, which in this context do not carry individual musical meaning themselves
but rather become properties of a global timbre (a whole that is other than the sum of
its parts).
Building a dictionary of piano sounds
In Ab-Tasten, the first step for modelling of target sounds was the creation of a dictionary
that is representative of the acoustic grand piano. Simply stated, this means recording
individual piano notes from the acoustic grand piano to be used as atoms, which represent
the smallest indivisible elements of the model. Sampling a keyboard instrument such as
the piano is comparatively straightforward due to its percussive sound production and
structure of discrete keys with orthogonal dimensions of pitch and dynamics. Since
the sound production of the Disklavier can be controlled via MIDI, this sampling process
could be automated via a computer program that triggered individual notes and recorded
the sounds with a microphone. Sampling the entire combinatoric space of possible MIDI
key numbers and velocities would require recording 11176 (88×127) samples, which at an
average duration of 10 seconds would correspond to a corpus of about 31 hours of sound.
Atomic decomposition with such a large dictionary would be impractical in terms of
computing time. Initial tests with the Disklavier revealed that sampling the keys at 127
individual MIDI velocities was unnecessary, as the JND (just-noticeable difference) for
dynamics was at a value of about 4. Moreover, MIDI velocities greater than 120 sounded
unnaturally harsh, while values below 20 would not always accelerate the piano hammers
enough to reach the string and produce a sound. Thus, the 88 keys of the piano were
sampled at 25 distinct MIDI velocities (in the range of 20 to 120 in steps of 4), resulting in
a total of 2200 individual recordings. Since the decay of a piano note varies with its pitch
the duration of the notes was fixed to 15 seconds and the recording trimmed off earlier
if the rms amplitude fell below a threshold of -24dB. The program for carrying out this
sampling process was realised in OpenMusic using the library OM-SoX, which provides
a suite of objects and functions for audio recording and processing [24]. Figure 2 shows
the OpenMusic patch sending MIDI events to the acoustic grand piano for triggering
individual notes and recording the resulting acoustic sounds as audio files.
346
On the top left of this figure we see the generation of the lists for MIDI key numbers
and velocities, a number specifying the maximum duration of the recording in seconds,
and a string used as prefix for naming the files (a). Visible on the top right is the outer
loop (sample-robot), which iterates over the list of key numbers (b). On the bottom
left we see the inner loop, which iterates over the list of MIDI velocities and generates
unique filenames (c). The abstraction sox-samplebot (d) performs the sampling: the
function sox-process starts the audio recording. After 0.4 seconds a MIDI “note on”
event is sent, and the program sleeps for 15 seconds. In the meantime, the function
sox-trimsilence trims off “silence” (i.e. audio with an RMS amplitude value below -24dB)
from the beginning and end of the recording, before the “note off” event is sent. The
resulting sound is written to disk as an audio file with a unique name.
Figure 2. The patch used for automated sampling of individual piano sounds.
347
Marlon Schumacher
The library OM-Pursuit

The functionalities for CBAD were implemented as an external OpenMusic library,
titled OM-Pursuit3 [23]. This library wraps signal processing functionalities in pydbm
[2] and uses the Sound Description Interchange Format (SDIF) as a container for storage
and interchange of data describing the model and the dictionary. OM-Pursuit imple-
ments a number of classes for representing atomic sound models and a set of functions for
processing them. These models can be used for audio synthesis or converted into different
structures to be manipulated and eventually transcribed into a symbolic score. Figure 3
shows an example of an OpenMusic patch used for creating a corpus-based atomic
model of a sound target. The patch contains a target sound (sound object) (A), the
Figure 3. A patch illustrating atomic sound modelling with the library OM-Pursuit.
temporal locations calculated as an arithmetic series with an interval of 16 milliseconds

(stored in an SDIFfile object) (B), the dictionary generated via the abstraction make-
sdif-dict (stored in another SDIFfile object) (C), and parameters for the decomposition
(maximum number of simultaneous atoms and total number of atoms) (D). Three types
of information are stored in the dictionary for each sound file (top right in the figure):
3 Thistext describes the library at the time of composing Ab-Tasten (2011). It has since then been
further developed and extended with spectral audio descriptors and a constraint programming system.
348
the file path, pitch in midicents, and MIDI velocity. The function soundgrain-decomp
then takes these data, carries out the matching process, and returns three values: a
sound file of the audio synthesis of the model (E), a sound file of the residual (F), and
an SDIF file containing a parametric description of the model (G). At the bottom we
can see OpenMusic’s SDIF editor, which displays information about the model; each
atom is represented as an array with fields for onset time, duration, magnitude and norm
(amplitude), corpus index, file index, and file path.
This parametric description (G) can be converted into OpenMusic objects and
integrated into compositional processes like any other musical data. Figure 4 shows an
OpenMusic patch containing objects and functions for representing and manipulating
a model. The SDIFfile object at the top left (containing the model) is converted to a
sgn-array object, a tabulated structure in which columns represent individual atoms (in
OM-Pursuit referred to as “soundgrains”) and rows represent different parameters of
the atoms (A). This object can be directly used to drive audio synthesis or spatialisation
processes, e.g. using the libraries OM-SoX or OMPrisma [20]. Since the sgn-array
object internally stores pitch and velocity information for the individual soundgrains, it
can also be converted into a score-array object, a similar structure, in which columns
represent MIDI notes (B). This score-array object can be directly converted into a chord-
seq object (visible at the bottom left of the figure), and eventually exported to a MIDI
file: a “score” which can be performed by the computer-controlled piano. OM-Pursuit
includes two higher-order functions which allow connecting a Lisp function or patch in
lambda mode for vertical and horizontal manipulation of these objects: process-array-
comp iterates over columns, e.g. for filtering soundgrains that fall outside a specified pitch
range (C), and process-array-slot allows selecting a row to perform global processing, such
as rescaling of a parameter for the entire object (D).
Figure 4. Left (A, B): Conversion from SDIFfile to sgn-array and score-array. Right (C, D):
two examples for vertical and horizontal processing before converting to chord-seq objects.
349
Temporal Dictionary
Locations
(SDIF) Model
(SDIF) (Audio)
MarlonSound
Schumacher
Target
Atomic Model Sgn-Array Score-Array Chord-Seq MIDI file
Decomposition (SDIF) (Matrix) (Matrix) (Symbolic) (MIDI)
(Audio)
Residual Audio Voice Score

(Audio) Synthesis (Symbolic) (Music XML)
Matching
Pursuit
Figure 5 illustrates the process of CBAD in OM-Pursuit and the possible represen-
tations and renderings of the model.
Temporal Dictionary
Locations
(SDIF) Model
(SDIF) (Audio)
Sound
Atomic Model Sgn-Array Score-Array Chord-Seq MIDI file
Target
(Audio)
Residual Synthesis Voice Score

(Audio) (Audio) (Symbolic) (Music XML)
Figure 5. Corpus-based Atomic Decomposition in OM-Pursuit.
Modelling a birdsong
Dictionary with an acoustic grand piano
Temporal
Locations
(SDIF) (SDIF)
Let us look at a concrete example of the creation of musical materials for the piece.
Figure Sound
6 shows a sonogram
Target
Atomic (LPC analysis)
Model of an excerpt
Sgn-Array Score-Array of the birdsong MIDI
Chord-Seq thatfile served as
a source for developing most of the materials. Besides being an hommage to Messiaen,
(Audio)
the sound was chosen because of its inherent musical qualities including its rhythmic and
motivic structure. Visible as vertical lines in the sonogram are the markers used to specify
Figure 6. Top: Sonogram of birdsong. Bottom: Three models with 10, 50, 850 atoms,
respectively. Grayscale represents dynamics. Note how the matching process determines
polyphony and temporal structure.
possible temporal locations for atoms, defined as an evenly spaced grid with an interval
of 16 milliseconds (256 audio samples at 16 kHz sampling rate). Below the sonogram we
can see three horizontally aligned models whose parameters for the decomposition are
identical except for the number of iterations (10, 50, and 850, respectively). The grayscale
values in the chord-seq objects represent dynamics (as in the sonogram). This example
demonstrates the perceptual shift discussed before as a function of density and resolution;
for the first model consisting of 10 iterations (10 piano notes) it will be hardly possible
to recognise an underlying target sound. Instead, we perceive an abstract musical phrase
350
in which each of the individual piano notes are clearly identifiable. At 850 iterations,
in contrast, the same piano notes become micro-scale elements, integrated into a cloud
of sounds whose global features (granularity, pitch contour, etc.) become perceptually
salient. Instead of the relationships between individual notes we perceive an emergent
form on a larger scale.
Once converted into a symbolic representation, models of the birdsong were then
further processed (e.g. quantised, filtered) to create musical materials for the piece. For
the development of these materials I often switched between alternative representations
(e.g. continuous vs. metric time), as each offers different possibilities for manipulation.
For instance, in some situations the model was treated as a continuous signal using
sound-processing techniques (filtering, stretching, resampling), while in other situations
the model was treated as a symbolic musical structure and developed based on intervallic
relationships, harmony, rhythm, etc. Figure 7 shows a model of the complete birdsong
(4 phrases) in the chord-seq object and its quantification via the function omquantify
into a metric representation in a voice object. This is the main theme of the piece as it
appears on the acoustic piano at 1’20”.
Figure 7. Piano model of the birdsong as a chord-seq object (top) and its metric representation
in a voice object (bottom).
351
Marlon Schumacher
From virtual ensemble to meta-instrument

