Content uploaded by Andor Fürjes
Author content
All content in this area was uploaded by Andor Fürjes on Dec 18, 2020
Content may be subject to copyright.
EXPLORING AN ALTERNATIVE SCHEME OF ROOM
ACOUSTICS SPECIFICATION
Andor T. Fürjes1,2
1 aQrate Acoustics Ltd., H8083 Csákvár, Hungary
furjes.andor@aqrate.hu
1 Animative Ltd., H1043 Budapest Dugonics street 11., Hungary
andor.f@animative.eu
NOTE: Revised version of the preprint uploaded to Forum Acusticum 2020 Conference.
ABSTRACT
To define a specification scheme for room acoustics is a
basic task that usually ends up by relying mainly on mean
reverberation time tolerances. Many papers investigate
how reverberation time correlates (or not) to other
measures. One shall also consider robustness and ease of
measurement methods of selected parameters when
creating regulations.
While aiming for a practical specification scheme, the
author found that instead of (late) reverberation times the
early decay time parameter should gain more emphasis in
general cases: it correlates well with speech definition
metrics and it is more robust to measure or model.
In addition, measurement data suggests that instead of
arithmetically averaging results at adjacent frequency
bands (mean reverberation time), evaluation of a pre-
filtered (e.g. A-weighted) wide band signal would yield a
more practical and robust single figure.
The paper presents both measurement and calculated
results.
1. INTRODUCTION
Room acoustics standards and legislations have a long
history several countries and there are numerous papers
that sum up or compare their set of requirements or just
evaluate their applications.
As of writing this paper, Hungary has no legislation on
room acoustics parameters directly. In everyday practice,
engineers of room acoustics have to refer to foreign
national standards and sometimes have to convince clients
in order to avoid obvious mistakes. Theoretically if there
is no explicit instruction of the client, room acoustics
treatment can be omitted without the legislative force.
The situation is not tragic, however. In most cases clients
or project engineers are usually aware of the importance of
room acoustics mostly due to experience with technical
specifications of foreign developers or former failures.
In 2016 a private initiative started discussion of a national
standard and since then a continuous debate is in process.
It would have been easy to just refer to a selected existing
foreign standard, but as one said in the film English
Patient: ‘You’re Hungarian. You always disagree.’
This paper presents some of the doubts and results of
investigations to pursue a meaningful room acoustics
specification scheme.
Please note, that the paper looks at general situations only,
rather than special or artistic cases (i.e. concert halls, opera
houses, theatres).
2. INITIAL THOUGHTS
Basically, any general room acoustics specification shall
set simply the minimum requirements in order to avoid
fundamental mistakes. At the same time clear instructions
shall be given to make clear what is better and what is not
acceptable.
Some would say, that instead of minimum requirements,
optimum requirements shall be set. However, despite the
numerous papers on topics of room acoustics there seems
to be no global agreement on what is exactly ‘optimal’.
Also, setting requirements to a so-called optimum suggests
that anything else is simply bad.
There might be trends towards using a unified letter
denoted classification scheme, but this should be treated
carefully. First of all, for any engineer it is awkward using
letters instead of numbers. Secondly, preferred value
ranges and set of parameters still seem to change over time
as understanding of acoustic impression evolves. Finally,
where a trained ear under laboratory conditions might hear
differences, we call JND (just noticeable difference), a
client would appreciate a more reasonable metric to make
decisions, like JMD (just meaningful difference, [1]).
3. REVERB OR DECAY
Most room acoustic schemes are based on or list at first
place some sort of reverberation time. Surely, if conditions
of the Sabine equations hold, and there are no irregularities
in a perfectly exponential decay, then a single
reverberation time says almost all one needs to know.
While there are papers supporting reverberation time as a
fundamental parameter (e.g. [2] showing good correlations
for concert halls), a lot of papers raise doubts and report
low correlation of reverberation times to other important
parameters (e.g. gain, definition, speech intelligibility,
etc., see e.g. [7]) in more general use.
