Audio Engineering Society

Convention Paper 10267

Presented at the 147th Convention
2019 October 16–19, New York, USA

This Convention paper was selected based on a submitted abstract and 750-word precis that have been peer reviewed by at
least two qualified anonymous reviewers. The complete manuscript was not peer reviewed. This convention paper has been
reproduced from the author's advance manuscript without editing, corrections, or consideration by the Review Board. The
AES takes no responsibility for the contents. This paper is available in the AES E-Library, http://www.aes.org/e-lib. All rights
reserved. Reproduction of this paper, or any portion thereof, is not permitted without direct permission from the Journal of
the Audio Engineering Society.

The effects of spectators on the speech intelligibility performance

of sound systems in stadia and other large venues
Peter Mapp1, Ross Hammond2
1
Peter Mapp Associates, Colchester UK
2
Department of Engineering, University of Derby, UK
Correspondence
shouldbe addressedto Author(petermapp@petermapp
.com)

ABSTRACT
Stadiums and similar venues in the UK and throughout most of Europe are subject to strict safety standards and
regulations, including the performance of their Public Address systems. The usual requirement is for the PA
system to achieve a potential speech intelligibility performance of 0.50 STI, though some authorities and
organisations require a higher value than this. However, a problem exists with measuring the performance of the
system, as this can only be carried out in the empty stadium. The paper shows that with occupancy, the acoustic
conditions change significantly, as the spectators introduce significant sound absorption and also increase the
background noise level. The effect this can have on the intelligibility performance of the sound system are
examined and discussed. The relationship between the unoccupied starting conditions and audience absorption
and distribution are also investigated.

conditions, many venues often do not meet the

1 Introduction prescribed STI (Speech Transmission Index) values;
The intelligibility of the sound system in stadiums, though subjective evaluations carried out during a
arenas or large venues, where spectators or an game or event suggest that the intelligibility is
audience may be present in large numbers, is of satisfactory when the venue is occupied.
paramount importance. Furthermore the installed One of the objectives of the paper is to see if
sound system normally forms part of the venue’s life computer modelling a typical stadium, with various
safety measures. However, realistically the PA degrees of occupancy, could help illustrate the
system can only be objectively measured when the effects of occupancy on the potential intelligibility
stadium is unoccupied, without spectators being of the sound system. In order to give the project
present. Under unoccupied conditions, the direct significance, a real stadium having a capacity
reverberation time can often be surprisingly high, of approximately 30,000 spectators was taken as a
even in semi enclosed stands. For example, test case. The performance of the sound system was
measurements made by the author in several soccer measured in the empty stadium and the reverberation
stadiums indicate that even in stands with time characteristics of each stand were measured. A
completely open fronts and open or only partially detailed acoustic, computer model of the stadium
closed in sides, the reverberation time can typically was then built and the in-situ measurements used to
be in the region of 3-4 seconds. Under such
Mapp & Hammond Effects of spectators on intelligibility

calibrate the model. The effect of occupancy was the STI technique takes account of signal to noise,
then modelled for a number of scenarios. the method and algorithms also incorporate a speech
A survey of the literature showed that a wide range level dependency function, such that the potential
of values for the absorption of spectators and intelligibility of speech at low and high sound levels
audiences have been published, though definitive is weighted (i.e. reduced) in order to mirror the
data does not appear to exist for stadia and similar characteristics of human ear. In stadiums it is the
venues. Most absorption data for audiences relates to reduction in STI at high sound levels that is of
concert halls and similar auditoria, where the importance. This reduction factor begins to operate
acoustic conditions are very different, as is the at approximately 83-85 dBA and is nonlinear in
disposition of the audience. Some limited absorption operation.
data does exist for standing audiences and other,
non-concert hall scenarios [1]. The reported
absorption coefficients were found to vary by
approximately 100%. This difference was therefore
expected to have a significant effect on the occupied
reverberation time within a stand and hence on the
STI and sound system performance. The extremes of
the absorption coefficient ranges were tested in the
various occupancy / noise scenarios and are reported
in the paper. Three different stadium scenarios were
modelled, each with a range of acoustic conditions.
A range of noise & PA system announcement levels
has also been investigated.
Figure 1 – Effect of SPL on STI

2 Stadium & Absorption Modelling The weighting (reduction) factor is not universal but
Three different stadium models were employed. The is highly dependent on the other modulation
first comprised one large, single stand modelled in reduction losses produced by reverberation and
isolation. The second model was that of a complete ambient noise / signal to noise ratio. To put this into
stadium, where predicted data could be could be perspective, figure 2 shows some effects that
compared with actual performance measurements. different noise and signal to noise ratios and hence
The third model was based on the second but the speech levels can have on STI.
reverberation time was increased (i.e. the residual
absorption of the structure and seating were reduced)
resulting in lower unoccupied STI values. The
objective of this was to see if the expected changes
(improvements) in STI were linear.

