Impulse response of the Bayreuth Festspielhaus

Kai Huang∗
Experimentalphysik V, Universität Bayreuth, 95440 Bayreuth, Germany
(Dated: September 11, 2018)
The Bayreuth Festspielhaus is well known for its architecture because its design is heavily in-
fluenced by composer Richard Wagner. Due to the special acoustic design, the reverberation time
(i.e., time scale for the sound pressure level to decay 60 dB) is larger than usual opera houses.
Using hand-claps and smart phone recordings, I measured the impulse response of the Bayreuth
arXiv:1703.07080v1 [physics.pop-ph] 21 Mar 2017

Festspielhaus in the auditorium, on the stage, as well as in the orchestra pit. The measured rever-
beration time shows quantitative agreement with the literature values within a certain frequency
range, demonstrating the possibility of using this approach to monitor room acoustics.

INTRODUCTION

The Bayreuth Festspielhaus is unique for its special

acoustic design by Richard Wagner in the 19th century
[1]. It has also been well preserved since then for Wagner-
ian operas. The hidden orchestra pit, for instance, clears
the way for the audience to focus on the stage. Together
with the reflecting board that blocks the direct sound,
the pit acts as a low-pass filter and sets the sound from
the orchestra in perspective with the singers [2]. Due to
the special design, the reverberation time (RT), which is
one of the most important room acoustic quantities, of
the Bayreuth Festspielhaus is relatively large compared
with the other opera houses of the same age [1].
Impulse response is often used to characterize room
acoustics, because it probes all frequency components of
a room. Based on the measured response, further charac-
terizations on the strength, speech intelligibility, echoes,
reverberation, and other features are straightforward to
be quantified [3, 4]. Such a quantification facilitates room
acoustics design and planning, so that incommunicable
rooms such as the one that Sabine was facing more than FIG. 1. Architects’ drawings of the Bayreuth Festspielhaus
100 years ago [5] can be avoided. The source of an im- adapted from Ref. [1]. (a) A floor plan with the borders of
pulse can be a loudspeaker fed with a pulse signal, ex- the auditorium, the stage and the orchestra pit, the location
of the sound source (SS) marked. The recording devices are
ploding objects such as air-balloons, or hand-claps. Fol-
located in the region shaded in gray. The red line represents
lowing a standard protocol [6], we can determine RT and the stage gate, which was closed during the measurement.
other room acoustic quantities from the recorded signals (b) Corresponding side-view sketch with a close-view of the
[3, 4]. The RT and other room acoustic properties of the orchestra pit.
Bayreuth Festspielhaus have been measured systemati-
cally for more than half a century [1]. Shortly before the
recent renovation, Garai and colleagues have conducted a smart phone recordings (HCSP) is presented. The corre-
systematic characterization of the Festspielhaus in 2014 sponding room acoustic parameters (RT and center time)
[7]. are obtained with the ITA-Toolbox [10] and compared
With the recent development of information technol- with literature values, which were collected in the same
ogy, room acoustic characterizations are becoming more month as the present investigation (September, 2014) [7].
convenient using hand claps and the microphones from Based on the comparison, I discuss the advantages and
smart phones [8, 9]. An interesting follow-up question typical drawbacks of the HCSP and provide a checklist
is: How well can the acoustics of a room be character- for collecting sensible data from this approach.
ized using this approach in comparison to the conven-
tional measuring protocol? Here, the impulse response of
the Bayreuth Festspielhaus acquired by hand-claps and
MEASURING PROCEDURE