The second part of this chapter is dedicated to the relationship between acoustic in-
strument and electronics, as well as the approach to spatialisation in this piece. I was
interested in using electronics as spectral and spatial augmentations of the piano, almost
indiscernible from the acoustic instrument, to immerse the listener in a soundscape of
electroacoustic piano sonorities. Inspired by the notion of different perceptual modes
discussed before, I was aiming to develop a dialectic between the fusion of acoustic
and electronic elements into single sound source (akin to a “meta-instrument”) and the
splitting apart into individual musical identities, similar to a chamber music setting
(“virtual ensemble”).
An interesting feature of CBAD is the possibility of decomposing a sound using a
mixed dictionary containing several corpora (sound collections). From the resulting
model, the elements from the respective corpora can be extracted and organised into
individual structures, similar to instrumental parts in an orchestral score. Using a
dictionary containing corpora for instrumental and electronic elements (e.g. processed
or synthetic sounds), it is possible to develop both types of materials as complementary
parts in an integrated approach, through the same formalism.
Electronics as microtonal augmentation

The idea of fusion and segregation of instrumental and electronic parts was realised
by conceiving the electronics as an ensemble of four “virtual pianos”, represented as
microtonal transpositions of the acoustic grand. To that end, four new sound corpora
were created, representing each of the virtual pianos. The original recordings of the
acoustic piano were copied and transposed upwards by 20, 40, 60, and 80 midicents
respectively, resulting in a total of five sound corpora in 12-tone equal temperament
(12TET), each “tuned” 1/10th tone higher. These sound corpora could then be combined
together into a mixed dictionary with a resolution of 60 pitches per octave (60TET). From
the model created with this mixed dictionary, the pitches corresponding to the acoustic
and virtual pianos can be extracted and assigned to the respective parts. Depending
on the perceptual properties of the resulting musical structure, these individual parts
(corresponding to individual pianos) may be integrated into a single auditory stream
(comparable to a musical “voice”), or segregated into individual voices, according to
principles of auditory organisation [3].
Figure 8 shows three examples to illustrate this effect. The chord-seq object on the top
left shows an ascending scale over one octave in 60TET (A). The function micro-›multi
parses the corresponding pitches into five microtonal parts (staves) in semitone resolution.
In this example, the notes are closer in pitch and time between adjacent parts than
within the same part, which creates the perception of a single auditory stream that
is alternating between the parts. The middle example (B) shows a chord with pitches
corresponding to the frequencies of the first 30 partials of a harmonic spectrum (quantised
to 60TET resolution), which are distributed to the five microtonal parts. Here, the
harmonic structure of the pitches and their simultaneous onset create the perceptual
fusion of the individual parts into a single auditory object. On the right (C) we see
eight chords in rapid succession consisting of three to eight random pitches between
3600 and 8600 midicents that are quantised to 60TET and distributed to the individual
352
piano parts. This creates an ambiguous situation in which cues for sequential and
simultaneous grouping compete with each other, and where other cognitive processes—
such as attention and expectation—can influence the forming of auditory streams.
Figure 8. Examples of musical structures in 60TET (chord-seq objects) dispatched into five
distinct parts in 12TET (multi-seq objects).
Fusing performance and listening spaces

An important aspect of the relationship between instrument and electronics in this piece
is the development of the approach to spatialisation and the transfer of concepts of spatial
sound synthesis to an electroacoustic setting. Historically, piano music was performed in
intimate settings, often for a small audience located around the instrument. Each of the
listeners had an individual listening position (see e.g. Josef Danhauser’s famous painting
Liszt am Flügel phantasierend). Stockhausen described this situation as “comparable
with people today who wear headphones and completely immerse themselves in the
music” [18]. Inspired by these reflections, my aim was to recreate such an immersive
listening situation in an electroacoustic context: each audience member would have
her personal perspective and experience of the piece, comparable to viewing a physical
artwork from different angles. This required the instrument to be positioned close to
353
Marlon Schumacher
the audience and developing an approach that offers heterogeneous, equally privileged
listening positions, rather than an idealised “projection” of the music which is degraded
for the majority of listeners.
To realise this idea, an unconventional setup was used. The acoustic grand piano
was positioned at the centre of the hall. The audience was seated around it, turned
towards the instrument. Four loudspeakers were placed around the audience, at the
corners of the hall. Each virtual piano part was projected from a dedicated loudspeaker,
which avoided the use of phantom sources and thus the distortion of the spatial image
depending on listener position. The result was an intimate listening situation with
individual auditory perspectives, fusing performance space and listening space. This
setup is shown in Figure 9.
445.11Hz 450.28Hz
440Hz
460.81Hz 455.52Hz
Figure 9. Performance setup for the piece. Note the physical piano at the centre of the
audience and the four loudspeakers at the corners of the hall representing the virtual pianos.
The numbers next to the pianos indicate their diapason in Hertz. (The speakers placed in
between the audience members were used for other electronic parts which are not discussed in
this article.)
354
The spatial disposition and the assignment of micro-tuned chromatic scales to differ-
ent loudspeakers result in a morphological correlation between pitch and spatial position.
If we recall the microtonal structures from Figure 8, it can be seen how adjacent micro-
tonal pitches translate to adjacent positions in space, whereas vertical pitch structures
(chords/harmonies) result in spatial constellations. Each listening position provides an
individual auditory perspective for an acoustic situation in which instrumental sounds
emanating from the centre are complemented by microtonal and spatial extensions in the
electronics.
Spatial sound synthesis as perceptual experience

