1 s2.0 S0003682X19311211 Main

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Applied Acoustics 164 (2020) 107284

Contents lists available at ScienceDirect

Applied Acoustics
journal homepage: www.elsevier.com/locate/apacoust

Sound model of an orchestral kettledrum considering viscoelastic effects


Erik U. Gallardo a, Miguel A. Alonso-Arévalo b, Eloísa García-Canseco c,⇑, Carlos Aguilar-Ibáñez d
a
Earth Sciences Division, Centre for Scientific Research and Higher Education (CICESE), Ensenada, B.C., Mexico
b
Dept. of Electronics & Telecommunications, Applied Physics Division, Centre for Scientific Research and Higher Education (CICESE), Ensenada, B.C., Mexico
c
Faculty of Sciences, Autonomous University of Baja California (UABC), Ensenada, B.C., Mexico
d
Computer Research Centre (CIC), National Polytechnic Institute, Mexico City, Mexico

a r t i c l e i n f o a b s t r a c t

Article history: The modeling of the mechanical systems that describe musical instruments has long been a topic of inter-
Received 9 October 2019 est for acoustic physicists. The orchestral kettledrum, distinguished by a broad spectrum of sounds, has
Received in revised form 3 February 2020 been the subject of multiple investigations related to the goal of generating a realistic digital sound syn-
Accepted 18 February 2020
thesis. In the present work, we apply the Green function method to estimate modal frequencies consid-
Available online 7 March 2020
ering an air-loaded viscoelastic membrane. We propose a method that includes viscoelasticity to
accurately predict the sound spectrum of the modeled kettledrum. Results are compared to real sound
Keywords:
recordings of an orchestral kettledrum obtained in controlled conditions. The calculated modal frequen-
Kettledrum
Green function
cies are found to coincide well with the real values with an absolute mean error of 1:25  0:76 Hz and
Modal synthesis 1:87  1:83 Hz for the A2 and B[2 tuned drumheads respectively. The spectral envelope of the synthesized
Viscoelasticity sound spectrum coincides well with the Fourier transform of the real sound. The viscoelastic term was
found to generally reduce the amplitude of the sound spectrum and in certain cases, better approximate
the modal frequencies and decay times. The modal synthesis method used here is numerically light-
weight and can be adapted to be used in real-time applications with low computational resources. To
reproduce our experiment, the recorded kettledrum sounds and Python source code of the model are
freely available.
Ó 2020 Elsevier Ltd. All rights reserved.

1. Introduction considered to be the underlying percussion instrument in the


orchestra. Moreover, the kettledrum sound can be tuned over a
Drums are considered to be among the first musical instru- range of more than one octave by varying the tension of the mem-
ments ever invented by humans. Throughout history and different brane by using the screws around the circumference or by using a
cultures, drums have played an essential role in music, religious tensioning pedal. Due to the large surface of the drumhead, the
ceremonies, battles, and social gatherings. Nowadays, drums con- striking position plays a crucial role in the resulting sound. As
tinue to drive the rhythm in every performance that includes them described by Rossing [1], in order to generate the desired harmonic
(See Table 1). sound, the kettledrum should be struck about a quarter of the way
It is well known that the sound produced by most drums, from the edge of the drumhead to the center. If the drum is struck
regardless of their size, has no pitch. In the present work, we in the center, other inharmonic fundamental modes will dominate
develop an analytic sound model for a particular type of percussive the sound.
instrument called the kettledrum, also known as the timpani. Although several physical modeling sound synthesis methods of
The kettledrum consists of a drumhead (calfskin or synthetic the kettledrum are available in the literature, the primary motiva-
high tensile strength film) stretched over a large empty bowl tradi- tion behind the present work is the lack of a highly realistic,
tionally made of copper. Its membrane is made to vibrate by the computationally-efficient and freely available model. Our physical
impact of a mallet or drumstick. The interaction of the membrane modeling approach belongs to the category usually known as
with the air on the top and beneath its surface generates the modal synthesis where the complex dynamic behavior of a vibrat-
kettledrum’s unique pitch and timbre [1]. For this reason, it is ing object may be decomposed into contributions from a set of
modes, with each mode oscillating at a single frequency [2].
Related research work has been done in the past. For instance,
⇑ Corresponding author. Christian et al. [3] were the first authors to propose a model for
E-mail address: eloisa.garcia@uabc.edu.mx (E. García-Canseco). the acoustic pressure using the Green’s function method. Their

https://doi.org/10.1016/j.apacoust.2020.107284
0003-682X/Ó 2020 Elsevier Ltd. All rights reserved.
2 E.U. Gallardo et al. / Applied Acoustics 164 (2020) 107284