Figure 1 shows the sketches of the Festspielhaus

∗ kai.huang@uni-bayreuth.de adapted from the architects’ drawings. The measurement
 2

was conducted during a visit with about 50 visitors on

September 11, 2014. As marked in Fig. 1(a), the sound
of five hand-claps was generated at location SS, directly
in front of the orchestra pit. The sound was recorded
by the microphones of three smart phones with a sam-
pling rate of 44.1 kHz. The distances between the source
and recorders were 5 ∼ 10 meters, roughly double the
reverberation distance [4]. In addition, five hand-claps
with both SS and microphones in the orchestra pit (with
shorter distances between SS and the recorders due to
the limited space) and on the stage were also performed.
Subsequently, the recorded signal of each individual clap
was extracted manually for post processing.
The data analysis was performed with Matlab using
the ITA-Toolbox [10], which provides a standard routine
for room acoustic characterizations following ISO 3382
[6]. More specifically, the raw signals are filtered into dif-
ferent octave bands and subsequently the energy decay
curve (EDC) for each band is obtained through applying
the backward integration method on the squared enve-
lope of each filtered signal [3]. From least square fits of
the EDC in the semi-logarithmic plane, the reverberation
time is obtained. Depending on the different ranges of
data used in the fitting, EDT (early reverberation time),
T15, T20, T30 (i.e., using the energy decay from −5 dB
to −20, −25, or −35 dB), etc., are obtained. In the
present study, T15 and the center time obtained in the
frequency range from 250 Hz to 4000 Hz, covering five
octave bands, are presented. T20 and T30 are not used
because they cannot be obtained for all frequencies due
to the limited signal-to-noise ratio (SNR). FIG. 2. (a) Representative raw data captured by the three
recorders in the Auditorium. (b) Corresponding energy decay
curves for center frequency 1000 Hz. The solid lines in (b) are
linear fits of the data in the semi-logarithmic plane to obtain
RESULTS AND DISCUSSION
T15.

Figure 2(a) shows the raw signals captured by the three

microphones. Although all the smart phones use the energy level represents the influence of the direct sound.
same program (Smart Voice Recorder) for recording, the Because of the strong direct sound from recorder 1, the
sound pressure differs one from another, owing to the dif- magnitude of the corresponding EDC drop, which repre-
ferent properties of the microphones and analogue-digital sents the energy level of the direct sound, is much larger
(AD) converters embedded, as well as the different dis- than the other two. Based on definition, RT is obtained
tances to the SS. The following features can be learned through a fit of the exponential decay part of EDC. The
from a comparison of the raw signals: (i) The direct and agreement of the EDCs in this regime indicates that the
reflected sound can be clearly distinguished (separated exact locations and gain levels of the microphones play a
at ∼ 0.05 s) for all recorded signals. (ii) the sound pres- minor role in obtaining RT, because those configurations
sure of the direct sound may vary dramatically from one only lead to different factors to the EDC, not the decay
recording device to another. For instance, the maximum exponent. Because the sound energy generated by hand-
sound pressure from recorder 1 is more than one order claps is limited and the waiting time between subsequent
of magnitude larger than that from recorder 3. (iii) The claps is not sufficiently long, the conventional T30 can-
direct sound is not always the strongest peak in the im- not be obtained for all octave bands. Instead, T15 is
pulse response. The difference of the direct sound energy used here. This difference can lead to additional uncer-
will lead to the scattering of the center time, which will tainty in determining RT. Moreover, as indicated by the
be discussed at the end of this section. EDC of recorder 1, too large direct sound energy may
The cumulative energy decay curves obtained with the lead to a rapid drop of the energy level to below −5 dB
backward integration method [11] are shown in Fig. 2. and consequently a shorter reverberation time.
For all three recordings, EDCs have a similar behavior: As shown in Fig. 3(a), no clear distinguish between the
A stepwise initial drop followed by an exponential decay direct and reflected sound can be made for the data col-
(note the semi-logarithmic scale). The initial drop of the lected in the orchestra pit. This can be attributed to the
 3

3.6

(s)
3.4

15
3.2
Literature values

T
3.0 (unoccupied)

Reverberation time
2.8

2.6

2.4

2.2

Recorder 1
2.0
Recorder 2
1.8 Recorder 3

mean
1.6

250 500 1000 2000 4000

Frequency (Hz)