The spatialisation of individual piano notes, which correspond to the components of
a sound model, can be related to the concept of spatial sound synthesis, first described
in [15]. In simplified terms, the idea of spatial sound synthesis is to consider spatialisation
as a parameter of a sound synthesis process. Rather than spatialising pre-existing sounds
and conceptualising musical space in terms of spatial sound scenes (based on the model
of sound sources in a physical space), spatial sound synthesis is an approach in which
spatial perceptions are created through composition and synthesis of sound components,
such as frequency-domain partials or time-domain grains, spatialised individually. When
cleverly controlled, this allows for the creation of physically impossible sound sources
and auditory illusions. Similar to how the synthesis of frequency structures, such as the
individual partials of a tone complex, can be used to create the sensation of pitch [12], in
spatial sound synthesis the individual spatialisation of sound components can give rise
to the perception of spatial auditory objects.
Indeed, auditory perception can be described as a heuristic process evaluating percep-
tual criteria to build a meaningful mental representation of an auditory scene [3]. Spatial
cues, such as time- and level-differences between ear signals, are part of these criteria,
but can be overridden by other perceptual cues. In the case of incomplete or paradoxical
auditory stimuli the perceptual system follows a “best guess” strategy in which certain
cues can dominate others in favour of the most plausible interpretation [14]. This can
result in auditory conflicts and illusions, such as the “Glissando Illusion” described by
Deutsch [6]. In this experiment, a continuously upwards/downwards gliding sine tone
is abruptly switched between left and right loudspeakers in a stereo setup. Despite the
spatial discontinuity, most people perceive an uninterrupted single glissando, in which
the spatial movement is correlated with the pitch movement. This illusion demonstrates
that the Gestalt principles of continuity and proximity in pitch can override localisation
cues, producing the auditory illusion of a single source that is continuously moving in
space.
Let us consider an excerpt from Ab-Tasten in which a similar auditory illusion is
produced in a musical context. The sonogram of the birdsong from Figure 10 shows
a number of rapidly descending glissandi (chirps). At 6’10” in the piece this birdsong
appears in a dilated and resampled form; it was time-stretched by a factor of 15 (using
a phase-vocoder) and the resulting sound file was then decomposed using a mixed
dictionary of the five corpora of the different pianos. For the temporal locations, the
same evenly spaced temporal grid (16 milliseconds) was used as in the examples shown
in Figure 6. The birdsong was decomposed into 2500 atoms (piano notes), which were
distributed into five parts for the respective pianos. This is another example of how
355
Marlon Schumacher
resolution and scale changes the mode of perception: the short-time morphological
features of the birdsong (e.g. pitch contour, amplitude envelope) are changing so slowly
that they shift into the macro-scale of the score, while the decomposition at 15 times
higher temporal resolution brings the micro-scale details in the fine structure of the sound
to the foreground, akin to a “zoom” effect. Figure 10 shows an excerpt from the piece
(6’47”–7’00”) displayed in five chord-seq objects (representing the physical piano and the
four virtual pianos). The rapid chirps from the original birdsong now resemble slowly
modulating, granular glissandi, traversing the individual piano parts.
Figure 10. Atomic model of a time-stretched birdsong displayed in 5 staves (chord-seq objects)
representing the physical and virtual piano parts. The labelled brackets indicate individual
glissandi.
This perceptual situation produces a striking effect: although the individual piano
sounds emanate from discrete spatial locations, their high repetition rate, as well as
the proximity and continuity of pitch, create the perception of distinct streams of piano
sounds that seem to move continuously through the hall and in between the acoustic and
virtual pianos. The global impression of this sonority can be described as an amorphous
granular sound texture in which individual sound sources seem to emerge, move through
space and disappear again. This auditory illusion can be explained by the general
preference of the perceptual system for organising sound events in a way that yields
the simplest interpretation of an auditory scene [14]. As in the Glissando Illusion, the
sounds produced at the different physical locations are timbrally quasi identical, making
it ambiguous whether sound events have been produced by distinct, static sound sources,
or by a single source that has changed its position. As a result, the perceptual system
356
prioritises other cues for forming auditory streams. Rather than parsing the acoustic
information in a consistent way that would create complicated temporal/harmonic pat-
terns for each physical sound source, it trades off the misinterpretation of spatial cues in
favour of forming more plausible, simple streams based on continuity and proximity in
pitch and time. The resulting effect is an illusion of spatial movement correlated with
pitch.
The perception of spatial depth and distance might be attributed to the similarity
between the characteristic changes of sound features as a function of distance to the
listener, and as an effect of dynamics on the piano. Indeed, studies of piano timbre have
revealed semantic congruencies and similar pianistic performance strategies to express
the labels “distant” (a spatial attribute), “dark”, and “muddled” (timbral attributes) [1].
Two important perceptual cues to determine the distance of a sound source are its overall
level (which decreases with distance due to the spreading of the wavefront) and relative
high-frequency energy (which decreases with distance due to atmospheric absorption by
water molecules) [21]. A similar change in sound characteristics can be observed as a
function of dynamics on the piano: the lower the velocity when striking a key, the lower
will be the overall level as well as the relative high-frequency energy (spectral centroid)
of the sound. These correlated spectral and level differences of piano tones might be
interpreted by the perceptual system as distance cues, such as a pp note on the piano
sounding more distant as compared with a ff note.4
Closing remarks
In this chapter I discussed some of the conceptual implications of abstract sound repre-
sentations as compositional models and introduced a corpus-based, atomic representation
that establishes a direct link between a concrete sound phenomenon and a collection of
sounds. Combined with the possibilities offered by the robotic grand piano, this model
allows an integrated approach for the composition of materials for both acoustic and
electronic parts by extracting structures at different resolutions and time scales, and
with different sound corpora.
The electronics were conceived as virtual microtonal copies of the acoustic instru-
ment, creating a perceptual ambiguity between acoustic and electronic sounds, between
cumulative whole and constituent parts, and between reality and imagination. Using a
specific spatial disposition of loudspeakers, acoustic instrument, and listeners, together
with a morphological correlation between pitch and spatial location, it was possible to
exploit principles of auditory organisation to create the illusion of depth and spatial
movement and transfer concepts of spatial sound synthesis to an electroacoustic setting.
The libraries OM-Pursuit and OM-SoX provided the functionalities to realise artistic
ideas that would have otherwise been difficult to achieve.
Although it might seem that CBAD is merely another analysis/synthesis technique,
I believe its true potential lies in its conceptual and compositional implications. The
process of atomic modelling raises interesting questions related to context, resolution, and
scale as constructors of modes of perception and, consequently, semantics of materials.
4Abinaural recording of the premiere of Ab-Tasten is available on

http://soundcloud.com/marleynoe/sets/ab-tasten.
357
Marlon Schumacher
Similar to how spectral representations paved the way for new approaches, the atomic
model offers an alternative paradigm that may inspire new directions for compositional
thinking.
The German title Ab-Tasten has multiple meanings: its literal translation means
“sampling”, as in sampling an audio signal. In this spelling, it also refers to the piano’s
performance interface, meaning “removing the keys”, in the sense of a renewed interpre-
tation of the piano as an acoustic synthesiser. Lastly, it can be interpreted as “touching”,
“sensing”, or “exploring”, as in exploring unknown lands.
References
[1] Michel Bernays. The expression and production of piano timbre: gestural control
and technique, perception and verbalisation in the context of piano performance and
practice. Ph.D. thesis, Université de Montréal, 2013.
[2] Graham Boyes. Dictionary-Based Analysis/Synthesis and Structured Representa-

tions of Musical Audio. Master’s thesis, McGill University, 2011.
[3] Albert S. Bregman. Auditory Scene Analysis: The Perceptual Organization of Sound.
Cambridge, MA: MIT Press, 1994.
[4] Jean Bresson, Carlos Agon. “Musical Representation of Sound in Computer-Aided

Composition: A Visual Programming Framework”. Journal of New Music Research,
36(4), 2007.
[5] Jean Bresson, Carlos Agon, Gérard Assayag (eds.) The OM Composer’s Book 2.
Editions Delatour France/IRCAM-Centre Pompidou, 2008.
[6] Diana Deutsch. “Grouping Mechanisms in Music”. In Diana Deutsch (ed.) The
Psychology of Music. San Diego: Academic Press, second edition, 1999.
[7] Aaron Einbond, Diemo Schwarz, Jean Bresson. “Corpus-based Transcription as

an Approach to the Compositional Control of Timbre”. In Proceedings of the
International Computer Music Conference. Montreal, 2009.
[8] Willis D. Ellis (ed.) A Source Book of Gestalt Psychology. Abingdon: Routledge,
1999.
[9] Ben Hackbarth, Norbert Schnell, Philippe Esling. “Composing Morphology:

Concatenative Synthesis as an Intuitive Medium for Prescribing Sound in Time”.
Contemporary Music Review, 32(1), 2013.
[10] Stéphane Mallat, Zhifeng Zhang. “Matching Pursuits with Time-Frequency

Dictionaries”. IEEE Transactions on Signal Processing, 41(12), 1993.
[11] Sune Hye Park. Elements of Impressionism evoked in Debussy and Ravel’s “Reflets
dans l’eau” and “Jeux d’eau”: The theme of water. Ph.D. thesis, University of
Washington, 2012.
358
[12] Jean-Claude Risset. “Pitch Control and Pitch Paradoxes Demonstrated with
Computer-Synthesized Sounds”. Journal of the Acoustical Society of America,
46(1A), 1969.
[13] Winfried Ritsch. “Robotic Piano Player Making Pianos Talk”. In Proceedings of the
Sound and Music Computing Conference. Padova, 2011.
[14] Angélique A. Scharine, Tomasz R. Letowski. “Auditory Conflicts and Illusions”. In
Clarence E. Rash, Michael B. Russo, Tomasz R. Letowski, Elmar T. Schmeisser
(eds.) Helmet-Mounted Displays: Sensation, Perception and Cognition Issues. Fort
Rucker, AL: US Army Aeromedical Research Laboratory, 2009.
[15] Marlon Schumacher, Jean Bresson. “Spatial Sound Synthesis in Computer-Aided
Composition”. Organised Sound, 15(03), 2010.
[16] Xavier Serra, Julius Smith. “Spectral Modeling Synthesis: A Sound Analy-
sis/Synthesis System Based on a Deterministic Plus Stochastic Decomposition”.
[17] Robert Silver. “Digital composition of a mosaic image”. US Patent Office, 2000.
[18] Karlheinz Stockhausen, Jerome Kohl. “Clavier Music 1992”. Perspectives of New
Music, 31(2), 1993.
[19] Bob L. Sturm. “Adaptive Concatenative Sound Synthesis and Its Application to
Micromontage Composition”. Computer Music Journal, 30(4), 2006.
[20] Bob L. Sturm, Curtis Roads, Aaron McLeran, John J. Shynk. “Analysis, Visu-
alization, and Transformation of Audio Signals Using Dictionary-Based Methods”.
Journal of New Music Research, 38(4), 2009.
[21] Pavel Zahorik, Douglas S. Brungart, Adelbert W. Bronkhorst. “Auditory distance
perception in humans: A summary of past and present research”. Acta Acustica
United with Acustica, 91(3), 2005.
[22] Aymeric Zils, Francois Pachet. “Musical Mosaicing”. In Proceedings of the
International Conference on Digital Audio Effects (DAFx). Limerick, 2001.
Online
[23] Marlon Schumacher. “OM-Pursuit: Dictionary-Based Sound Modelling in