Causes of deviations are usually identified as deviations
from the ideal exponential decay for any reason. Either the
room is too damped, sizes are disproportionate, absorption
is unevenly distributed or surface diffusion is too low.
However, there might be another obvious reason:
reverberation time by definition tells only the length of the
decay but not its quality. How can one tell the quality of a
book by only measuring its thickness?
Since direct measurement of reverberation time is hardly
feasible, measurement standards (e.g. [9]) suggest to
estimate reverberation time by best fitting lines between
pairs of points of the energy decay curve (see Figure 1.),
which can be calculated from the impulse response h(τ):
𝐸𝐷𝐶(𝑡)= 10 ∙ 𝑙𝑜𝑔 ∫()
∫()
(1)
Figure 1. Schematic drawings of early decay time (blue)
compared to (late) reverberation time (red) on the EDC.
Shaded areas show range of possible EDCs with the same
decay properties.
Energetically speaking this means, that reverberation time
estimates characterize only the last and lesser portion of
the impulse response (the -5 dB…-35 dB is about 32% of
the energy of the impulse response). By comparison, early
decay time (or EDT10) is the length of the first 10 dB drop
in the decay, which characterizes the first 90% energetic
portion of the impulse response.
It is known, that impression of room acoustic quality is
based by the very first part of the impulse response. If we
accept this, why do we expect that late reverberation times
would always correlate well with parameters that
characterize the early and larger portion of the impulse
response? In case of EDT10 compared to late measures:
what do we expect to tell from the rest of the impulse
response, when 90% of the energy is already delivered?
Choosing EDT10 as a qualitative measure seems
reasonable, because it is relatively easy to evaluate and
even less prone to in situ background noise than T10, T20 or
T30 (see e.g. [3]).
4. CORRELATIONS
To check usability and correlations of EDT10, both
measurements and results of computer models were
analyzed.
4.1 Measurement results (see also [10])
4.1.1 Reverberation time vs. early decay time
Measurement results of a broad range of situations
(600…800 measurements, V = 50…20,000 m3) is shown
in scatter plots of Figure 2. In the figure T20 is T5-25 (see
[9]) and “m2” denotes arithmetic average of results in
500 Hz and 1 kHz octave bands. For better resolution, both
horizontal and vertical axes are shown with log2 scale.
Please note, that most measurements were taken using
omnidirectional loudspeakers, but some measurements
results were collected using directional loudspeakers and
even turned in different directions, which caused even
larger fluctuations (see ranges of EDT10 at T20,m2 0.8 s and
6.4 s).
Figure 3 shows the same measurements but instead of
averaging results at 500 Hz and 1 kHz octave bands, A-
weighted wide band impulse responses were evaluated.
Here the correlation is clearer and more stable.
Overall statistical results are summarized in Table 1.
These observations suggest that evaluation of pre-filtered
wide band impulse responses might be more suitable than
averaging octave band results, also because evaluation is
faster and more robust against background noise levels.
Even if EDT10/T20 can highly vary from measurement to
measurement, taking ratios of arithmetic averages for
whole rooms give more stable results. According to results
in Table 1, EDT10,A is expected to be lower than T20,A and
more than 85% of cases between 80…100% of T20,A.
Table 1. Overall statistic figures of EDT10/T20 ratios.
time
(sec)
EDC
(dB)
0
-5
-10
-15
-25
t
0
t
5
t
10
t
15
t
25
EDT
10
/6
T
15
/6
T
25
/3
time
(sec)
h
2
(lin.)
early decay includes
more interaction of
source-room-receiver
late reverberation
supresses effects of
source-room-receiver
interaction
direct sound
reflections from
nearest boundaries
EDT
10,m2
/ T
20,m2
EDT
10,A
/ T
20,A
min 0,31 0,29
max 2,28 1,55
average 0,95 0,82
std. dev. 0,28 0,17
min 0,56 0,51
max 1,91 0,99
average 1,01 0,87
std. dev. 0,26 0,09
for each room
for each
measurement
Figure 2. Comparison of reverberation time and early
decay time measurement results of a wide range of rooms
(○ each measurement point, ● average of rooms).