Table 1 2. Material and methods


List of symbols.

Symbol Meaning Due to the symmetry of the orchestral kettledrum, it can be


Drumhead abstracted as a cylindrical cavity, as shown in Fig. 1. Besides the
P surface of the drumhead mathematical convenience of this abstraction, [7] showed that
t the difference between the calculated frequency modal ratios
q0 obtained using this model and a better-defined kettle-shape model
/0
is small and of little musical significance. The acoustical model of
r0 ¼ ðq0 ; /0 ; tÞ point on P at time t
a radius of the kettledrum the instrument requires solving a coupled boundary problem for
T membrane tension of drumhead (in N/m) membrane vibration and air pressure distribution [4]. To this
r superficial density of drumhead (in kg/m2 ) end, we apply an existing Green’s function method proposed by
m viscoelastic coefficient of drumhead (in seconds) [8], to solve the problem by obtaining natural frequencies and
g transverse displacement of drumhead
decay times of the cavity backed membrane.
xð0Þ
mn
frequency of normal modes in vacuum
xms frequency of normal modes with air loading
2.1. Modeling of membrane-air interaction
Air
Xe external space (outside the kettledrum)
Xi internal space (inside the kettledrum) A circular membrane fixed at the edge and uniformly stretched
q by a tension T is allowed to vibrate in the transversely. As the
/ membrane vibrates, air pressures loads the membrane from out-
z
side, due to ambient atmosphere, and from inside, due to the
r ¼ ðq; /; zÞ point in R3 (¼ Xe [ Xi )
closed cylindrical cavity. Additionally, the membrane is damped
ðr o ; /o ; zo Þ point of observer
pj ðr; /; zÞ by a linear viscoelastic force defined by the relationship between
pressure in Xj for j ¼ e; i in R3 (¼ Xe [ Xi )
pj ðnÞ pressure in Xj as a discrete function of time stress and strain [9]. With respect to a cylindrical coordinate sys-
qa air density in Xe [ Xi tem ðq; /; zÞ, the transverse motion of the membrane is governed
ca speed of sound in air by:
Kettledrum  
@2g @g
v surface of the kettledrum r ¼ T r2
g þ m þ Pðq; /; L ; tÞ  Pðq; /; Lþ ; tÞ; ð1Þ
L height of the kettledrum @t2 @t
Mallet where r is the superficial density of the membrane, m is the vis-
uðx; y; zÞ position of the point of oscillation of the mallet
coelastic coefficient and P is the acoustic pressure field.
M total mass of the mallet
K coefficient of mallet stiffness Pðq; /; L ; tÞ represents the pressure field just below the drumhead
a angle of inclination of the mallet while Pðq; /; Lþ ; tÞ the pressure just above it.
Drumhead-mallet interaction The acoustic pressure field on R3 ðXe [ Xi Þ satisfies the wave
F Initial interaction force between mallet and drumhead equation
rs ¼ ðqs ; /s Þ striking point !
1 @2
r p ¼ 0; ð2Þ
ca @t 2
model was later improved and extended to a composite two mem-
brane case by Sankalp and Anurag [4]. However, none of those where it is assumed that the kettle walls act as a sound baffle. This
models included viscoelastic dependency. In comparison, Chaigne assumption requires that:
and Kergomard [5], Rhaouti and Chaigne [6] include the viscoelas-
tic damping coefficient, which accounts for average loses, but by
employing a numerical model instead of an analytical solution.
Previous kettledrum models report unreasonably high decay times
for the musically relevant modes Sankalp and Anurag [4] which is
mostly attributed to the vibrating membrane that continuously
dissipates the energy from it in the form of sound waves. Our
model differentiates from former approaches, in addition to includ-
ing a linear viscoelastic drumhead model, we also incorporate a
model sound synthesis technique allowing us to approximate the
acoustic pressure field directly.
To verify our kettledrum model, we compared the synthesized
sounds with the ones produced by a real instrument. More specif-
ically, we have recorded a set of kettledrum sounds in a controlled
environment, then we have thoroughly contrasted the frequency
spectrum produced by our model with the spectrum produced by
a 32” inch orchestral kettledrum. The similarity between both
sounds is remarkable, particularly in the preferred modes.
The remainder of the paper is structured as follows. In Section 2,
we formulate the problem and describe the methodology used to
obtain the mathematical model. In Section 3, we present the exper-
imental setup, discuss the numerical procedure, as well as the Fig. 1. Proposed kettledrum model. An infinite baffle surrounds the circular
spectral analysis and model parameters sensitivity analysis. drumhead. A cylindrical system is used where the origin is located at the
Finally, we present concluding remarks in Section 4. drumhead’s center. Table 1 lists the symbols used in this work.
E.U. Gallardo et al. / Applied Acoustics 164 (2020) 107284 3