Computer-Aided Composition”. http://www.idmil.org/software/OM-Pursuit.
[24] Marlon Schumacher. “OM-SoX: Multichannel Audio Manipulation and Functional
Batch Processing in Computer-Aided Composition”.
http://www.idmil.org/software/OM-SoX.
Acknowledgements: This work would not have been possible without the analytical spirit and creative
input of Graham Boyes. I would like to express my gratitude to Philippe Depalle for enlightening
discussions on the signal processing aspects and to Jean Bresson for support with the implementation
in OpenMusic.
359
Appendix
OpenMusic
OpenMusic is a visual programming environment designed for computer-aided music

composition. This environment can be regarded as a graphical interface and superset of
the Common Lisp programming language. It allows composers to develop functional
processes for generating or transforming data in order to produce musical material
(chords, scores, sounds, etc.)
The basic terms and concepts of this environment are explained here in order to
facilitate the understanding of the texts and illustrations in The OM Composer’s Book.
Visual programs/Patches
A patch is a visual graph representing a program. This is the main programming
interface displayed in the figures of this book. In a patch, boxes are connected, or
“patched” together, in order to define the functional layout of the program. Each box
represents a function and has a set of inputs (at the top of the box) that correspond
to the function’s arguments (or parameters) and a set of outputs (at the bottom of
the box) that correspond to the returned/computed value(s).
The evaluation of a box refers to the operation of executing this function to compute
and return values. If an input of the box is connected to the output of another box, it
means that the value of the second box must be computed and used as an argument for
the first function.
Figure 1 shows a simple patch using the functions for the addition (+) and multipli-
cation (×) of numbers. It is equivalent to the expression (3 + 6) × 100. The value of the
box × at the bottom is the result of the execution of the function × with two arguments:
the first argument is the value of the box +, and the second argument is 100. The value
of the box + is the result of the execution of + with two arguments, 3 and 6.
Figure 1. An OpenMusic patch that corresponds to the expression (3 + 6) × 100.
OpenMusic provides a number of built-in functions of varying complexity and spe-

cialisation. Others may be created by the users either graphically or by programming
textually in the Common Lisp language.
363
OpenMusic
Abstractions
If the number 3 in the expression (3 + 6) × 100 is made into a variable, the previous
program becomes a single-parameter function defined as: f (x) = (x + 6) × 100. This
simple procedure is called functional abstraction.
Abstraction in OpenMusic consists of creating custom functions graphically. In
Figure 2 (a) the variable x is represented by an arrow-shaped input-box visible at the
top-left of the window. The output-box (at the bottom) corresponds to the value that
will be returned.
The patch (or abstraction) can then be used as an internal function in other Open-
Music patches, as shown in Figure 2 (b). The value(s) connected to the patch box
input(s) will then be bound to the corresponding variable(s).
(a) (b)
Figure 2. (a) The patch patch1 corresponds to the function f (x) = (x + 6) × 100. (b) patch1
is used as a sub-patch in another patch. The value 5 is bound to the patch input. This second
patch corresponds to the expression f (5)/10 = (5 + 6) × 100/10.
Users usually refer to internal abstractions as sub-patches. Within the limits of

what can be set down on paper, the articles in this book contain various illustrations
for purposes of explaining compositional processes in different levels of detail. These
illustrations may therefore refer to the main patches used to implement these processes
or to their various internal sub-patches.
364
OpenMusic
Objects
Data structures can be used in OpenMusic patches in order to create and manipulate
objects (particularly musical ones). In terms of object-oriented programming, the
definition of an object is referred to as a class, and the actual object created from a
given class is called an instance.
Object constructors are also represented as graphical boxes in OpenMusic. The
inputs and outputs of these boxes represent the attributes of the object (called the slots
of the class). The evaluation of an object box produces an instance of the corresponding
class: inputs are used as arguments for the construction of such an instance, while outputs
allow access to the values of these attributes for the last created instance.
The patch in Figure 3 contains two boxes corresponding to the class note. The first
input (and the first output) of an object box always corresponds to the instance itself
(the input allows the creation of a new instance by copying an existing one, and the
output simply returns the created instance). The other inputs (and outputs), from left
to right, correspond to the pitch, duration, intensity, and MIDI channel of the note. The
pitch input of the upper note box in Figure 3 is connected to the value 6700 (which
corresponds to G4).1 This value is used by the box +, which increments it by 1200 (12
semitones = one octave). The note box at the bottom of the patch uses this new pitch
value and creates a new note instance (G5).
Figure 3. Musical objects.
Object boxes are generally associated with a graphical editor that enables the display
and manual editing of musical data.
1 Pitchesin OpenMusic are expressed in midicents, that is, using the MIDI conventional values (one
unit = one semitone, 60 = middle C) multiplied by 100, in order to allow for the manipulation of
micro-intervals as small as 1/100th semitone. Throughout this book we follow the convention MIDI
note 60 = C4.
365
OpenMusic
The main types of objects available and generally cited in this book are:
• note: A simple note (see previous example).
• chord: A set of simultaneous notes. A chord is defined with lists of values for
each of its attributes (list of pitches, list of durations, list of dynamics, etc.).
• chord-seq: A sequence of chords labelled with absolute time values in millisec-
onds. The chord-seq is also defined with separate lists for each of its attributes
(lists of lists of pitches, durations, dynamics, etc.) and with a list of onsets for
setting the time-position of each chord.
• voice: A sequence of chords expressed using traditional rhythmic notation. The
voice is defined with a list of chords and a rhythm tree determining the metrical
and rhythmic structure.
• multi-seq/poly: Polyphonic objects made up of several chord-seqs, or voices,

respectively.
• bpf: break-point function – A function defined as an ordered sequence of (x, y)
pairs. Defined with two separate lists (x-points and y-points).
• bpc: break-point curve – A sequence of unordered 2D points defining a curve.
Also defined with two separate lists (x-points and y-points).
• bpf-lib, bpc-lib: Collections of bpfs or bpcs.
• n-cercle: Circular representation of pitch class sets or other periodic structures.
• textfile: Editable text buffer, manipulable from visual programs and sometimes
linked to a file on disk.
• midifile/sound/sdiffile: File pointers linking respectively to MIDI, Audio, or
SDIF files, which make it possible to view and access their contents from Open-
Music patches.
366
OpenMusic
Higher-order functions
In functional programming, the term “higher-order” designates a function that generates
another function, or uses a function as one of its arguments. A typical example is the
sort function, which sorts a list according to a given criterion. This criterion can be
any function allowing the comparison of two elements of the list: it can be a simple
test of magnitude (< or >) for comparing numbers, or any other more or less complex
test, depending on the nature of the elements in the list. (Imagine, for instance, a
comparison of harmonicity between chord objects.) This test function could therefore be
(and actually is) a parameter of the generic sort function.
Other frequent higher-order functions are mapping utilities such as mapcar. In
functional programming a mapping consists of applying the same function to all of the
elements of a list: the arguments of the mapcar function are therefore another function
followed by a list of parameters (see Figure 4).
In OpenMusic patches, this notion of higher-order functions is realised through the
use of lambda boxes. A lambda box is a standard function or abstraction box set in a
special state (called “lambda”). When a box is in this state, a small λ icon is displayed
in its upper-left corner. The evaluation of a lambda box does not return the result from
the box computation, but the function defined by this box. The patch in Figure 4 shows
a function box used in the “lambda” mode as an argument of mapcar. The mapcar box
applies this function to the elements of its second argument (the list of pitches at the top
of the patch).
Figure 4. Mapping the function om+ to the elements of a list. The function om+ generates
a list of pitches by adding its two arguments, which can be either numbers or lists of numbers.
In the “lambda” mode this box, connected to (0 700 1200), represents “the function applying
om+ with (0 700 1200)”; that is, f (x) = om + (x, [0, 700, 1200]). The mapcar evaluation returns
the list of resulting lists (a list of lists of pitches, that is, a sequence of chords). The two object
boxes in this patch refer to the class chord-seq.
367
OpenMusic
Control and iterations

