Professional Documents
Culture Documents
In Situ Calibration of The Sound Strength Parameter G - Bryan Katz
In Situ Calibration of The Sound Strength Parameter G - Bryan Katz
In Situ Calibration of The Sound Strength Parameter G - Bryan Katz
Brian F. G. Katz
Citation: The Journal of the Acoustical Society of America 138, EL167 (2015); doi: 10.1121/1.4928292
View online: https://doi.org/10.1121/1.4928292
View Table of Contents: https://asa.scitation.org/toc/jas/138/2
Published by the Acoustical Society of America
The sound strength parameter G and its importance in evaluating and planning the acoustics of halls for music
The Journal of the Acoustical Society of America 129, 3020 (2011); https://doi.org/10.1121/1.3573983
International Round Robin on Room Acoustical Impulse Response Analysis Software 2004
Acoustics Research Letters Online 5, 158 (2004); https://doi.org/10.1121/1.1758239
Brian F. G. Katz: JASA Express Letters [http://dx.doi.org/10.1121/1.4928292] Published Online 10 August 2015
1. Introduction
For almost two decades, the sound strength parameter G has been employed as a mea-
sure of the amplification of a hall. This criterion, and its derivatives like Glate, has been
found to be perceptually highly relevant. However, this parameter is often viewed as
too difficult to measure by many practitioners. As such, the widespread usage of G has
not manifested despite studies supporting its use (e.g., Bradley1 and Beranek2). One of
the major reasons for such reticence can be attributed to the difficulty in carrying
out the measurement according to the ISO measurement standard.3 This difficulty lies
in the need to have a reference measurement of the free-field sound pressure level of
the measurement chain at a distance of 10 m.
Two approaches are described in the standard for obtaining the reference mea-
surement. The first employs a direct or extrapolated free-field mean sound pressure
level (SPL), such as acquired in an anechoic room, requiring rotation of the source to
account for source directivity variations. Alternatively, the mean SPL can be deduced
via a measure of the radiated sound power level (PWL) using a spatial averaged
diffuse-field measurement in a reverberation chamber or a spatially averaged intensity
measurement around the source. Both of these approaches are cumbersome in their
requirements and impractical if not impossible for most acousticians. As the calibra-
tion considers the entire measurement chain, the requirement extends beyond just the
loudspeaker and includes the amplifier and signal conditioning as configured at the
time of the measurement of both source and receiver. The result of this is that signal
gains must be fixed in the equipment and considered stable between the on-site and
calibration measurement sessions. This can be difficult in reality, e.g., due to drifts in
potentiometers or, more practically, to the need to adjust gains if various rooms or
configurations are measured in the same session. A final non-standard method, termed
the field method, has been employed by some acousticians,4 consisting in SPL measure-
ments made at a short distance from the source on stage, which are then extrapolated
to 10 m according to the free-field method.
2. Previous comparisons
In an early work, Beranek5 compared data reported from different acousticians, sepa-
rating them into two categories: reverberation chamber method (also referred to as the
Japanese method) and Western data sources. A difference between data sets concluded
the Japanese data to be approximately 1.2 dB greater. Each set was then adjusted by
60.6 dB to compensate for this difference in comparisons.
Tabulation of Gmid, taking the average of G for 500 and 1000 Hz octave
bands, as reported for all halls where data are provided from more than one source
(excluding Tanglewood Music Shed) results in data for 10 halls with both a Western
and Japanese data set. The mean of individual differences between these Western and
J. Acoust. Soc. Am. 138 (2), August 2015 C 2015 Acoustical Society of America EL167
V
Brian F. G. Katz: JASA Express Letters [http://dx.doi.org/10.1121/1.4928292] Published Online 10 August 2015
Japanese data is 1.10 dB (minimum: 0.10 dB, maximum: 2.03 dB), comparable to the
stated 1.2 dB in the reference. The mean magnitude difference among Western data is
0.45 dB (minimum: 0.02 dB, maximum: 0.77 dB).