Figure 3. Same results as in Figure 2, but with A-filtered
impulse response evaluation.
4.1.2 STI vs. reverberation time and early decay time
Figure 4 and 6 show scatter plots of T20,m2 and EDT10,m2 vs.
STI (EN 60268-16, indirect calculation, no masking,
standard spectra), where only the horizontal axis is shown
in logarithmic scale.
Figure 5 and 7 show the same but with A-weighted mean
reverberation time (T20,A) and early decay time (EDT10,A).
As one can see, EDT10,A correlates to STI much better,
namely regression model of
𝑆𝑇𝐼 ≈ −0.38 ∙ 𝑙𝑜𝑔𝐸𝐷𝑇,+ 0.59 (2)
yields R2 = 0.93.
Figure 4. Comparison of mean reverberation times
(T20,m2) and standard speech transmission index (STI).
Estimate is given in Eq. (3).
Figure 5. Comparison of A-weighted reverberation times
(T20,A) and standard speech transmission index (STI).
In addition, both graphs seem to provide a general ‘worst
case estimate’ for STI, true in at least 95% of all measured
cases:
𝑆𝑇𝐼 > −0.45 ∙ 𝑙𝑜𝑔𝑇,+ 0.57 (3)
𝑆𝑇𝐼 > −0.34 ∙ 𝑙𝑜𝑔𝐸𝐷𝑇,+ 0.54 (4)
These observations suggest that a room response can
degrade intelligibility down to a quite certain point, and
that intelligibility can be increased e.g. if the speaker has
higher directionality or the speaker is closer to the listener.
Especially in Figure 4 it is clear that for a high and almost
constant reverberation time, STI can vary significantly
depending on these conditions (directional characteristics
and proximity). On the other hand, these conditions are
already imprinted in early decay times. Also, correct pre-
filtering seems to support evaluating early decay time
y = 0,9582x + 0,05
R² = 0,9458
0,1
0,2
0,4
0,8
1,6
3,2
6,4
0,3 0,6 1,3 2,6 5,1
early decay time, EDT
10,m2
(sec)
reverberation time, T
20,m2
(sec)
y = 0,7875x + 0,0721
R² = 0,9738
0,1
0,2
0,4
0,8
1,6
3,2
6,4
0,3 0,6 1,3 2,6 5,1
early decay time, EDT
10,A
(sec)
reverberation time, T
20,A
(sec)
church with
Ø30m dome
regular rooms including
classrooms, offices,
meeting rooms, lecture
halls and theatres
0,20
0,30
0,40
0,50
0,60
0,70
0,80
0,90
1,00
0,3 0,6 1,2 2,4 4,8 9,6
speech transmission index (STI standard, -)
reverberation time (T
20,m2
, sec)
school
office
theatre
church
cultural center
event hall
music school rehearsal, untreated
music school rehearsal room, treated
STI worst estimate
0,20
0,30
0,40
0,50
0,60
0,70
0,80
0,90
1,00
0,3 0,6 1,2 2,4 4,8 9,6
speech transmission index (STI standard, -)
reverberation time (T
20,A
, sec)
school
office
theatre
church
cultural center
event hall
music school rehearsal, untreated
music school rehearsal room, treated
more meaningfully. Finally, reverberation time by itself
seems unable to become a reliable estimator of
intelligibility (or clarity).
Figure 6. Comparison of mean early decay times
(EDT10,m2) and standard speech transmission index (STI).
Figure 7. Comparison of A-weighted early decay times
(EDT10,A) and standard speech transmission index (STI).
Estimates are in Eq (2) and (4).
4.1.3 Single number ratings
While a single number rating is comfy for simple
specifications, it is not clear, why certain types of mean
reverberation times are believed to be better or worse for
the purpose.
In addition to the mean “m2” it is possible to add results
calculated in 2 kHz (Tm3 or Tmf) or even 125 Hz (Tm4)
octave bands to get a mean value, but according to
measured results, there seems to be no real benefit of using
more octave bands.