The usual control structures used in conventional programming languages (loops, condi-
tional expressions, etc.) also exist in OpenMusic.
An omloop is a special patch allowing one to program iterations (or loops). The
patch in Figure 5 is an example using an omloop to process the same list as in the previ-
ous example in a slightly more complex way. The incoming list of values is represented by
the input arrow-shaped box in the omloop program window. The list-loop box outputs
the elements of this lists one by one at each iteration (6000 at the first iteration, 6500 at
the second iteration, etc.)
The omif box represents the conditional (if ) statement. At each iteration the current
element from list-loop is tested at the left (first input of omif ): if the value is found among
the pitches of the chord object, then this value is returned as such (the second input
is connected directly to list-loop); else (third input) the previous chord construction
process (using om+) is applied. Each time, the result is collected into a new list by the
collect box. Once the last value in the input list has been reached and processed, the
collected list is pulled by the finally box, which represents the result of the omloop.
Figure 5. The omloop box carries out an iteration on a pitch list in order to enrich each
element by two additional pitches. A test (using omif ) decides whether a given pitch should be
processed or not.
368
OpenMusic
Maquettes
The maquette is an original interface designed for the purpose of unifying the concepts
of program and score in OpenMusic. It is a 2-dimensional space in which graphic boxes
(called temporal boxes) can be laid out and organised in time. A temporal box may
contain an isolated musical object, a program (patch) that generates an object, or it
can contain an internal maquette (which enables the user to construct hierarchical
structures).
OpenMusic users typically use this interface to sketch and program the general
structure or form of musical sequences or pieces.
The horizontal axis of the maquette represents time in such a way that the position
and size of the graphic boxes correspond to offset and duration values. The vertical axis
is a dimension that can be used freely in the calculation of the temporal boxes. Temporal
boxes may also be connected to each other and evaluated as standard boxes in a visual
program. Figure 6 is an example of a maquette in which temporal boxes arranged
along the time axis are functionally connected.
Figure 6. Example of a maquette. Each box contains a program computing a musical object.
369
OpenMusic
Extensions and User Libraries

As a visual programming language, OpenMusic provides all the programming possibil-
ities of his own underlying language (Common Lisp). OpenMusic users can therefore
create musical processes or functions using patches, like the one presented above and
throughout the chapters of this book, or directly by programming in Lisp.
Programmed Lisp functions can be easily integrated in an OpenMusic patch in the
form of a box. User-defined functions are usually grouped in libraries, which can be
dynamically loaded in the OpenMusic environment. References to such libraries are
frequently found in this book. Here is a list of some of the OpenMusic libraries cited:
• Profile is mostly dedicated to the manipulation of pitches, using geometric repre-

sentations and transformations.
• OMChaos is a library of chaotic and fractal functions, containing tools for the
deterministic generation of data using nonlinear models and auto-similar structures.
• OMAlea contains a set of functions for generating aleatoric sequences following
different probability distributions and random walk models.
• OMRC is a library allowing the user to describe and solve rhythmic constraint-
satisfaction problems.
• Morphologie is a set of tools and algorithms for the analysis, classification,
recognition, and reconstruction of symbolic sequences.
• LZ is a library for style modelling using statistical tools, permitting the analysis
and generation of sequences in the “style” of a reference model.
• OM-SuperVP is dedicated to the control of the SuperVP sound processing en-
gine. SuperVP is the main kernel used by AudioSculpt software for performing
sound analyses and transformations such as time stretching, transposition, filtering,
cross-synthesis, etc.
• OM-pm2 is dedicated to the control of the pm2 engine, another digital sound
processing kernel mostly used for partial tracking analysis (see p. 374).
• OM-Spat provides tools to connect sound sources to spatialisation parameters and
trajectories, in order to spatialise audio files in multiple channels using the Spat
spatialisation software package.
• OM2Csound allows one to format Csound score and instrument files, and to
perform calls to the sound synthesis kernel from OpenMusic patches.
• OMChroma is a library dedicated to the high-level control of sound synthesis.
The library contains a set of classes corresponding to predefined synthesis processes
(Csound orchestras) to be parameterised and controlled in OpenMusic.
• OMPrisma is a library for spatial sound synthesis extending OMChroma with
classes and processing tools dedicated to spatialisation.
• OM-SoX permits multichannel audio manipulations and batch processing using
the SoundeXchange sound processing kernel.
370
OpenMusic
Online resources
Please consult the OpenMusic website for more information, links to scientific publica-
tions, and download/installation instructions:
http://repmus.ircam.fr/openmusic/
A complete online User Manual is available at:

http://support.ircam.fr/docs/om/om6-manual/
371
Computer music tools and technology
Computer music tools used to complement OpenMusic are cited at different places in
this book. This section provides a brief description and pointers to these technologies
and software.
Software cited
Max [Cycling’74]. Real-time graphical environment for music and audio processing,
widely used for the development of interactive music applications.
https://cycling74.com/products/max/
A number of external objects and libraries for Max are also cited, such as:
• bach, a library of patches and externals for music notation, sequencing, and
computer-aided composition.
http://www.bachproject.net/
• MuBu (Multi-Buffer), a container for sound and motion data providing structured
memory for the processing of recorded data.
http://forumnet.ircam.fr/product/mubu-en/
• Antescofo, a score following system and synchronous programming language for
computer music composition and performance.
http://forumnet.ircam.fr/product/antescofo-en/
Patchwork. Graphical programming environment designed for musical applications.
Antecedent of OpenMusic, developed at IRCAM by Mikael Laurson, Jacques Duthen,
Camilo Rueda, Gérard Assayag and Carlos Agon. Not available anymore on modern
computer platforms.
PWGL. A modern version of Patchwork based on OpenGL. PWGL includes a number of

specialised packages for music notation, constraints programming, or real-time synthesis.
Developed by Mikael Laurson, Mika Kuuskankare, Vesa Norilo, and Kilian Sprotte.
http://www2.siba.fi/PWGL/
Csound. Open-source programming language designed and optimised for sound render-
ing and signal processing.
http://www.csounds.com/
CataRT [IRCAM]. Real-time concatenative sound synthesis system, which selects and
plays grains from a corpus of recorded audio, segmented and analysed by descriptors,
according to proximity to a target position in descriptor space. Available both as a
standalone application and as a library for Max.
http://ismm.ircam.fr/catart/
373
AudioSculpt [IRCAM]. Sound editing and processing program based on the short-term
Fourier transform representation. Provides graphical editing tools for sound analysis,
filtering, cross synthesis, time stretching, transposition, segmentation, etc.
http://forumnet.ircam.fr/product/audiosculpt-en/
The main use of AudioSculpt (and Spear, see below) with OpenMusic consists of
generating spectral analyses that are exported and loaded as input data in the computer-
aided composition environment. The analysis can be more or less precise and faithful
with regard to the original sound, and yield more or less dense sets of functions of time,
frequency, and amplitude (called partials), imported in OpenMusic as SDIF or text files
(see p. 375). Two principal types of analysis are used:
• Partial tracking analyses model the sound signal as a sum of pseudo-sinusoidal

signals which can vary in frequency and intensity.
• Chord-sequence analyses rely on a preliminary time-segmentation of the input

sound and estimate the main frequency components in each segment. The result-
ing data comprises simple time, frequency, and amplitude values, similar to the
description of notes in a musical score.
Spear. An application for audio analysis, editing, and synthesis based on the “partial
tracking” technique, representing a sound with many individual sinusoidal tracks.
Developed by Michael Klingbeil.
http://www.klingbeil.com/spear/
Modalys [IRCAM]. Physical modelling sound synthesiser. Modalys is also the name
of the OpenMusic library that works as a front-end to the sound synthesis kernel.
http://forumnet.ircam.fr/product/modalys-en/
Orchids, ATO-MS, Orchidée [IRCAM]. Computer-assisted orchestration and timbral

mixture optimisation systems. Orchids is the latest generation of orchestration software,
a successor to ATO-MS and Orchidée.
http://forumnet.ircam.fr/product/orchids-en/
Music notation software
Finale [MakeMusic]. http://www.finalemusic.com/

Sibelius [Avid]. http://www.avid.com/US/products/sibelius/
Digital Audio Workstations (DAWs)
Digital Performer [MOTU]. http://www.motu.com/products/software/dp

Reaper [cockos]. http://www.reaper.fm/
374
Data interchange formats and protocols

MIDI. Musical Instrument Digital Interface. Standard message specification and storage
format for the sharing and transfer of musical data between digital instruments and
computer systems.
MIDI Manufacturers Association: http://www.midi.org/
OSC. Open Sound Control. Protocol for communication of music and media data
through UDP among computers, sound synthesisers, and other multimedia devices.
Originally developed at CNMAT, UC Berkeley.
http://opensoundcontrol.org/
SDIF. Sound Description Interchange Format. Standard for the interchange of sound
description data, consisting of a basic data format framework and an extensible set of
standard sound descriptions (including spectral, sinusoidal, time-domain, and higher-
level models). Created by IRCAM, CNMAT, and IUA-UPF.
http://www.ircam.fr/sdif
MusicXML [MakeMusic]. Open file format for music notation, based on XML (Exten-
sible Markup Language). MusicXML is used as an import/export format for scores in
many music notation programs.
http://www.musicxml.com/
375
About the authors
Julián Ávila is currently completing his PhD at NOVARS Research