It should be noted that STI [2] takes account of both

the effects of reverberation and noise on speech
intelligibility, as well as a number of other factors
which need not be addressed here.
It is common practice to design PA and VA (Voice
Alarm) systems with a target signal to noise ratio of
10 dBA. Furthermore, many emergency sound
system standards such as BS5839-8, IEC 50849 and
Figure 2 – Effect of SPL & SNR on STI
NFPA72 stipulate a signal to noise ratio of ≥ 10
dBA. This requirement however, is no guarantee that
a system will achieve the STI requirement. Whilst

AES 147th Convention, New York, USA, 2019 October 16–19

Page 2 of 8
Mapp & Hammond Effects of spectators on intelligibility

From figure 2, it can be seen that a signal to noise

ratio of 12 dBA for example results in STI values of
0.74, 0.70 and 0.65 respectively for the background
noise conditions of 60, 70 & 80 dBA respectively.
As the figure shows, at the higher SPLs, increasing
the signal to noise ratio, begins to decrease the STI.
When dealing with announcement levels with high
SPLs, such as are needed in stadia etc, this masking
effect and STI reduction can have a very significant
impact, making high intelligibility difficult to
achieve.
The foregoing discussion needs to be borne in mind
when considering the modelling results presented in
the next section as it will be shown to have a very
significant impact on the results.

As noted in the introduction, a very wide range of

absorption coefficients was found to exist within the
literature. It was therefore decided to run two
scenarios representing the minimum and maximum
reported coefficients. These are shown in figure 3.

Figure 4 – Stadium 1 – Stand model

Figure 3 – Absorption coefficients used in the computer

models

Stadium 1 Modelling

This modelling exercise was carried out on a large, 3

tier stand, about to be constructed, to show that the
0.50 STI criterion could be met. The stand was a
‘standalone’ structure and modelled in isolation to
the rest of the stadium. Figure 4 shows an image of
the model employed.

Figure 5 – Modelling STI Result Plots

AES 147th Convention, New York, USA, 2019 October 16–19

Page 3 of 8
Mapp & Hammond Effects of spectators on intelligibility

The ambient noise level measured during a game

was 93LAEQ and so this value and associated
spectrum was used for the STI calculations. The PA
system was able to produce a broadcast speech level
of 109 dBA – thereby providing a signal to noise
ratio of up to 16 dBA. Figure 5 shows a number of
the results plots which are summarized in the table
below.

Ambient Noise level (LAEQ) 93 dB

PA system level 109 dBA
STI with occupancy, without 0.72 Figure 6 – 3D image of test stadium
noise or masking
STI with noise (16 dBA snr) 0.66
STI with noise & masking 0.54
STI with masking – no noise 0.60
STI – without audience abs 0.46
Table 1 – Stadium 1, summary of results

From table 1, it can be seen that with the absorption

of the spectators taken into account, the potential
mean STI is 0.72. When the background noise is
included (i.e. 16 dBA snr) the STI reduces to 0.66.
However, when the high SPL penalty (masking) is
introduced (as it must be) the STI reduces to 0.54.
This still means that the system passes and meets the Figure 7 – Measured RT characteristics
0.50 STI criterion and is shown to potentially
provide adequate intelligibility. However the
significant effect that the masking function has is
clearly demonstrated reducing the STI from 0.66 to
0.54. Equally, calculating the effect of the masking
alone (i.e. without noise, infinite snr) the effect is to
reduce the STI from 0.72 to 0.60. In other words,
operating the PA system at 109 dBA reduces the
potential STI by 0.12.

Stadium 2 Modelling

Stadium 2 is fairly typical of many UK soccer Figure 8 - Computer Model

stadiums. It has a capacity of 30,000 spectators and
one of the stands is standalone. A 3D image of the The computer model was calibrated against the
stadium is shown in figure 6. Considering the size reverberation time and STI measurements made in
and open frontage of the stands, they are remarkably the real stadium under unoccupied conditions.
reverberant, typically exhibiting a mid-frequency Figure 9 compares the RT predictions and
reverberation time of around 3.5 seconds, as shown measurements. The installed sound system achieves
in figure 7. Figure 8 shows three views taken from an average STI value of 0.53 in the unoccupied
the computer model. stadium.

AES 147th Convention, New York, USA, 2019 October 16–19

Page 4 of 8
Mapp & Hammond Effects of spectators on intelligibility

dBA had no significant effect on these results. In

other words, in this case, 5 dBA snr was adequate.
These results are summarised in figure 11 below.