Figure 8 compares different single number ratings of
reverberation time and it is obvious, that only TA has a very
slightly differentiated behavior.
Figure 8. Comparison of single number ratings of
reverberation times T500Hz, Tm2, Tm3, Tm4 and TA.
4.1.4 Spectral tolerances
One might argue that including more frequency bands in
the mean might provide a better view on spectral balance
of reverberation time. In the view of the author however,
this statement is misleading, because an averaged
spectrum cannot substitute a properly chosen spectral
tolerance requirement.
Experience does not support that inclusion of additional
frequency bands is able to express evenness of spectral
balance, so a spectral tolerance shall be applied
independently from the definition of the single number
rating.
In general, there are two main reasons to specify some kind
of spectral tolerance:
a) to avoid excessive high/low ratios
b) to avoid excessive single band deviation.
Figure 9 shows relative deviations from T20,m2 mean results
for each measurement considering all measurements.
Figure 10 shows the same but for the case of early decay
time. Natural tendencies of reverberation and decay times
in untreated rooms seem to fit within an acceptable
tolerance. High (1kHz, 2kHz, 4kHz) and low (125 Hz,
250 Hz, 500 Hz) mean reverberation time ratios showed an
average of 0.87 with a deviation of 0.23.
Usage of spectral tolerances is validly advised for musical
applications or if high SPL is considered in the room below
250 Hz bands.
However, in general cases when considering only normal
speech signal (see e.g. [4]) and typical (LAeq≥30 dBA)
background noise spectra, the perceived signal to noise
ratio can be much smaller than 30 dB at 125 Hz and below
0,20
0,30
0,40
0,50
0,60
0,70
0,80
0,90
0,1 0,2 0,4 0,8 1,6 3,2 6,4
speech transmission index (STI standard, -)
early decay time (EDT
10,m2
, sec)
school
office
theatre
church
cultural center
event hall
music school rehearsal, untreated
music school rehearsal room, treated
0,20
0,30
0,40
0,50
0,60
0,70
0,80
0,90
0,1 0,2 0,4 0,8 1,6 3,2 6,4
speech transmission index (STI standard, -)
early decay time (EDT
10,A
, sec)
school
office
theatre
church
cultural center
event hall
music school rehearsal, untreated
music school rehearsal room, treated
STI estimate
STI worst case
y = 1,11x - 0,07
R² = 1,00
0,1
0,2
0,4
0,8
1,6
3,2
6,4
0,1 0,2 0,4 0,8 1,6 3,2 6,4
T
500Hz
(sec)
T
m2
(sec)
y = 0,91x + 0,06
R² = 1,00
0,1
0,2
0,4
0,8
1,6
3,2
6,4
0,1 0,2 0,4 0,8 1,6 3,2 6,4
T
m3
(sec)
T
m2
(sec)
y = 0,87x + 0,09
R² = 1,00
0,1
0,2
0,4
0,8
1,6
3,2
6,4
0,1 0,2 0,4 0,8 1,6 3,2 6,4
T
m4
(sec)
T
m2
(sec)
y = 0,78x + 0,13
R² = 0,98
0,1
0,2
0,4
0,8
1,6
3,2
6,4
0,1 0,2 0,4 0,8 1,6 3,2 6,4
T
A
(sec)
T
m2
(sec)
(see also [5]), making the use of late reverberation metrics
and <250 Hz frequency tolerances somewhat meaningless.
On the other hand, EDT10 seems still more meaningful at
low frequencies, because it expresses level change rates
that are surely perceptible in the cases of normal speech
dynamics under ordinary conditions.
Figure 9. Deviation of reverberation times at each
frequency band relative to Tm2 mean values. Percentages
denote statistical percentiles.
Figure 10. Deviation of early decay times at each
frequency band relative to EDT10,m2 mean values.
Percentages denote statistical percentiles.
4.1.5 Room conditions
Usually room acoustic specifications are meant either for
furnished/unfurnished or occupied/unoccupied situations.