A closer inspection of the data presented in the associated references5,6 is car-
ried out here and presents somewhat different results. It is noted that the calibration
method is not reported for the various Western datasets. After contacting the principal
sources of data in the references, it was determined that many Western data sets
actually employed the reverberation method.7
According to this revised tabulation (see Table 1), four halls had data using
both the free-field and reverberation methods, nine halls had multiple datasets using
the reverberation method, while only one hall offered multiple datasets for the free-field
method. Repeating the data analysis with data classified according to confirmed cali-
bration method shows a mean difference in Gmid between free-field and reverberation
methods of 0.71 dB (minimum: 0.39 dB, maximum: 1.03 dB) with a mean magni-
tude difference between reverberation data repetitions of 0.91 dB (minimum: 0.07 dB,
maximum: 2.03 dB), and between the two free-field data repetitions of 0.02 dB. It seems
clear that the variations between practitioners is greater than, or at least comparable
to, variations attributable to choice of calibration method when considering the rever-
beration method.
As such, it could be reasonable to assume that factors other than calibration
method are involved in the observed differences, such as measurement protocol and
errors. The reverberation method may be more prone to variation due to the additional
stage of calibration of the reverberant room, which requires accurate determination of
the reverberation time and volume of the room. Variations in reverberation time calcu-
lations alone have been shown to be on the order 5% among different algorithms for
Table 1. Tabulated values of Gmid for repeated measures according to calibration method. Numerical references
indicate literature source as (1) Beranek (Ref. 5) or (2) Beranek (Ref. 6) followed by the associated room ID in
the corresponding reference.
Gmid
Hall Source
Free-field Reverberation
EL168 J. Acoust. Soc. Am. 138 (2), August 2015 Brian F. G. Katz
Brian F. G. Katz: JASA Express Letters [http://dx.doi.org/10.1121/1.4928292] Published Online 10 August 2015
8
the same impulse response. Any errors in these calculations will propagate to errors in
the G calibration factor.
In addition, as the nature of G changes with distance, a spatial average will be
heavily affected by the positions chosen, and this could be an important factor to con-
sider. If calibration method was the only factor, such within method variations would
not be present, only the bias shift between methods. The observed differences in Gmid
for the same hall by different acousticians of up to 2 dB should raise concerns about
the repeatability of such measurements and the standard protocol. Similar concerns
have been raised about other measurement standard procedures.8
Several recent studies have made direct comparisons between calibration meth-
ods. Beranek6 presented a comparison between reverberation and field methods with
the field method being 1.4 dB greater for Gmid. This difference can be easily attributed
to the stage floor reflected energy contribution and the distance error due to the size of
the source. For a dodecahedron of diameter 30 cm, a source height of 1.5 m, and a
measurement distance of 1 m as reported, a back of the envelope calculation shows a
gain of 1.4 dB due to the 15 cm distance error (one-half the source diameter, see Hak
et al.9) and an additional gain of 0.3 dB due to the stage reflection with a 3.16 m path
length, assuming uncorrelated dB summation.10 This error can be minimized by apply-
ing time windowing of the measured room impulse response (RIR) to isolate the direct
sound.4,9
A more thorough study by Hak et al.9 has shown that differences on the order
of 1.2 dB were observable between various calibration methods of Gmid for two halls.
The free-field method results were 0.5 dB lower than the reverberation method. The
field results were 1.25 dB lower than reverberation method when using noise stimuli
while being just 0.2 dB higher for impulse response (IR) data via the sweep excitation
method. It is noted that the IR method allows for direct determination of source-
receiver distance. While the origin of these discrepancies was not discussed, the
observed variations are less than those indicated by Beranek. Concerning the field
method, in contrast to Beranek6 who used only two measurement positions on oppo-
site sides of the source, Hak et al.9 and Pires12 have presented results employing a
rotating turntable to account for source directivity averaging, resulting in variations of
less than 1 dB for Gmid. This procedure, however, requires additional time and costly
or cumbersome equipment.
The foundations of the free-field method rely on the extrapolation of the cali-
bration data from the single selected measured distance to the 10 m reference distance
according to the basic spherical spreading law, while accounting for source directivity
through rotation of the source. As an alternative, this study proposes a simple method
to calibrate G exploiting the actual RIR measurements throughout the hall, at various
known distances, and determining the calibration value necessary to fit a collection of
points to spherical spreading decay theory. Source directivity is considered through the
inherent spatial averaging across the various measurement directions. In contrast to
the noise-stimuli field method, time windowing is employed to isolate the direct sound
component in the measured RIR.