dP dP where A0 is related to the initial force as A0 ¼ F=M; ðqs ; /s Þ is the


ða; /; z; tÞ ¼ ðq; /; L; tÞ ¼ 0; ð3Þ
dq dz striking point and qw is the impact radius. Expanding aðq; /Þ in
terms of the eigenmodes (gms ), equating to (10) and integrating
at Xi . The outside pressure (at Xe ) is assumed to be limited by an
with respect to gm0 s0 along the drumhead surface yields;
infinite baffle in the z ¼ L plane. This implies that:
K Z
X a Z 2p X
K X
S
dP ð0Þ
ai ðq; /ÞUkm0 s0 gm0 k0 ¼ bm0 s x2m0 s Ukm0 s Ukm0 s0 ; ð12Þ
ðq; /; L; tÞ ¼ 0 8q > a ð4Þ
dq k¼0 0 0 k¼0 s¼0

This last assumption has been shown to have moderate effects on where bm0 s are the coefficients associated to each eigenmode gm0 s .
the results [8], nonetheless, its incorporation greatly simplifies the The evaluation of (12) results in a N ¼ S  ðK þ 1Þ linear equations
analytical calculation. In general, Eqs. (1)–(4) form the coupled to solve for di ¼ ðb01 x201 ; . . . ; bKS x2ms Þ in the form:
boundary problem for the transverse displacement and pressure
[4]. ANN dN1 ¼ C N1 : ð13Þ
It is assumed that both transverse displacement and pressure Once the vector d is solved from the previous equation, both these
have harmonic time dependence i.e. eixt , thus the internal and values and the corresponding eigenvectors Ukm0 s are subsitituted
external pressure can be expressed by the Green functions Gin back to (5) and (6). (5) is numerically calculated for a considerably
and Gout defined in Appendix A as: large time interval. We further compute the spectral density of the
Z 2p Z a signal and compare the result with the spectrum obtained experi-
Pðq; /; L Þ ¼ gðq0 ; /0 ÞGin ðrjr0 Þq0 dq0 d/0 at Xi ð5Þ mentally for the corresponding stroke.
0 0

and 3. Experimental results and discussion


Z 2p Z a
Pðq; /; Lþ Þ ¼  gðq0 ; /0 ÞGout ðrjr0 Þq0 dq0 d/0 at Xe : ð6Þ Our results are divided in 5 subsections. First, we describe our
0 0
experimental setup and then we discuss the numerical procedure.
By taking a similar approach to Sankalp and Anurag [4], where gms is Next, we compare the calculated and measured frequencies of the
defined in terms of the basis functions provided by the normal first 16 modes using the Power Spectral Density (PSD) of the
modes of a membrane without air-loading: recorded sounds. Then, we analyze the effect of the viscous term
X
K in the calculated spectral density and compare both models with
ð0Þ
gms ðq; /Þ ¼ Ukms gmk ðq; /Þ; ð7Þ the experimental results. Finally, we investigate the effect of the
k¼0 microphone position (the point where acoustic pressure is evalu-
the following results: ated) on the calculated synthesized sound. Here, we emphasize
on how inaccuracies in the parameters of the model affect the pre-
0
  4x2ms qa a X
N
0 cision of the estimated sound.
x2ms  xð0Þ
mn ð1  imxms Þ Ums ¼
k
Unms xmn xmn0 ½C mnn0
r n0 ¼1