FIG. 3. (a) A sample recording obtained in the orchestra pit,

illustrating the lack of clear separation between direct and re-
flected sound. (b) Corresponding EDCs at 1 kHz for all three
recorders. Solid, dashed and short dashed lines correspond to
the linear fits of the data in the semi-logarithmic plane, using
the energy decay between 0.30 and 0.70 s.

relatively small volume and the complex geometry of the

pit [see the inset of Fig. 1(b)]. In addition, the sound
FIG. 4. (a) Reverberation time of the auditorium for differ-
reflecting board, which was designed to block the direct
ent octave bands. The error bars correspond to the standard
sound to the stage, effectively mixes the directed and re- deviation from five hand-claps. The asteroids are mean val-
flected sound together. Last but not least, the opening ues from the different recorders. The region shaded with gray
to the stage gives rise to additional influence from the lines corresponds to the literature values [7, 12]. (b) Corre-
stage and auditorium. Consequently, the EDCs shown sponding center time at different frequency.
in Fig. 3(b) exhibit a gradual decay instead of the rapid
drop. Thus, using the conventional ways of determin-
ing RT is inappropriate, because the exponential decay below −20dB within the recording time period because of
starts at an energy level lower than −5 dB. Thus, I fit the slow decay rate for low frequency sound. Therefore,
the tails of the EDCs and obtain RT from the slopes of the statistics is not sufficient for an accurate quantifica-
the fits. An average of the three EDCs yields an RT of tion. (ii) The strong influence of the direct sound may
1.38 ± 0.08 s for the octave band centered at 1000 Hz. lead to large fit error because the exponential decay may
The reverberation time obtained from different hand- start at an energy level smaller than the standard −5 dB.
claps and recorders is shown in Fig. 4(a). A comparison (iii) The frequency response of MEMS (microelectrome-
between the mean value obtained from the current mea- chanical systems) microphones, which are typically used
surement and the literature values [7, 12] demonstrates in smart phones, has a roll-off at low frequencies and a
that RT can be extracted from the simple HCSP ap- peak at ∼ 15kHz due to the design of the chamber geom-
proach, at least in a certain frequency range. However, etry [13]. Consequently, the reliability of data obtained
due to the lack of high repeatability, different hand-claps at low frequencies may suffer from the low SNR.
yield slightly different RTs (represented as errors). Thus, RT obtained at different locations is compared in Ta-
an average of the obtained RTs from different clapping ble I. Quantitative agreement between the data from the
events and recorders is necessary for a sensible charac- auditorium and from the stage is found for frequency
terization. The error from the exponential fits can be ≥ 500 Hz, suggesting that the characterization of RT is
ignored as it is typically much smaller than the uncer- weakly dependent on where the hand-claps and recorders
tainty from different clapping events. are located. Due to the same reason described above,
The deviation from the literature values at low frequen- there is also an overestimation of RT at 250 Hz for the
cies (≤ 250 Hz) can be attributed to the following three RT obtained on the stage.
reasons: (i) The sound energy does not always decay to Figure 4(b) shows the center time obtained from
 4