Centre, University of Manchester, UK on the topic of “Spectral Diffusion
and Spectral Energy in Electroacoustic Composition”. His research
focuses on electroacoustic composition and spatialisation as well as
multidisciplinary projects, which he has pursued at IRCAM in Paris,
University of Alcala de Henares in Madrid, and CDMC (Centre for the
Diffusion of Contemporary Music) in Madrid. Julián holds a degree in
composition and saxophone from RCSMM (Madrid Royal Conservatory)
and a master’s degree in performing arts from URJC (King Juan Carlos University). He
has taken the advanced contemporary composition course at CSMA as well as a course
in audio production at the Centre of Technology for the Performing Arts of INAEM
(Ministry of Culture, Spain). He received the Francisco Guerrero Marín 2013 award of
the Fundación SGAE-CNDM, the Composer’s Marathon V 2013 (Vienna), the national
first prize in the competition INJUVE Creation 2012, first prize in Vacances Percutantes
2011 (Bordeaux, France), second prize in INJUVE Creation 2011, and first prize in
Hui Hui Música 2007 (Valencia, Spain). He currently works as Graduate Teaching
Assistant at NOVARS, lecturer of Electroacoustic Composition at CSMCLM (Castilla la
Mancha Conservatory), and invited lecturer on the Master of Ephemeral Arts at Madrid
Polytechnic University.
Philippe Boivin was born in 1954. He studied musicology at the

Sorbonne, harmony at the Paris Conservatoire, and composition with
Max Deutsch. The SACEM awarded him a prize for best pedagogical
composition in 1985 as well as two further prizes for composition: the
Georges Enesco Award in 1988 and the Pierre and Germain Labole Award
in 2002. Thanks to his varied musical activities, Philippe Boivin belongs
to a generation of composers for whom the act of creation is not a
restricted field of experience. Mostly focused on chamber music, his work shows a great
diversity of interests. His entire production, however, can be defined by one term: rigour,
as revealed both in the precision of his style as well as in the thoroughly elaborated forms
into which he pours his music. This quality, however, does not hinder the instruments
from expressing lyrical, dramatic, and even theatrical affects.
[Cécile Gilly, Éditions Salabert]
377
About the authors
Federico Bonacossa is a composer and guitarist based in Miami,

Florida. He studied classical guitar at the Conservatorio Statale
Giovanni Pierluigi da Palestrina in Italy before moving to the United
States in 2001. He holds a master’s degree from the Peabody
Conservatory, a doctorate from the University of Miami in classical
guitar performance and music theory, as well as a master’s degree
from Florida International University where he studied composition and
electronic music. His recent work as a composer explores various forms of interaction
between live performers and the computer, the relationship between pitch and rhythm,
and the computer-aided transcription of spontaneous vocal gestures. He is the company
composer for Dance NOW Miami and is a member of the Miami Guitar Trio. As a
performer he is involved in promoting new music for guitar, especially works that feature
electronics. He currently teaches Music Theory and Musicology at Florida International
University.
Jean Bresson is a researcher specialising in computer-aided composition.

He studied sciences and specialised in computer science at the Universities
of Toulouse and Nice, France; Granada, Spain; and holds a PhD from the
Université Pierre et Marie Curie, Paris. He is currently the main researcher
and developer in charge of the OpenMusic environment at IRCAM.
His research projects and software development have mostly concerned
extended applications of computer-aided composition in the fields of sound
synthesis and processing, sound spatialisation, or more recently, interaction and reactive
systems. In 2013 he assisted Marco Stroppa for aspects of computer-aided composition
in the opera Re Orso.
Luiz Castelões is a Brazilian composer, Professor at Universidade Federal

de Juiz de Fora, Brazil since 2009, and was a postdoctoral researcher at the
Integra Lab, Birmingham Conservatoire, UK in 2015-16. He holds degrees
from Boston University (DMA, 2009) and UNIRIO (MM, 2004, and BM,
2001) as well as prizes, residencies, and grants from CMMAS (Mexico),
CAPES/Fulbright, UFRJ School of Music (honourable mention at their
1st National Composition Contest, 2012), Festival Primeiro Plano (Prize
for Best Sound Editing, 2003), and Funarte (1st prize in the XIV Brazilian
Contemporary Music Biennial, 2001). Recent performances and recordings of his music
have been given by the Szlachta String Quintet (USA, 2015), Duo Promenade Sauvage
(Italy, 2015), Quartetto Maurice (Italy, 2014), and Ensemble Arsenale (Italy, 2013).
He has recently published articles in Sonic Ideas (Mexico, 2015), El Oído Pensante
(Argentina, 2013), and the International Review of the Aesthetics and Sociology of
Music (IRASM, Croatia, 2009). His research interests include musical onomatopœia,
algorithmic composition, image-to-music conversion, and popular music.
378
About the authors
Fabio De Sanctis De Benedictis was born in Pisa in 1963.

He graduated in Livorno with degrees in violin, choral music and
conducting, and composition (the latter cum laude under the guidance
of Claudio Vaira). At the same time he attended Giacomo Manzoni’s
classes at the Fiesole Music School. He has won several national and
international composition competitions, published music theory research
in specialised international journals, and presented lectures and seminars
throughout Europe. His works have been performed in Italy, Croatia,
Portugal, the United Kingdom, Spain, and Belgium. Since 1992 he has taught at several
Italian conservatories and at Pisa University. At present he teaches harmony, musical
analysis, and fundamentals of composition at ISSM “Pietro Mascagni” in Livorno and
continues his work in composition, electronic music, and musical analysis.
Aaron Einbond’s work explores the intersection of instrumental music,

sound installation, field recording, and technology, bringing the spontane-
ity of live performance together with computer interactivity to impact and
challenge the listener. His recent music has focused on audio transcription
as the centre of a creative process bridging composition, improvisation,
and interpretation, questioning the thresholds of perception between
instrument, stage, room, and loudspeaker. In 2014 Chicago-based
Ensemble Dal Niente released his portrait album Without Words on Carrier Records
and he became Co-Artistic Director of Qubit New Music Initiative with whom he curates
and produces experimental media in New York. Einbond currently teaches composition
at City University of London and has taught at Columbia University, the University of
Huddersfield, and Harvard University as well as held a John Simon Guggenheim Memorial
Foundation Fellowship and an Artistic Research Residency at IRCAM. He was born in
New York in 1978 and studied at Harvard University, the University of Cambridge, the
University of California Berkeley, and in the Cursus at IRCAM with teachers including
Mario Davidovsky, Julian Anderson, Edmund Campion, and Philippe Leroux.
Alireza Farhang, Iranian-French composer, was introduced to music

at a young age as he grew up in a family of musicians. He took piano
classes with Emmanuel Melikaslanian and Raphaël Minaskanian and
studied composition with Alireza Machayeki at the University of
Tehran. He pursued further studies in orchestration with Michel
Merlet at the École Normale de Musique de Paris and continued his
composition studies with Ivan Fedele at the Conservatoire National
Régional de Strasbourg. He was part of the inaugural programme of the European Course
for Music Composition and Technologies (ECMCT) jointly developed by IRCAM and the
Technische Universität, Universität der Künste, and Hochschule für Musik Hanns Eisler
in Berlin. He has worked closely with Brice Pauset, Toshio Hosokawa, Kaija Saariaho,
Michael Jarrell, Yan Maresz, Gérard Pesson, and Tristan Murail.
379
About the authors
Ambrose Field is a British composer and Head of Department

of Music at the University of York, UK. His music is recorded
on ECM (Munich) and Sargasso (London), and broadcast by the
BBC and other international networks (SVR, RTE, ORF). His
works have been performed at the Vienna Konzerthaus, the Chicago
Early Music Festival, Parco della Musica Rome, the A-Cappella
Festival Leipzig, the Perth International Festival Australia, Konvergencie Bratislava,
Kultursommer Rheinland-Pfalz Germany, the ICA London, Ultima Norway, the Tampere
Vocal Music Festival Finland, Kryptonale Berlin, CBSO Centre Birmingham, Voce-
Versa Milan, Huset Copenhagen, Warsaw Autumn, and other international venues.
Field studied Education at the University of Cambridge and in 1994 moved to London.
Here, funded by the British Academy, he received a PhD in Composition from City
University. Field served as a board director of the UK’s Sonic Arts Network (forerunner
to today’s Sound and Music Organisation) in the late 1990s. His work with technology
and composition has received several international awards including three Prix Ars
Electronica Honorary Mentions in 2006, 1997, and 1996. He has been a resident artist at
Asphodel/Recombinant Media Labs in San Francisco, the Hochschule für Gestaltung in
Karlsruhe, Germany, and a composer at Hungarian National Radio funded by UNESCO.
Gonçalo Gato was born in 1979 in Lisbon, Portugal, where he initially

studied composition. He moved to London in 2011 to start a doctorate
on the subject of “Algorithm and Decision in Musical Composition”
under the supervision of Julian Anderson at the Guildhall School of
Music and Drama. Recently he was selected to take part in the London
Symphony Orchestra’s Panufnik Scheme to write and publicly rehearse a
new orchestral work. Recent works include A Walk in the Countryside
(2016) for solo flute, developed in collaboration with Martin Fahlenbock of ensemble
recherche; Vacuum Instability (2013), premiered by musicians of the BBC Symphony
Orchestra at Maida Vale studios; Canti Firmi (2014), based on a madrigal by Carlo
Gesualdo and premiered by the Guildhall New Music Ensemble in London; and
Comendador u m’eu quitei (2015), a song based on an Iberian medieval poem by Rui
Pais de Ribela and premiered by the Ensemble MPMP at the Gulbenkian Foundation in
Lisbon, Portugal. His orchestral works include Vectorial-modular (2011), awarded first
prize in the Póvoa de Varzim composition competition, and A Vida é Nossa (2013) for
symphonic wind band, premiered by the Banda Sinfónica Portuguesa at the Casa da
Música in Porto.
380
About the authors
Jean-Luc Hervé was born in 1960. He studied composition at the