Figure 9 – RT measured & predicted values

The effect of the audience absorption, for the two

sets of data employed, can be seen in figure 10. The
differences in the RT were smaller than might be
expected from the large differences in the Figure 11 – Effect of Absorption coefficient on STI
coefficients. For example at 1 kHz, the average RT
reduced from 3.1 seconds for the empty condition to The effect that different audience distributions had
2.3 seconds for the lower absorption tested and to on the overall STI was also investigated. As there
1.8 seconds for the higher absorption characteristic. was little difference between the two sets of
absorption data at the emergency announcement
level for the occupied stadium, the predictions were
carried out using the lower, more conservative case.
Three different distributions were investigated, Full,
½ full – (all at front) and ½ full (alternate rows
occupied). Figure 12 summarises the results.

Figure 10 – RT unoccupied and & predicted occupied

The above results show that spectators can indeed

make a significant difference to the RT in a stand.
The corresponding, mean STI values were 0.59 for
the low absorption and 0.63 for the high absorption,
assuming a low ambient noise level and a PA level
of 70 dBA. Whilst statistically the difference is
Figure 12 – Effect of spectator distribution
significant, it is only just above the JND.
However, when the PA level was increased to 95 As figure 12 shows, increasing the PA
dBA, signal masking reduced this difference to 0.02 announcement level from 70 dBA to 95 dBA in the
STI, with the mean STI values reducing to 0.52 and empty stadium decreases the mean STI value from
0.54 respectively. Introducing spectator noise of 90
0.53 (pass) to 0.48 (fail). However, whilst some

AES 147th Convention, New York, USA, 2019 October 16–19

Page 5 of 8
Mapp & Hammond Effects of spectators on intelligibility

authorities and consultants test a stadium like this, it Stadium 3 Modelling

is a completely meaningless exercise, as it is not
possible to have the crowd noise without the Stadium 3 was based on stadium 2 but the
additional absorption that the spectators inherently reverberation time was increased to produce a lower,
bring with them. As figure 11 shows, the audience initial, mean STI value. The unoccupied, mid-
absorption effectively restores the STI back to its frequency RT was increased to around 4.5 seconds
original, unoccupied value (0.53 pass) at the higher as shown in figure 14. Again the two extremes for
announcement level of 95 dBA. The distribution of audience absorption data were investigated reducing
the spectators was found to make a small difference, the unoccupied RT to approximately 3.6 and 3.1
with unsurprisingly, the more evenly distributed seconds respectively.
absorption giving rise to the better result, though the
difference is within the normal calculation
uncertainty. The effect of signal masking and signal
to noise ratio was further investigated over the range
from -5 to +15 dBA, with two further conditions of
+95 dBA (i.e. infinite snr) also being considered.
The lower occupancy absorption data was employed.
Firstly the condition without signal masking was
investigated. With no background noise, the mean
STI in the empty condition (predicted and measured)
is 0.54 which is predicted to increase to 0.59 when
the stadium is full. With 15 & 10 dBA signal to
noise ratios this value of 0.59 does not change but
reduces fractionally with 5 dBA snr and even at -5
Figure 14 – RT Characteristics – Stadium 3
dB snr, the STI still achieved 0.56 - as shown by the
orange bars in figure 13. However, when speech
An updated crowd noise spectrum was also
level masking is considered (as it has to be) the STI
employed, being similar to that used in model 1.
drops significantly but still achieved the 0.50 target
Figure 15 compares the two spectra. As can be seen,
STI value even with a zero dBA signal to noise ratio.
the newer spectrum (upper curve) has more high
The noise spectrum employed was based on crowd
frequency energy – particularly at 4kHz. The latter
noise measurements made by the primary author and
spectrum is based on a 45 minute LEQ measurement
reported in [3].
of the crowd noise from a soccer game, whilst the
former spectrum is based on a much shorter sample
related to the crowd noise produced immediately
after a goal had been scored [3]. The crowd noise
level was initially set at 90 dBA for the STI
calculations although a range of levels was also
investigated.

The increase in RT caused the mean unoccupied STI

to reduce to 0.45 for the empty stadium (i.e. no noise
or low noise). The effect of audience absorption on
the STI was then investigated. Using the higher
sound absorption coefficient for the spectators (full
occupation) increased the STI to 0.57 and 0.54 for
the lower absorption. So again, the choice of
absorption coefficient does not appear to be highly
Figure 13 – Effect of noise and masking, PA = 95dBA
critical, but the difference of 0.03 STI can be of

AES 147th Convention, New York, USA, 2019 October 16–19

Page 6 of 8
Mapp & Hammond Effects of spectators on intelligibility

significance near to target boundaries. The effect of

absorption was further investigated with the
inclusion of crowd noise.