1
 iImnn0 ; ð8Þ
2
where the left and right subscripts in x indicate the number of dia-
metrical nodes and circular nodes respectively, xmn is the nth root of
the mth order Bessel function, C mnn0 and Imnn0 are defined as follows:
N00
X cotðcðymn00 ÞL=aÞ
C mnn0 ¼ ð9Þ
n00 ¼1 cðymn00 Þðx2mn  y2mn00 Þðx2mn0  y2mn00 Þð1  ym2 2 Þ
mn00

Z 1
kJ 2m ðkÞ
Imnn0 ¼ dk: ð10Þ
0 cðkÞðk  x2mn Þðk2  x2mn0 Þ
2

Fig. 2. Kettledrum and drumsticks used during the experimental evaluation of the
where Jm ðxÞ refers to the Bessel function of the mth order and xmn is proposed model.
defined as the solution of J m ðymn Þ ¼ 0. It should be noted that Eq. (8)
is identical to that obtained by [7] with the exception of the vis-
coelastic term. Eq. (8) can be solved as an iterative eigenvalue– Table 2
Parameter values used for calculations. Striking points and microphone positions are
eigenvector problem where xms is the eigenvalue and Ukms the
defined in cylindrical coordinate system.
eigenvector to be determined. At the end of each iteration, a relax-
ation parameter is applied to xms before it is substituted back. The Parameter Values used for each tonality
process is repeated until frequencies agree to at least two decimal A B[ F G
places.
a 0:4015 m – – –
r 0:262 kg/m2 – – –
2.2. Modal sound synthesis model qa 1:21 kg/m3 – – –
ca 344 m/s – – –
The initial acceleration provided by the impact of the mallet is m 0:6  10 6 m/s – – –
L 0:4015 m – – –
modeled by the cosine profile:
T 3600 N 4075 N 2290 N 2770 N
( h  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi i )
A0 S1 ð0:2576; 0; LÞ – – –
1 þ cos qp 2 q2 þ q2s  2qqs cosð/  /s Þ if jq  qs j 6 qw
ai ðq;/Þ ¼ 2 w S2 ð0:3226; 0; LÞ – – –
0; otherwise M1 ð0:115; p; L þ 0:14Þ – – –
ð11Þ M2 ð0:115; p; L þ 0:79Þ – – –
4 E.U. Gallardo et al. / Applied Acoustics 164 (2020) 107284

Fig. 3. Spectral density of the recorded sound of a A2-tuned timpani drumhead stroke at S2 measured at M2 . Predicted frequency values are outlined in doted lines. The
spectrum was normalized such that the highest amplitude is located at 0 dB.

3.1. Experimental procedure where n ¼ 1; . . . ; N. The periodogram is defined as a function of


the frequency f in Hertz (Hz) as
All of the measurements reported in this paper were made on a
2
copper timpani with an Adams Renaissance 32 inch drumhead. The 1 XN