The advantages of this method include the absence of any additional equip-
ment relative to that necessary for other acoustical parameters, the ability to adjust
gain levels of the equipment on-site, and the added advantage of simplifying the proto-
col to allow for multiple sources and multiple receivers if available for large measure-
ment installations.
Sections 3–5 describe the proposed method for a test case, providing a com-
parison to results obtained in the Stavanger Concert Hall, Norway, using the free-field
anechoic room calibration method. Additional comparisons are made to measurements
made in the same hall with an alternate uncalibrated source.
J. Acoust. Soc. Am. 138 (2), August 2015 Brian F. G. Katz EL169
Brian F. G. Katz: JASA Express Letters [http://dx.doi.org/10.1121/1.4928292] Published Online 10 August 2015
determined from the time-of-flight relative to the stimulus using the start time of the
RIR. Deconvolution and post-processing steps were carried out in MATLAB.
The case study was the Fartein Valen hall in Stavanger, Norway. This 1500
seat concert hall by acoustician Eckhard Kahle, Kahle Acoustics, opened in September
2012 and is notable for its suspended balconies, movable canopy, and movable ceiling.
Measurements were carried out over a period of 2 days. The input gains on the RME
Octamic and Fireface were calibrated using a test tone with variations for each channel
between start and finish of the sessions 0.16 dB. Orchestra risers, seating, and stands
were in place on stage for 33 musicians. Src-Rec positions are indicated in Fig. 1(a),
being at heights of 1.5 and 1.2 m, respectively. Receiver position E was the engineer
position behind the mixing console. Some measurements were repeated as part of a
secondary study. Not all Src-Rec combinations were measured due to time constraints.
A total of 30 measurements were carried out, for the five microphones, resulting in 150
RIRs.
The reference calibration of the measurement chain was achieved in the
anechoic chamber at IRCAM, Paris. Impulse response measurements were carried out
with a Src-Rec distance of 1.86 m with source rotations in 5 steps. While this distance
is less than the 3 m stated in the standard,3 this was the measurement configuration pos-
sible in this anechoic room. Variations due to directivity [see Fig. 1(b)] were <1 dB for
frequency bands under 4 kHz as expected following the study by Leishman et al.,13 due
to the small size of the speaker. As such, it can be assumed that there should be mini-
mal effect of source orientation during the measurements. The mean sound pressure
level was calculated in octave bands and extrapolated to its equivalent level at 10 m via
the spherical spreading factor 20 log10(1.86/10).
As an alternative to previously cited calibration methods, an in situ calibration
method is proposed. This method is based on a comparison of the curve-fit of the
direct sound from a collection of existing Src-Rec measurements to the theoretical free-
field decay to derive the calibration factor. Specifically, the proposed in situ method
extracts the direct sound level from the measured data, which is then fit to free-field ex-
ponential decay theory, normalized to 10 m. To estimate the direct sound level, we
define here G(very early), or Gvearly, to represent G calculated on the first 5 ms of the
RIR, relative to the start of the RIR. If theory holds, the direct sound level Gvearly
should follow the free-field decay curve. A window of 5 ms is considered short enough
to ignore most reflections while inspection of the employed Src’s anechoic IR shows
the level at 5 ms was at least 40 dB below the peak value. For lower frequencies, and
Fig. 1. (Color online) (a) Measurement position in the Stavanger hall. (b) Octave-band directivity pattern of
dodecahedral speaker, measured at 5 steps. (c) Gvearly calibrated using the anechoic chamber method. (d)
Gvearly calibrated using the on-site method, based on data points indicated by (䊊), with indicated unused data
points lacking line-of-sight () and overly present early reflections (þ).
EL170 J. Acoust. Soc. Am. 138 (2), August 2015 Brian F. G. Katz
Brian F. G. Katz: JASA Express Letters [http://dx.doi.org/10.1121/1.4928292] Published Online 10 August 2015
4
larger sources, the window length may need to be extended. In applying the in situ
method, it is necessary to select Src-Rec configurations where the 5 ms assumption of
no substantial reflections is valid, where 5 ms equates to a path difference between
direct and reflected sound of 1.7 m. This eliminates positions too close to reflecting
surfaces and positions where grazing reflections over seating or balcony fronts arrive
within the time window. In addition, only positions with line-of-sight to the source are
considered.