Conservatoire de Paris with Gérard Grisey, where he received a Premier
Prix in composition. In 1997 he received the “Goffredo Petrassi” prize
for his composition Ciels for orchestra. He was composer-in-research at
IRCAM and received a fellowship from the DAAD in Berlin (2003). The
profound effect of a residence at Villa Kujoyama in Kyoto, along with a
doctoral thesis in æsthetics and subsequent research at IRCAM, have helped
to shape Hervé’s compositional outlook. He founded the group Biotop(e)
with Thierry Blondeau and Oliver Schneller in 2004. His works have been performed by
ensembles such as Orchestre National de France, Orchestre Philharmonique de Radio-
France, Orchestra Sinfonica dell’Emilia-Romagna “Arturo Toscanini”, Instant Donné,
Court-Circuit, Ensemble Intercontemporain, 2E2M, Contrechamps, Berliner Symphonie
Orchester, KNM Berlin, Musik Fabrik, and Orchestra della Toscana. He is currently a
teacher of composition at the Conservatoire de Boulogne-Billancourt.
Geof Holbrook completed his doctoral studies in composition at

Columbia University and holds an MMus from McGill University in
Montréal, Canada. His works have been performed by the Nouvel
Ensemble Moderne, Wet Ink Ensemble, Esprit Orchestra, the Windsor
Symphony Orchestra, Collectif 9, In Extensio, Toca Loca, Ensemble of
the National Arts Centre Orchestra, Quasar, Sixtrum, and the Ensemble
Orchestral Contemporain in Lyon. He is a recipient of multiple awards
and fellowships, including a Prix Opus for “Création de l’Année” in 2007, and has been
a finalist for the Jules Léger Prize for New Chamber Music. He has participated in
composition courses at Domaine Forget, Royaumont, the National Arts Centre, and
IRCAM in Paris, where he pursued the Cursus computer training programme and later
an Artistic Research Residency.
Matthew Lane, originally from Kingston, Ontario, studied composition

at Mount Allison University in New Brunswick and further at Université
de Montréal under the tutelage of Alan Belkin, Pierre Michaud, Hugues
Leclair, and Robert Normandeau. His compositions include both
instrumental works and mixed instrumental-electronic works, and he
draws material often from Québecois poetry or Swedish folk music. Much
of his work in computer-assisted composition involves the application of
processes to this existing material, with the goal of creating tools further to bridge the
gap between programming and more traditional composition.
Serge Lemouton studied violin, musicology, theory, and composition

before specialising in computer music at the SONVS department of the
Conservatoire National Supérieur de Musique de Lyon. Since 1992 he has
been a computer music designer at IRCAM. There he works with researchers
to develop computer tools and has participated in the production of musical
projects with composers including Florence Baschet, Laurent Cuniot, Michael
Jarrell, Jacques Lenot, Jean-Luc Hervé, Michaël Levinas, Magnus Lindberg,
Tristan Murail, Marco Stroppa, and Fréderic Durieux. He was responsible for the
production and performance in real time of several works by Philippe Manoury such as
K..., La Frontiére, On-Iron, the two Partitas, and the opera Quartet by Luca Francesconi.
381
About the authors
Eduardo Reck Miranda’s distinctive work is informed by his unique

background as an Artificial Intelligence scientist and classically trained
composer with an early involvement in electroacoustic and avant-garde
pop music. Currently he is Professor of Computer Music at Plymouth
University in the United Kingdom, where he founded the Interdisciplinary
Centre for Computer Music Research (ICCMR). The inside story of his
acclaimed choral symphony Sound to Sea is revealed in the book Thinking
Music (University of Plymouth Press), which also includes the full score of
the piece and a CD of the recording of its premiere.
Alessandro Ratoci was born in Tuscany in 1980. After studying piano,

composition, and electronic music in Italy he completed his studies in Geneva
with Michael Jarrel and Louis Naon. In 2014-2015 he was selected for the
IRCAM Cursus in Paris. He is an active composer, electronic performer, and
music teacher. Devoted to the pedagogy of interpretation and improvisation
with live electronics, he is currently teaching at the Haute École de Musique
(HEMU) in Lausanne, Switzerland and at the Conservatorio G. B. Martini
in Bologna, Italy.
Marlon Schumacher holds degrees in music theory/digital media and

composition from the Staatliche Hochschule für Musik, Stuttgart, and a
PhD in Music Technology from McGill University, Montreal. After a visiting
research year at the Input Devices and Music Interaction Lab, he joined the
Expanded Musical Practice group at CIRMMT to work on research projects
for composition and performance of interactive media. In 2009 he initiated
a collaboration with IRCAM’s Music Representations team on compositional control of
spatial sound synthesis. In this context he developed a number of OM libraries dedicated
to gestures, sound processing, corpus-based analysis/synthesis. As composer and digital
artist, Schumacher has realised works for a broad spectrum of instrumentations and
media, exploring the extension of human expression using computational means. Besides
his artistic and scientific work, he is also active as a lecturer in Europe and North America.
In 2015 he joined the Institute for Music Informatics and Musicology of the Hochschule
für Musik, Karlsruhe, as a visiting professor.
Matthew Schumaker is a native of San Francisco, where he is based.

He earned a BA in Music and Philosophy from Dartmouth, an MA in
Music Composition from Princeton, and his doctorate from UC Berkeley
in 2015. While at Berkeley he studied Composition with Professors Edmund
Campion, Cindy Cox, Franck Bedrossian, and Ken Ueno. In 2014 he received
Berkeley’s Georges Ladd Fellowship, allowing him to work in-depth with
composer Martin Matalon in Paris. Matthew is currently a Lecturer at UC
Berkeley.
382
About the authors
Marco Stroppa is a composer, researcher, and teacher who studied music in

Italy (piano, choir direction, composition, and electronic music) and pursued
further studies at the MIT Media Laboratory (computer science, cognitive
psychology, and artificial intelligence). From 1980-84 he worked at the Centre
for Computational Sonology (Padua), where he wrote Traiettoria for piano
and electronics. In 1982 Pierre Boulez invited him to join IRCAM, and his
uninterrupted association there has been crucial for his musical growth. At
IRCAM, his main interest is the compositional control of sound synthesis.
Together with Carlos Agon and Jean Bresson he developed the libraries OMChroma
and OM-Chant. Since its birth in 2007, he has also been strongly involved with score-
following software Antescofo developed by Arshia Cont. A respected educator, Stroppa
founded the composition course at the Bartók Festival (Hungary), where he taught for
13 years. Since 1999 he has been professor of composition in Stuttgart. He also taught
at the Conservatoire National Supérieur de Musique et de Danse de Paris (CNSMDP).
[Photo by Roberto Masotti,
c Casa Ricordi, Milano]
Christopher Trapani was born in New Orleans, Louisiana. He earned

a Bachelor’s degree from Harvard, then spent most of his twenties in
Europe: a year in London, working on a master’s degree at the Royal
College of Music with Julian Anderson; a year in Istanbul, studying
microtonality in Ottoman music on a Fulbright grant; and seven years
in Paris, where he studied with Philippe Leroux and worked at IRCAM,
both on the composition Cursus and a six-month Musical Research
Residency. He is currently based in New York City. Christopher is the winner of the
2007 Gaudeamus Prize. His scores have been performed by Ensemble Modern, ICTUS,
Ensemble L’Itinéraire, Nieuw Ensemble, Talea Ensemble, and the JACK Quartet,
amongst others. Recent projects include a new string quartet with electronics for Quatuor
Béla and Grame (Lyon), and a commission for orchestra and electronics (IRCAM) for
the Orchestre Philharmonique de Radio France and Festival Présences 2015.
Takéshi Tsuchiya majored in composition and music theory at