measurement devices for monitoring room acoustics, pro-

TABLE I. Mean values of the reverberation time obtained in
vided that the following precautions are properly taken
the auditorium (A) and on the stage (S), unoccupied.
care of:
250 500 1000 2000 4000 (Hz)
A 3.1 2.7 2.5 2.3 1.8 (s) - The sound generated by hand-claps should be as
S 3.5 2.6 2.5 2.1 1.8 (s) loud as possible to have sufficient initial sound en-
ergy, otherwise the accuracy for individual mea-
surements will suffer from the low SNR.
the same measurement. It is defined as the first - Sufficient waiting time (at least 1.5ts with ts the
moment
R∞ of the
R∞ squared impulse response ts = expected RT) between individual claps is necessary
2 2
0
[p(t) ]tdt/ 0
[p(t) ]dt [3]. As ts characterizes the bal- for extracting the response at low frequencies, be-
ance of direct and reflected sound, it is expected to vary cause the low frequency components decay slower
with the locations of the recorders and sound sources, than the high frequency ones.
as well as with the source signals generated. Therefore,
the data scattering is strong in comparison to the RT re- - Multiple measurements (hand-claps, recording de-
sults. As described above, the influence of direct sound vices and locations) are needed for better statistics.
on recorder 1 is the largest among the three, therefore
the corresponding ts obtained is the smallest for most In addition to RT, the center time obtained with the
frequencies. An average over the results from different same recordings shows strong scattering because of its
frequencies yields 0.125 ± 0.023 s, which also compares dependence on the positions of the sound source and
fairly well with a previous measurement [7]. This value recording devices. Therefore, further analysis on the spa-
suggests that the speech intelligibility is ≥ 80% [3] if the tial distribution of ts is needed for a better comparison
speaker is standing close to the proscenium. with the other measurements.
The possibility of quantifying room acoustics proper-
ties with easily accessible devices helps, for instance, to
monitor an opera house in occupation on a more regular
CONCLUSIONS AND OUTLOOK basis and to provide instantaneous feedback on building
open-air theaters with the help of real-time analysis tools
To summarize, room acoustic characterizations are [8, 9].
conducted for the Bayreuth Festspielhaus using hand-
claps as sound source and smart phones as recording
devices. The reverberation time obtained with the ITA- ACKNOWLEDGMENTS
Toolbox agrees quantitatively with another measurement
[7] taken place in the same month within a certain fre- I acknowledge the organizers of the 19th Dynamics
quency range. Possible reasons for the data scattering Days Europe for arranging the visit to the Bayreuth
among different recorders and hand-claps, as well as de- Festspielhaus. I would also like to thank Erik Werner,
viations from the literature values are discussed. This in- Thomas Müller, and Stephan Messlinger for their kind
vestigation demonstrates the possibility of using amateur help in audio recording.

[1] L. Beranek, Concert Halls and opera houses (Springer, [8] P. Seetharaman and S. P. Tarzia, in
New York, NY, 2010). Audio Engineering Society Convention 132 (2012).
[2] J. D. Polack and M. Retbi, in 8th International Confer- [9] A. Rosenkranz, R. Burgmayer, D. Ackermann,
ence on Auditorium Acoustics (2011) pp. 198–206. M. Hädrich, and S. Weinzierl, in the 43th German
[3] H. Kuttruff, Room Acoustics, 4th ed. (CRC Press, Lon- Annual Conference on Acoustics (DAGA) (2017).
don, England ; New York, NY, 2000). [10] P. Dietrich, M. Guski, J. Klein, M. Müller-Trapet,
[4] G. Müller and M. Möser, Handbook of Engineering M. Pollow, R. Scharrer, and M. Vorländer, in 40th Italian
Acoustics (Springer, Berlin, Heidelberg, 2012). (AIA) Annual Conference on Acoustics and the 39th Ger-
[5] W. C. Sabine, Collected papers on acoustics (Harvard man Annual Conference on Acoustics (DAGA) (2013).
University Press, 1922). [11] M. R. Schroeder, J. Acoust. Soc. Am. 37, 409 (1965).
[6] “ISO3382-1, acoustics measurement of room acoustic pa- [12] M. K., in Das Richard Wagner Festspielhaus Bayreuth,
rameters. part 1: Performance spaces,” (2009). edited by M. Kiesel (Nettpress, Düsseldorf, 2007) pp.
[7] M. Garai, D. D’Orazio, K. Ito, 174–197.
S. De Cesaris, and F. Morandi, in [13] T. A. Michaelis and C. Kling, in the 43th German Annual
22th International Conference of Sound and Vibration Conference on Acoustics (DAGA) (2017).
(Florence, Italy, 2015).