This is a rather simplified scheme, because room acoustics
is significantly different also depending on whether a room
is inhabited or uninhabited.
In certain applications (e.g. kindergarten rooms,
elementary schools) operational conditions can
significantly differ from what a room acoustic engineer
can validly consider or measure before the rooms is used.
The above can explain, why unoccupied kindergarten
rooms could result Tm2<0.6 s and unoccupied elementary
school classrooms could result Tm2<0.7 s reverberation
times without any intentional room acoustic treatment.
Since vocal comfort can degrade by adding too much
damping (see [6]), consideration of expected additional
absorptions is advised in such cases.
In the view of the author, specifications for partial or fully
occupied situations shall be avoided due to the uncertain
nature of absorption occupants can present and the lack of
controlled environment during verification. Besides,
specifying the unoccupied condition is in most cases
practically the worst-case scenario specification.
Table 2. Differentiation of conditions of a room to
measure.
Due to the above considerations, comparable room
acoustic specifications shall be formed for furnished,
unoccupied, uninhabited situation, without any furniture
that can be removed under normal operational conditions
(e.g. an event hall can be used without mobile chairs, but
a class room without chairs is not functional).
4.2 Modelling results
Numerous papers (e.g. [7]) demonstrate how Sabine,
Eyring or other statistical reverberation time formula
cannot reliably correlate other monaural acoustic
parameters like G, clarity, definition or STI, if conditions
(e.g. room proportions, distribution of absorption) are not
ideal.
To test a wide range of scenarios, a series of modelling
calculations were carried out on 2 types of geometries (see
Figure 11). Both geometries have 1000 m3 volume, 200 m2
floor area and 5 m height. Boundaries are divided to
5×10 m (50 m2) patches. In case of Geometry #B patches
12 and 13 are the same. Source is at (6.0;6.0;1.5) m
coordinates. Omni receivers of 1×1 m raster are at 1.1 m
height in two parts of the room, each 8×8 m (1 m away
from side walls).
Initially all boundaries are reflective (smooth concrete)
and only 2 patches (so 100 m2 in total) are set to absorptive
(4” glass fiber). Different combinations of absorptive
patches (see Table 3) are calculated at 5%, 10%, 20% and
40% minimum diffusity settings (128 model runs total).
0,0
0,5
1,0
1,5
2,0
2,5
3,0
63 125 250 500 1000 2000 4000 8000
T
20
/T
20,m2
(-)
frequency band (Hz, octave)
2% 98% average 5%
10% 50% 90% 95%
0,0
0,5
1,0
1,5
2,0
2,5
3,0
63 125 250 500 1000 2000 4000 8000
EDT
10
/EDT
10,m2
(-)
frequency band (Hz, octave)
2% 98% average 5%
10% 50% 90% 95%
operational
building
handover,
commissioning
suggested
reference for
specifications
furnished yes no yes
occupied yes no no
inhabited yes no no
conditions to consider
factors to
consider
Finally, mean (m2) and A-weighted results of EDT10, T20,
G, C50 , C80 parameters are compared with each other and
STI.
Modelling calculations were carried out using EASE Aura
Module (4.4.6) and PETRA1. Both handle non-diffuse
(specular) and diffuse reflections and overall diffuse
reflection ratios can be set for each calculation.
Figure 11. Tested geometries with boundaries.
Table 3. Overview of calculated model variants. (X = 4”
thick glass fiber i.e. patch is absorptive, while - means
reflective (smooth concrete).
1 PETRA is an experimental acoustic modelling
environment, here combining only phased beam tracing
and radiosity methods.
Due to the high number of results, only some examples are
shown in Figure 12 (test geometry #A) and Figure 13 (test
geometry #B). Bars show average of mean values of
different models with 5%, 10%, 20% and 40% diffuse
settings.
Table 4 shows relation of different parameter results
expressed by coefficient of determination (R2) of linear
regression.
Reverberation times calculated by the Sabine formula is
1.48/1.47 s for the #A/#B geometries and A…Q cases.
Reverberation times according to the Norris-Eyring
formula are 1.36/1.36 s for the same.