Fig. 2. (Color online) (a) Gmid calibrated using the on-site method. (b) Glate,mid calibrated using the on-site
method. Small dodecahedron (), large dodecahedron (䊊).
J. Acoust. Soc. Am. 138 (2), August 2015 Brian F. G. Katz EL171
Brian F. G. Katz: JASA Express Letters [http://dx.doi.org/10.1121/1.4928292] Published Online 10 August 2015
EL172 J. Acoust. Soc. Am. 138 (2), August 2015 Brian F. G. Katz
Brian F. G. Katz: JASA Express Letters [http://dx.doi.org/10.1121/1.4928292] Published Online 10 August 2015
Additional thanks are offered to Magnus Ognedal and Tønnes Ognedal of Sinus AS for
their assistance in the measurements. Finally, thanks to the staff at Stavanger Konserthus
who were both accommodating and helpful with all our requests.
References and links
1
J. S. Bradley, “Using ISO 3382 measures, and their extensions, to evaluate acoustical conditions in con-
cert halls,” Acoust. Sci. Technol. 26, 170–178 (2005).
2
L. Beranek, “The sound strength parameter G and its importance in evaluating and planning the acous-
tics of halls for music,” J. Acoust. Soc. Am. 129, 3020–3026 (2011).
3
ISO 3382-1:2009, “Acoustics—Measurement of room acoustic parameters. Part 1: Performance spaces”
(International Organization for Standardization, Geneva, Switzerland, 2009).
4
J. J. Dammerud, “Stage acoustics for symphony orchestras in concert halls,” Ph.D. thesis, University of
Bath, 2009, https://stageac.wordpress.com/phd/.
5
L. Beranek, Concert and Opera Halls: How They Sound (Acoustical Society of America, New York,
1996).
6
L. Beranek, Concert and Opera Halls: Music, Acoustics, and Architecture, 2nd ed. (Springer, New York,
2004).
7
It was confirmed that data from Bradley was acquired using the reverberation method. Data from Gade
were acquired using the reverberation method with a correction applied using a reference field
measurement and anechoic measured directivity data to allow for on-site level adjustments. Data from
Hyde were acquired using the free-field method. Calibration methods for Matsuzawa and M€ uller-BBM
datasets could not be confirmed, and the sets were no longer considered. Data for the Salle Pleyel by
CSTB differed by more than 3 dB from other data and were removed as an outlier.
8
B. F. G. Katz, “International round robin on room acoustical impulse response analysis software 2004,”
Appl. Res. Lett. Online 5, 158–164 (2004).
9
C. Hak, R. Wenmaekers, J. Hak, L. Van Luxemburg, and A. Gade, “Sound strength calibration meth-
ods,” in International Congress on Acoustics, Sydney, Australia (2010), pp. 1–6.
10
It should be reiterated that the field and free field methods are not the same. Such confusion is further
perpetuated by lack of apparent clarity in studies discussing a so-called Western calibration method
(Refs. 5, 6, and 11).
11
L. Beranek and N. Nishihara, “Mean-free-paths in concert and chamber music halls and the correct
method for calibrating dodecahedral sound sources,” J. Acoust. Soc. Am. 135, 223–230 (2014).
12
G. Pires, “Determination of the sound power of an omni-directional sound source on behalf of a mea-
surement of the sound strength in a room,” Master’s thesis, Eindhoven University of Technology, 2004,
http://sts.bwk.tue.nl/bps/publications/fp655.pdf.
13
T. W. Leishman, S. Rollins, and H. M. Smith, “An experimental evaluation of regular polyhedron loud-
speakers as omnidirectional sources of sound” J. Acoust. Soc. Am. 120, 1411–1422 (2006).
14
It is noted that there was a 2 wk delay in making the calibration measurements, the equipment was
transported by airplane, and the equipment employed had analog gain knobs, which where taped into
position at the start of the measurement session, and remained taped until after the calibration
measurement.
15
It should be noted that certain Src-Rec combinations were only measured with one of the two sources,
due to time constraints (e.g., S3 was not measured with the smaller source).
J. Acoust. Soc. Am. 138 (2), August 2015 Brian F. G. Katz EL173