kettledrum was placed at the center of a recording studio lined Sðf Þ ¼ p ðnÞei2pfn : ð14Þ
N n¼1 j
with sound baffles to reduce reverberation. Two striking positions
(S1 and S2 ) known by the player to provide a wide range of har-
monic tones were selected and marked. Two Shure SM57 instru- 3.2.1. Comparison between calculated and observed modal frequencies
ment microphones were positioned at different locations (M1 and Using the parametric values defined in Table II, we obtained
M2 ) on top of the timpani, although the recordings from only one the frequency values of the first 16 modes that coincided well
of them were used in our evaluation. Four different mallets were with the peaks identified in the real sound spectrum.1 Results
used to strike the timpani. The drum was tuned by a professional are shown in Figs. 3 and 4 for the A and B[ tuned timpani. Both
timpanist so that frequency of the g11 mode coincided with the fre- figures compare the frequency values with the normalized PSD
quencies of the notes A2 , G2, B[2 and F2. This was done by tightening of the sound recorded at M2 due to a strike in S2 . The computed
the drum head with the use of a pedal mechanism. The sound of values of the first 10 modes are shown in Table 3. The values
each drum strike was recorded with a sampling frequency of obtained with N,N0 ; N00 and K set to 12 for the B[2 tuned membrane
48 kHz with 24-bit resolution using an X32 Behringer Digital Mixer are also shown The PSD of the sound recorded at M1 was not
and Cubase 8 Pro, the recording hardware was located in a contigu- taken into account due to its higher distance with respect to the
ous separate room. All sound recordings obtained for the present striking point. Hence non-radiative dissipative mechanisms that
work are provided free of charge. The instrument and mallets used were not considered become dominant as the distance to the
are presented in Fig. 2. source increases.
We observed that the predicted frequencies generally coincide
3.2. Numerical procedure well with the peaks identified in both spectrums as reflected in
similarity between the real and computed values. The absolute
The parameter values for the timpani membrane and kettle, mean error  of the first N ¼ 10 modes was obtained using the fol-
along with those of the air used in our calculations are shown in lowing formula:
Table 2.
1X N
Modal frequencies xms and eigenvectors Ums were obtained for  ¼ f
true analytical
 fi : ð15Þ
the modes m ¼ 0; . . . 12; s ¼ 0; . . . ; 3. The values of T, m and L were N i¼1 i
chosen so that the calculated frequency of the mode g21 coincided
with the experimental value. For the sake of simplicity, the upper These values were found to be 1:25  0:76 Hz and 1:87  1:83 Hz
limits of the sums defined in (8) and (9) (N, N0 ; N00 and K) were for the A and B[ tuned drums respectively.
all set to 6, while S and K in (12) were set to 3 and 6 respectively. In the case of B[ , the highest error obtained was attributed to
To evaluate the importance of the number of terms, a second the g12 modal frequency, two solutions were identified while vary-
experiment was conducted were N, N0 ; N00 and K were set to 12. ing the penalty factor from 0 to 1. However, only one of them mod-
By following the already discussed procedure, the acoustic pres- erately coincided with the observed frequency value.
sure field was computed at each position (M1 and M2 ) and its PSD
was estimated using the Periodogram method [10]. More specifi- 1
Neither the striking position S nor the microphone position M were considered
cally, let pj ðnÞ be the pressure in Xj as a discrete function of time, yet.
E.U. Gallardo et al. / Applied Acoustics 164 (2020) 107284 5

Fig. 4. Spectral density of the recorded sound of a B[2 -tuned timpani drumhead stroke at S2 measured at M 2 . Predicted frequency values are outlined in doted lines. The
spectrum was normalized such that the highest amplitude is located at 0 dB.

Table 3
Comparison of real vs. synthetic (i.e. estimated) values of the first 10 modes for the A and B[ tuned drums. The Syn 2. column refers to the synthetic values obtained with more
terms in the sums.

A B[
Mode Real (Hz) Syn. (Hz) Real (Hz) Syn. (Hz) Syn. 2 (Hz)
g11 109.8 107.9 116.15 114.3 114.3
g21 165.8 165.4 175.7 175.6 175.6
g31 219.3 219.8 232.5 233.4 233.4
g12 236.4 235.0 251.8 245.6 245.6
g41 272.0 272.8 288.2 289.8 289.8
g51 323.8 325.0 343.0 345.3 345.3
g32 363.2 365.2 385.6 387.9 387.7
g61 375.3 376.7 397.3 400.4 400.3
g13 385.1 384.8 408.0 408.0 407.8
g42 421.5 424.1 451.0 450.6 450.5