Results confirmed the following observations:
- T20,mf or T20,A has poor or weak correlation to G or C50
and only modest correlation to STI;
- EDT10,mf or EDT10,A has poor correlation to G, modest
correlation to C50 and firm correlation to STI,
correlation is better for EDT10,A;
- spatial average minus standard deviation of STI seems
to correlate to both T20 and EDT10;
- C50 is highly correlated to STI;
- G is not correlated to decay length or clarity and its
spatial average depends mainly on absorption quantity,
but barely depends on its position or room shape;
- settings of diffusity affect T20 more than EDT10; and
arithmetic mean values more than A-weighted values.
Results also show, that reverberation or decay lengths are
lowest, STI and clarity are highest, while G is lowest if
absorption is arranged according D, I and K versions. The
opposite is true for arrangement M. Other arrangements
seem to perform broadly the same.
Generally speaking and not surprisingly, the more evenly
absorption is distributed, the more efficiently it reduces
reverberation and its consequences.
5. DISCUSSION
It is known, that early decay time (EDT10) is subjectively
more important than reverberation time (see e.g. [9]) and
that maybe reverberation times shall be taken only as
quantitative parameter.
This paper presented experience from wide range of
measurements and a systematic series of modelled
scenarios and concluded, that EDT10 is indeed a better
candidate to represent fundamental acoustic qualities of
rooms of general use:
- firm correlation to clarity and intelligibility;
geom. version 01 02 03 04 05 06 07 08 09 10 11 12+13
X - - - - - - - - - - - -
A X X - - - - - - - - - -
B X - X - - - - - - - - -
C X - - X - - - - - - - -
D X - - - X - - - - - - -
E X - - - - X - - - - - -
F X - - - - - X - - - - -
G X - - - - - - X - - - -
H - X X - - - - - - - - -
I - X - - X - - - - - - -
J - X - - - - X - - - - -
K - - - - X - - X - - - -
L - - - - - - X - X - - -
M - - - - X X - - - - - -
N X - - - - - - - - - - X
O - X - - - - - - - - - X
P - - - - X - - - - - - X
Q - - - - - X - - - - - X
#A, #B
boundary patch
#B
variant
- measurement is less sensitive to background noise and
does not necessarily require impulse response
measurements (easier in situ evaluation possible).
Figure 12. Modelling results summarized and compared
(test geometry #A).
There are problems to solve and concerns to clear, though:
- there is no simple formula to predict EDT;
- EDT measurement and evaluation practice: correct
direct sound detection and higher spatial variance can
cause ambiguity.
Figure 13. Modelling results summarized and compared
(test geometry #B).
0,0
1,0
2,0
3,0
4,0
5,0
6,0
7,0
8,0
9,0
X
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
T
20,mf
(sec)
0,0
1,0
2,0
3,0
4,0
5,0
6,0
7,0
8,0
9,0
X
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
EDT
10,mf
(sec)
14,0
16,0
18,0
20,0
22,0
24,0
X A B C D E F G H I J K L M N O P Q
G
mf
(dB)
5% 10% 20% 40%
-10,0
-8,0
-6,0
-4,0
-2,0
0,0
2,0
4,0
X
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
C
50,mf
(dB)
0,20
0,30
0,40
0,50
0,60
X
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
STI (-)
0,20
0,30
0,40
0,50
0,60
X A B C D E F G H I J K L M N O P Q
STI mean-dev (-)
5% 10% 20% 40%
0,0
1,0
2,0
3,0
4,0
5,0
6,0
7,0
8,0
9,0
X
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
T
20,mf
(sec)
0,0
1,0
2,0
3,0
4,0
5,0
6,0
7,0
8,0
9,0
X A B C D E F G H I J K L M N O P Q
EDT
10,mf
(sec)
14,0
16,0
18,0
20,0
22,0
24,0
X A B C D E F G H I J K L M N O P Q
G
mf
(dB)
5% 10% 20% 40%
-10,0
-8,0
-6,0
-4,0
-2,0
0,0
2,0
4,0
X A B C D E F G H I J K L M N O P Q
C
50,mf
(dB)
0,20
0,30
0,40
0,50
0,60
X
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
STI (-)
0,20
0,30
0,40
0,50
0,60
X A B C D E F G H I J K L M N O P Q
STI mean-dev (-)
5% 10% 20% 40%
Table 4. Relation of different parameters expressed by
coefficients of determination, calculated at 10% diffuse
ratio.