the Graduate School of the Tokyo College of Music. He studied
composition under Joji Yuasa, Shin-ichiro Ikebe, and Akira Nishimura.
He additionally studied Conducting under Sei-ichi Mitsuishi and Ondes
Martenot under Takashi Harada. He was finalist of the 13th Japan
Society for Contemporary Music New Composer Competition and finalist
of the 66th Japan Music Competition. In 2008, he received the Special
Jury Award of the 31st Valentino Bucchi International Composition
Competition. His most important works have been broadcast by NHK among others. As
part of his activities in the field of electronic music, he plays an active part in producing
music with technology, including taking charge of programming and electroacoustics
for works by Toshio Hosokawa, Ichiro Nodaira, and others. He is currently Associate
Professor at the Tokyo College of Music and resides in Tokyo.
383
About the authors
Julien Vincenot was born in 1985 in France. After various instru-

mental studies (latin-american harp, piano, jazz, and improvisation) he
specialised in computer music and electroacoustic composition at Paris 8
University with Anne Sèdes, Horacio Vaggione, and José Manuel López
López. There in 2007 he co-founded the Unmapped collective, dedicated
to improvisation involving instruments and live computers, which today
includes eleven permanent members. In parallel, he studied composition
at the Montbéliard Conservatory with Jacopo Baboni Schilingi, Frédéric Voisin, Lorenzo
Bianchi, and Giacomo Platini. Since 2010 he has also been a member of the international
research group PRISMA. In 2013 he received the first prize in composition from the
SACEM. The same year he attend the Cursus 1 in composition and computer music at
IRCAM under the artistic direction of Hèctor Parra. His works have been presented
throughout Europe, in China, and in the United States.
Anders Vinjar is a Norwegian composer. While studying ethnomusicology

and linguistics he experimented with the potentials of programming
languages and artificial intelligence techniques to work on issues of music
analysis. He became interested in using the same tools to create music,
stopped studying, and started composing. His main interests are acousmatic
music and other electroacoustic art, algorithmic composition, digital signal
processing, and programming for music. He spends most of his composing-
hours either making field recordings or working in functional programming
environments for music such as OpenMusic, Common Music, CLM, SuperCollider, and
other FLOSS-ware. His output includes concert music of various kinds, installations,
music for movies, streams/web-art, hacks, applications, workshops, lectures, and
occasional articles.
Michele Zaccagnini, born 1974, is a composer from Rome, Italy. He

studied Economics at the Universitá La Sapienza and Clarinet at the
Conservatorio Santa Cecilia. He moved to Los Angeles in 2004 to study
film scoring at UCLA and then to Boston where he completed his PhD
in Music Theory and Composition at Brandeis University studying with
Martin Boykan, Yu-Hui Chang, Eric Chasalow, and David Rakowski. His dissertation
composition Variazioni su AlDo ClEmenti has received wide praise and was selected
and performed as a finalist at the International Composition Competition “A Camillo
Togni” by the jury of Enzo Restagno, Mauro Bonifacio, Magnus Lindberg, Gabrio
Taglietti, and Jesus Villa-Rojo. Aside from traditional acoustic and electroacoustic
composition, Zaccagnini is also interested in interactive application of musical algorithms
and has recently developed a musical interface that links meditation practice, an
electroencephalogram device, and generative algorithms (Musically-induced Interactive
Mediation, Mi-iM). The project was presented at the first Brain Computer Musical
Interface Conference at Plymouth University, UK.
384
Dépôt légal : 3ème trimestre 2016
Imprimé en France

OM Composer's Book 3 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

OM Composer's Book 3 PDF

Uploaded by

Copyright:

Available Formats

THE OM COMPOSER’S

The Musique/Sciences series contributes to our understanding of the relationship between

• Gérard Assayag, Frano̧is Nicolas, Guerino Mazzola (dir.)

Variazioni su AlDo ClEmenti

Folk material transformations

Tri Bhuwana for 12 voices and Balinese gamelan:

Programming modular progressions in OpenMusic

Musicalising sonification: Image-to-music

On “slow” computer-aided composition

OM-Darwin: Generative and descriptive

Computer-aided composition in the creation

Materials and techniques in D’improvviso

Musique instrumentale concrète: Timbral transcription

Composing for the resonance: Finding new relationships

Sketching, synthesis, spatialisation:

Dialogue with OpenMusic in the process of composing

Electronic sound creation in Balænoptera for

Modelling a gesture: Tak-Sı̄m

Electronic dramaturgy and

Rima Flow: Oral tradition and composition

Ab-Tasten: Atomic sound modelling

Computer-assisted composition and algorithmic composition are the modern realisations

1A presentation of OpenMusic is given in the Appendix of this book.

Jean Bresson, Carlos Agon, Gérard Assayag, Editors.

[...] our everyday experience in building things tends to give

Figure 1. Bouncing object: space=10, speed=6, duration=2. First four moves.

3 The object’s position is always measured from its left edge.

Figure 2. Scored movement of the example in Figure 1.

• They are periodic.

6 Iterations take place as in stop-frame motion.

Figure 3. The implementation of the “bouncing” formula in OpenMusic.

Figure 5. Multi-seq rendition of “Invenzione”.

So far these processes are relatively straightforward: simple permutations operating

11 Proportional to prime numbers 2, 3, 5, and 7.

Manual complementation: manual composition performed so as to coexist with al-

4 3 2 b˙ 4 ˙ b-œ ™ œ 2 b>œbœ œ >œœœ œ 4 b>œbœœ œ œœ Gonçalo Gato

Clarinete 1 em Sib & 4

Analysis and resynthesis

Figure 2. Transcription of the folk melody Senhora do Almurtão.

Part Melodic Intervals

Table 1. Intervallic analysis of Senhora do Almurtão.

2A música portuguesa a gostar dela própria: https://vimeo.com/mpagdp.

Figure 3. Rhythmic polymerisation from cells and pauses.

Figure 4. Initial brass melody with rhythmic cells in parenthesis.

Figure 6 details the patch intervallic_melody_range, which generates a number of

Figure 6. The patch intervallic_melody_range, generating a sequence of pitches in a given

Figure 7. Rhythm-focusing process by diminution.

5 https://vimeo.com/39572090. Video directed by Tiago Pereira, 2012.

& œ™ œ ˙ œ œ œ™ œ #œ œ™ œ œ œ œ œ™ œ œ œ ™ œ œ Œ ≈ œ œ œ #œ ™ œœœ œ œœœœ

& <#>œ#œ ™ œ œ œ ™ œ œ œ œ œ ™ œ ‰ ‰ ™ œr œ œ œœœ œ œ™ œ œ ˙ œ œ œ™ œ #œ œ™ œ

& <#>œ œ œ œ™ œ œ œ ™ œ œ Œ ≈ œ™j œ™ #œ œ œœœ œ œ™ œ œ ˙ œ œ œ™ œ #œ œ™ œ

& œ œ ™ œ œ ™ œ œ™#œ œ™#œ œ™ œ œ ™ œ œ™ œ œ

& <#>œ œ œ œ™ œ œ œ ™ œ œ Œ ≈ œ œ œ #œ ™ œ œ œ œ œ ™ œ œ ˙ œ œ™ œ œ œ™#œ œ œ œ œ ™ œ œ œ™ œ œ

& <#>œ œ œ œ™ œ œ œ ™ œ œ Œ ≈ œ œ œ #œ ™ œ œ œ œ œ ™ œ œ ˙ œ œ™ œ œ œ™#œ œ œ œ œ ™ œ œ œ™ œ œ

final score (some of them >œcome b œ b œ nœ#œ into sync with

were composed manually (based >œ™ #œ ™

below)& and∑ work as a complement to theœ™

Ob. 1 & œ™ œœ œ- œœ ™ œœ™ œ ˙ œ œ #œ œ- œ œ œ œ- Œ œ œ- #œ œ œ- œ œ ˙ œ œ #œ œ- œ œ

Sax. Alto 2 & œ™ œ˙ œ œ™ œ œ œ™ #œ œ œ- œ œ ™ œ œ œ™ œœ œ- œ ‰™ ‰ œ œ- #œ œ™ œ œ œ- œœ ™ œœ™ œ ˙ œ œ #-œ œ œ œ

While& I was∑ exploring∑ techniques for dealing with the# œ folk

Exploring the same technique, I tried to achieve a degree of harmonic amalgamation

Sax. Alto 1 & ˙ œ œ #œ œ- œ œ œ œ- Œ œ œ- #œ œ œ- œ œ ˙ œ œ #œ œ- œ œ œ œ- Œ œ œ- #œ œ œ- œ œ

Figure 12. Score implementation