The frequencies obtained for the 12  12 matrix system can be Table 4


seen to differ very little from the original 6  6 problem. An error of Computational times for each task.
less than 2 decimal places was reported for the first 6 modes. This Task Time (s) Time
suggests that the eigenvalue problem of a large computationally (%)
expensive matrix system will yield a very similar result to a small Solve the 6  6 eigenvalue-eigenvector 720 s (40 s per 91.1%
less expensive 6  6 problem. problem frequency)
Solve the 18  18 linear system 23 s 2.9%
Evaluate the acoustic pressure 47 s 6%
3.2.2. Computational cost analysis
In summary, the modal sound synthesis procedure applied here Total time 790 s 100%
involved 3 basic steps. The first step required solving a 6  6 eigen-
value - eigenvector problem iteratively to obtain each modal fre-
quency. This was done for all 18 modes. The number of iterations
was set to 10 for each mode. The second procedure, involved Although the first task required a relevant computational time of
obtaining the coefficients bms associated with each eigenmode 12 min, its results only depend on the physical properties of the
gms . This required assembling an 18  18 matrix system and solv- membrane (tension, density, etc.) and the air surrounding it.
ing it. The third and final procedure involved numerically evaluat- Hence, the drumhead - drumstick interaction is not involved. The
ing (6) for each eigenmode considering the parameters in Table 2 low times of the second and third task makes the model ideal for
and applying the discrete Fourier transform to calculate the acous- real-time applications. For instance, if the parameters of the drum
tic pressure. To synthesize the sounds, a sampling frequency of were to be known beforehand, only the second and third task
36,000 Hz was used. would need to be performed in real-time. In contrast, other numer-
Prototyping was performed using Python. To benchmark we ical methods (finite difference, finite element) involving timpani
used an Intel Core i7-6500 processor at 2.6 GHz with 8 GB of drum synthesis are time consuming and will require parallel pro-
RAM gave the computational times listed in Table 4 for each step. cessing to achieve comparable results in performance [11].
6 E.U. Gallardo et al. / Applied Acoustics 164 (2020) 107284

3.3. Spectral analysis and viscoelasticity effect M1 using the modal frequencies obtained in the previous section
for the A and B[ tuned drumheads. This is done with and with-
By following the modal sound synthesis procedure already out the viscoelastic term. The PSDs of the synthetic time
defined in Section 2.2, the acoustic pressure field is calculated at series obtained for the A and B[ tuned drumheads are shown in

A2 real
A2 synthesized
-40 A2 synthesized with viscoelastic term

-60
Power level (dB)

-80

-100

-30

-120
-40

-50
-140
-60

105 110 115


-160
0 100 200 300 400 500 600

Fig. 5. Spectral density of real sound and synthesized signal with and without viscoelasticity of an A2-tuned timpani. The dB scale employed is relative to a digitized signal at
full scale.

-20

B b2 real
B b2 synthesized
-40
B b2 synthesized with viscoelastic term

-60
Power level (dB)

-80

-100

-30
-120
-40

-140 -50

-60
110 115 120
-160
0 100 200 300 400 500 600

Fig. 6. Spectral density of real sound and synthesized signal with and without viscoelasticity of a B[2 -tuned timpani. The dB scale employed is relative to a digitized signal at
full scale.
E.U. Gallardo et al. / Applied Acoustics 164 (2020) 107284 7