To estimate EDT10 of a room, the rate of level change
caused by the effect of the first absorptions is calculated:
∆
∆ =∙.
.
= 10∙
.
()
(5)
where αavg. is the average absorption coefficient, lavg. is the
mean free path between reflections. For a -60 dB decay:
𝑇 =
∆ ∆
⁄=()
∙.{𝑠𝑒𝑐} (6)
which is exactly the Norris-Eyring equation.
Of course, more detailed estimates can consider effects of
directional behavior and source-receiver distance (see [2]).
As suspected, a pre-filtered (e.g. A-weighted) wide band
evaluation might be more meaningful than an arithmetic
average of frequency dependent evaluation results. Oddly
however, simple averages can proximate well:
𝑇, ≈∙∙∙
− 0,06 (7)
and
𝐸𝐷𝑇, ≈∙∙
− 0,06 (8)
where indices 500, 1k, 2k and 4k denote results at 500 Hz,
1 kHz, 2 kHz and 4 kHz octave bands respectively.
6. CONCLUSION
Late reverberation times do not correlate well to clarity or
other monaural measures indeed. Yet they can still be used
to estimate worst case limits of C50 or STI, which is exactly
what a minimum specification should be about.
More general and emphasized use of early decay time is
encouraged in general specifications, since it correlates
well to clarity measures and perceived quality of a room as
well.
Use of STI seems to be superfluous, as it is more complex
to measure or calculate and is meant basically to qualify
nonlinear and noisy transmission, none of which is a room
acoustic feature. If it is important to express clarity in a
simple way, EDT10 is more practical than STI or even C50.
Specifying spectral tolerances cannot be avoided if the
source signal level is high in low frequency bands (e.g.
music or amplification is active).
It seems adequate to specify quantity, uniform distribution
and visibility of absorption to specify sound level reducing
room acoustics measures.
Rooms shall be specified in their furnished, unoccupied
condition, with special consideration if non-architectural
objects affect initial room acoustic conditions when the
room is inhabited.
7. RECOMMENDATIONS
Proposed additions to aim a more complete room acoustic
specification scheme:
- There seems to be a lack of general consensus of what
the optimum limits of values are the best for each room
acoustic parameter to each use of a room. Therefore, a
future-proof scheme shall be used, which aims not the
‘optimum’, but to avoid mistakes.
- Reverberation time is still the best and the simplest
overall descriptor of room acoustic quality. A
maximum reverberation time value guarantees, that
other measures (clarity, intelligibility etc.) will not be
worse than a certain limit.
- A single number figure of reverberation time can be
used both by averaging single band results or
evaluating a frequency-weighted wide band signal.
Difference between usual mean values seems to be
negligible. Using more frequency bands to calculate a
mean value will not substitute using frequency
tolerances. T500Hz+1kHz or TA are the most practical
choices.
- Tolerances of frequency dependent maximum and
minimum limits shall be formed relative to a mean
value to avoid single band deviances and to ensure
overall balance.
- If the source or the receiver is not fixed in a room, only
those parameters shall be used that are proved to be
meaningful even after spatial averaging. Majority of
room acoustic parameters depend on source and
receiver positions, but spatial averages of some
parameters (e.g EDT10, C50, C80, G, etc.) seem to be
able to qualify room acoustic impression quite well.