Figure 16 – Effect of Occupancy & Noise on STI

To further establish the dominant effect of the STI

Figure 15 – Comparison of crowd noise data employed
signal masking function, a series of tests were run at
Operating the PA system at 100 dBA, but in the various background noise levels ranging from 80 to
unoccupied stadium, reduced the STI from 0.45 to 100 dBA. The PA broadcast level was maintained at
0.42, due to signal masking. In the fully occupied 100 dBA, though in the unoccupied condition, a
state, with an ambient noise level of 90 dBA and 100 level of 75 dBA was also employed. The results are
dBA PA speech level (i.e. a signal to noise ratio of summarised in figure 17.
10 dBA), the resultant STI values for the two sets of
coefficients were 0.51 & 0.49. Perceptually, there is
no difference between these, the difference being
less than the JND. However, contractually one is a
pass and the other could be construed as a fail !
Interestingly, without the masking function being
applied, the 10 dBA snr was sufficient to ensure that
the STI did not change from the ‘noiseless’
condition i.e. the STI values remained at 0.57 and
0.54. This again demonstrates the dominating effect
of signal masking at high SPLs. Figure 16
summarises these results.

It is arguable as to whether the stadium, when half Figure 17 – Effect of various background noise levels
full, is as noisy as when it is fully occupied. A
further set of computations were therefore carried In the empty stadium, the PA system with
out with a crowd noise level of 87 dBA for the half reverberation only as the intelligibility degrading
occupancy condition – using the lower absorption factor (i.e. high snr), produced a mean STI value of
coefficient (B). Under these conditions, again with a 0.45 with a 75 dBA speech signal. Increasing the PA
PA level of 100 dBA, the predicted STI values are level to 100 dBA reduced the STI to 0.42. With a
0.46 and 0.49 for the half and fully occupied stadium background noise level of 80 dBA and a PA
(or 0.51 and 0.54 without masking). Again, the broadcast level of 100 dBA, in the fully occupied
signal to noise ratio of 10 dBA is sufficient to stadium, the mean STI was 0.49 (0.54 without
overcome the noise level with the reduction in STI masking). This value was maintained up to and
being purely dependent on masking.

AES 147th Convention, New York, USA, 2019 October 16–19

Page 7 of 8
Mapp & Hammond Effects of spectators on intelligibility

including 95 dBA. With 100 dBA of background 7 The results indicate that a signal to noise ratio
noise, the STI reduced to 0.48. of 10 dBA is not always required in order to
achieve an STI value of ≥ 0.50.
3 Comments on Results & Conclusions
The paper has established and highlighted a number 8 The results indicate that it may be prudent to
of important findings highly relevant to the design minimise the snr at high sound pressure levels
and performance testing of stadia and other large in order to minimise the potential effects of
reverberant spaces. The following comments and speech masking.
conclusions are made:
9 A well-adjusted Ambient Noise Sensing
1 The sound absorption introduced by the system (ANS) could be beneficial in keeping
presence of spectators and audiences has a very PA announcement SPL requirements to a
significant effect on the acoustic conditions in minimum and so optimise performance, by
the stadiums investigated and must not be potentially minimising speech masking.
ignored when checking the intelligibility
performance of the PA system. 10 The STI results obtained were calculated using
the standard STI speech spectrum as per the
2 A wide range of values for the sound 2011 version of the STI standard (IEC 60286-
absorption attributable to a seated audience or 16). However, in practice, the actual spectrum
spectators was found within the literature. of a particular announcer’s voice may be
However, in practice, even quite large different – particularly at high frequencies
differences in the absorption coefficients only (4kHz & 8 kHz) where greater speech energy is
had a small effect on the resultant STI and so often present [4].
the exact absorption values applied do not
seem to be hyper-critical. 11 Work is continuing to see if a general ‘rule of
thumb’ can be determined and applied to
3 In general, the absorption increased the STI by unoccupied measurement data to predict the
approximately 0.1 – which is in keeping with likely STI values under occupied conditions.
the previous experience and measurements
made by the author.
References
4 It has been shown that STI measurements made [1] IEC 60268-16 (2011) – Objective rating of
in an empty stadium do not reflect the likely speech intelligibility by speech
result for the occupied case. transmission index.

5 Testing an unoccupied stadium with high [2] Hammond, R et al, On the accuracy of
speech SPLs has been shown to be an audience implementations in acoustic models.
erroneous practice, except to ascertain that the Audio Society Convention 145, New York,
PA system is capable of delivering the SPLs 2018. Paper 1007
required for emergency announcements.
[3] Mapp, P. “The effects of spectators,
6 The speech level masking factor within the STI audiences and buildings on sound system
assessment method has been shown to be a performance”. Audio Engineering Society
dominating factor in determining the potential Convention 98, Paris, 1995, paper 3964.
intelligibility of high SPL announcements.
[4] Mapp, P. Some Practical Aspects of STI
Measurement & Prediction. Audio Society
Convention 134, Rome, 2013, paper 8864.

AES 147th Convention, New York, USA, 2019 October 16–19

Page 8 of 8