Figs. 5 and 6. Since we are computing the squared value of the dig- where the Root Mean Square (RMS) envelope of the viscosynthetic
ital signal according to Eq. 14, in analogy with electrical signals we signal with and without the viscoelastic term is compared to the
refer to the vertical axis as the power spectrum, although in reality RMS envelope of the real sound. The n-th sample of RMS envelope
there is no physical power involved. In our case, the dB scale that pRMS ðnÞ is computed using a sliding window as follows:
we use is relative to a amplitude Amax Þ of a digitized signal at full
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
scale, more specifically 20log10 ðA=Amax Þ. We note that the spectral u
u1 X N
 2
envelope of the synthesized signal generally coincides well with pRMS ðnÞ ¼ t pj ðn þ kÞ ; ð16Þ
the real PSD. It is seen that the amplitudes of the g11 ; g21 ; g31 and N k¼1
g41 modes better match the observed amplitudes. This correspon-
dence is less accurate for higher frequencies. We believe that this is where N corresponds to the number of samples that fit in a 20 ms
due to reflective effects inside the recording room that were not data segment.
considered in the model. In terms of sound quality, no significant difference was noted in
The effect of including viscoelasticity was noticeable in two dif- the initial attack of the synthezised sounds. However, as seen in
ferent ways. First, the viscoelastic term was found to generally Fig. 7, the viscoelastic synthetic signal sounds closer to the real
reduce the amplitude of the observed PSD. As a noticeable feature, one during the fade-out phase. Sound quality evaluation should
the g12 mode appeared to coincide well with the observed fre- not be limited only to waveform or spectral similarity, a formal
quency value only in the synthesized PSD of the viscoelastic model subjective evaluation should be conducted to fully determine if
for both A and B[ . The effect of viscoelasticity was also noticeable in the viscoelasticity improves the perception of the kettledrum
the decay time of the sound. This effect is illustrated in Fig. 7, sound model.

0.7
Real
0.6 Synthetic with viscoelasticity
Synthetic

0.5
Amplitude

0.4

0.3

0.2

0.1

0
0 0.5 1 1.5 2 2.5 3 3.5 4
Time (s)

Fig. 7. Root Mean Square (RMS) envelope of the viscosynthetic signal with and without the viscoelastic term is compared to the RMS envelope of the real sound for the B[ -
tuned timpani. Envelopes were computed using a sliding window of 20 ms.

-30 -30 -30


A2 R A2 R A2 R
A2 S A2 S A2 S
-35 -35 -35
A 2 S5TC A 2 S5U A 2 S5C
A 2 S5TE A 2 S5D A 2 S5E
-40 -40 -40
Power level (dB)

-45 -45 -45

-50 -50 -50

-55 -55 -55

-60 -60 -60

-65 -65 -65


100 105 110 115 120 100 105 110 115 120 100 105 110 115 120

Fig. 8. PSD of real sound and synthesized signal without viscoelasticity of an A-tuned timpani. The meaning of the acronyms is the following for (a): A2R, real sound; A2S,
synthesized sound; A2S5TC, synthesized sound microphone position 5 cm towards the center; A2S5TC, synthesized sound microphone position 5 cm towards the edge. For
(b): A2S5U, microphone position 5 cm upwards; A2S5D, microphone position 5 cm downwards. For (c): A25SC, strike position 5 cm towards center; A25SC, strike position 5 cm
towards edge. The dB scale employed is relative to a digitized signal at full scale.
8 E.U. Gallardo et al. / Applied Acoustics 164 (2020) 107284

3.4. Sensitivity analysis of model parameters port during the recording of the kettledrums sounds. The authors
also wish to thank the anonymous reviewers for their valuable
To analyze the uniqueness of the solution, we slightly vary the comments and suggestions that helped us to improve the quality
striking and microphone positions used in the model. Fig. 8(a)–(c) of this manuscript.
show the PSD of the signal calculated at different microphone and
striking positions at a 100–120 Hz frequency range corresponding Appendix A. Expressions for Gin and Gout
to the g11 mode for the A-tuned membrane. We note that the cal-
culated amplitude of the g11 mode coincides well with the real The Green functions for the outer and inner regions defined in
value. We observe that the calculated amplitude is more sensitive Fig. 1 are given as:
to the striking position than the microphone position. For instance, p
2
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ið x Þ q2 þq02 2qq0 cosð//0 Þþðzz0 Þ2
in cases A2S5TC and A2S5D the amplitude increases while in the Gout ðq; /; zjq0 ; /0 ; z0 Þ ¼ ep ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ca
2
cases A2S5TE and A2S5U it decreases. This same effect is amplified 4p q2 þq02 2qq0 cosð//0 Þþðzz0 Þ2
p
2
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðA:1Þ
for the cases A25SC and A25SE respectively. The results obtained for ið x Þ q2 þq02 2qq0 cosð//0 Þþðzþz0 Þ2
ep ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
þ
ca
2
the A2S5U and A2S5D cases may be physically interpreted as sound 4p q2 þq02 2qq0 cosð//0 Þþðzþz0 Þ2
intensity is expected to decay as the distance from the source
increases. The high amplitude reached in the A2S5C case may be X
1
0 X
1
0
eimð// Þ m ½ymn ðq=aÞJ m ½ymn ðq =aÞ
Gin ðq; /; zjq0 ; /0 ; z0 Þ ¼ 4p 2p
 2J
a2 ð1m2 =y2mn ÞJ 2m ðymn Þ
accounted for by its proximity to the center where the preferred
m¼1 n¼1
modes (e.g g11 ; g21 ; g31 ) are more likely to contribute to the result-
ing spectrum, while the low amplitude in the A2S5E case may
 cos½cmnc z< sinð
cos½cmn ðLz> Þ
c LÞ ;
mn mn