- Early decay time is a good candidate to be used
generally instead of clarity of speech transmission
index. The benefit of using EDT10 is that it does not
necessarily require to measure room impulse
responses. Nevertheless, due to measurement
uncertainties of direct sound, a more robust version of
EDT10 shall be used. A direct wide-band (or weighted
wide band) evaluation can help to overcome this
problem.
mf A mf A mf A mf A avg avg-dev
- 1,00 0,96 0,89 0,83 0,27 0,14 0,34 0,23 0,52 0,67
log 1,00 0,95 0,90 0,84 0,29 0,15 0,36 0,25 0,55 0,69
- - 1,00 0,94 0,90 0,19 0,07 0,38 0,27 0,58 0,77
log - 1,00 0,95 0,91 0,21 0,07 0,41 0,29 0,60 0,79
- - - 1,00 0,98 0,30 0,12 0,56 0,42 0,73 0,83
log - - 1,00 0,98 0,33 0,14 0,58 0,44 0,75 0,84
- - - - 1,00 0,31 0,12 0,64 0,51 0,80 0,84
log - - - 1,00 0,32 0,12 0,66 0,52 0,82 0,85
mf - - - - - 1,00 0,92 0,43 0,37 0,43 0,14
A - - - - - - 1,00 0,19 0,15 0,18 0,02
mf - - - - - - - 1,00 0,98 0,96 0,64
A - - - - - - - - 1,00 0,89 0,54
G
C
50
STI
std
mf
A
mf
A
T
20
EDT
10
T
20
EDT
10
G C
50
- Looking at values of usual room acoustic parameters
by themselves can be misleading, if rate of diffusion or
the overall diffuse reflection ratio is not known.
Unfortunately there is no objective measure for that
yet.
- Gain or G seems to depend on mainly the quantity of
absorption and does barely correlate to other
parameters. Since G is directly meaningful and is easy
to measure, G shall be used more often to express
related quality criteria.
- STI is not a room acoustic parameter. It shall be used
only for electroacoustic systems, as it is meant to be.
8. ACKNOWLEDGEMENTS
The author would like to thank colleagues Éva Arató,
Gergely Borsi and Attila B. Nagy to provide additional
room impulse response measurement data for this study.
The author would like to thank the Section for Acoustics
of the Hungarian Engineering Chamber to support the
preparation of a room acoustic design guideline, during
which numerous observations were revealed.
9. REFERENCES
[1] W. M. Whitmer, D. McShefferty, M. A. Akeroyd,
“On Detectable and Meaningful Speech-
Intelligibility Benefits” Adv. Exp. Med. Biol. 2016;
894:447-455
[2] M. Skålevik: “Reverberation Time – The Mother of
All Room Acoustic Parameters” proc. 20th ICA, 2010
[3] C. C. J. M. Hak, R. H. C. Wenmaekers, L. C. J. van
Luxemburg: “Measuring Room Impulse Responses:
Impact of the Decay Range on Derived Room
Acoustic Parameters”, Acta Acustica, Vol. 98 (2012)
907-915
[4] D. Byrne, H. Dillon, K. Tran, S. Arlinger, „An
international comparison of long-term average
speech spectra”, JASA 96(4):2108-2120 October
1996
[5] ISO 226:2003 Acoustics — Normal equal-loudness-
level contours
[6] D. Pelegrín-Garcia, B. Rasmussen, J. Brunskog,
“Speaker-Oriented Classroom Acoustics Design
Guidelines in the Context of Current Regulations in
European Countries” Acta Acustica, Vol. 100 (2014)
1073-1089
[7] C. Campbell, C. Svensson, E. Nilsson, “The same
reverberation time in two identical rooms does not
necessarily mean the same level of speech clarity and
sound levels when we look at impact of different
ceiling and wall absorbers”, Inter-noise 2014
[8] Barron, M.: “Theory and measurement of early, late
and total sound levels in rooms.” JASA Vol. 137, No.
6, 2015
[9] ISO 3382-1:2009 Acoustics. Measurement of room
acoustic parameters.
[10] A. T. Fürjes, A. B. Nagy, “Tales of more than One
Thousand and One Measurements (STI vs. room
acoustic parameters - a study on extensive
measurement data)”, preprint uploaded to
www.researchgate.net in May of 2020.