reflect its tendency to excite higher modes. ðA:2Þ


where ymn is the solution of J0m ðymn Þ
¼ 0, z> ; z< denote the greater
4. Conclusions
and lesser of z and z0 , respectively, and
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Among orchestral drums, the kettledrum is generally consid- cmn ¼ x2 =c2a  y2mn =a2 for ymn < xa=ca ;
ered the most important due to its capacity to convey a clear sense pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðA:3Þ
cmn ¼ i y2mn =a2  x2 =c2a for ymn > xa=ca ;
of pitch. Hence it is relevant to adequately estimate the sound
spectrum. Our results show that the modal frequencies of an ca is the speed in which sounds travels above and below the
orchestral kettledrum can be accurately estimated using the air- drumhead.
loaded membrane formulation defined in Section 2.2. The modal
sound synthesis technique introduced here to estimate a synthetic References
spectrum that closely matches the real sound spectrum. The vis-
[1] Rossing TD. The physics of kettledrums. Sci Am 1982;247(5):172–9.
coelastic term was found to generally reduce the amplitude of
[2] Bilbao S. Numerical sound synthesis: finite difference schemes and simulation
the modal frequency peaks. Noticeably, the calculated frequency in musical acoustics. John Wiley & Sons; 2009.
values of the g12 mode coincided with the observed experimental [3] Christian RS, Davies RE, Tubis AB, Anderson CA. Effects of air loading on
tympani membrane vibrations. J Acoust Soc Am 1984;76:1336–45.
value only in the viscoelastic case. The sensitivity analysis done
[4] Sankalp T, Anurag G. Effects of air loading on the acoustics of an Indian musical
for the model parameters in Section 3.4 shows that the striking drum. J Acoust Soc Am 2017;4:2611–21.
position is more relevant for the spectrum synthesis than the [5] Chaigne A, Kergomard J. Acoustics of musical instruments. Springer; 2016.
microphone location. In contrast to other numerical approaches, [6] Rhaouti L, Chaigne A. Time-domain modeling and numerical simulation of a
kettledrum. J Acoust Soc Am 1999;105:43–75.
our model is not only realistic but also computationally efficient. [7] Davis RE. Mathematical modeling of the orchestral timpani. Purdue University;
The proposed approach can be extended to other types of orches- 1988. Ph.D. dissertation.
tral drums. We consider that the presented model could be [8] Christian D, Tubis. Effects of air loading on timpani membrane vibrations. J
Acoust Soc Am 1984;76(5):1336–45.
improved by involving the drumstick-drumhead interaction in [9] Wang Y. Generalized viscoelastic equation. Geophys J Int 2016;204
the model. The recorded kettledrum sounds and Python source (5):1216–21.
code of the model are freely available. [10] Stoica P, Moses R. Spectral analysis of signals. NJ: Pearson Prentice Hall Upper
Saddle River; 2005.
[11] Bilbao S, Webb C. Timpani drum synthesis in 3D on GPGPUS; 2012.
Acknowledgments

The authors wish to thank Miguel Ángel Cuevas and the Faculty
of Arts at the Autonomous University of Baja California for the sup-

You might also like