Designing for Speech Intelligibility 1391ularly the elderly or those suffering even a mild hearing
loss, may miss certain words or strain to hear, even under
quiet conditions. Although normal face to face conversa-
tion may take place at around 60 dBA, regularly listeners
demand higher sound pressure levels from sound
systems, with 70–75 dBA being typical for conference
systems even when under quiet listening conditions.
In noisy situations, it is essential that a good SNR is
achieved. Various rules of thumb have been developed
over the years. As a general minimum, 6 dBA isrequired and at least 10 dBA should be aimed for.
Above 15 dBA there is still some improvement to be
had, but the law of diminishing returns sets in for most
practical systems.
There is some disagreement among the generally
accepted reference data. Fig. 36-11, for example, shows
the general relationship between SNR and intelligibility.
As the curve shows, this is an essentially linear relation-
ship. In practice, the improvement curve flattens out at
high signal-to-noise ratios—though this is highly depen-
dent on the test conditions. This fact is shown in Fig.
36-12, which compares the results of a number of stud-
ies, using different test conditions and signals.The curve, for example, shows that for more difficult
listening tasks, the greater the SNR has to be in order to
achieve good intelligibility. Fig. 36-13 shows the effect
of SNR on the %Alcons intelligibility scale. Here, the
improvement can be seen to clearly flatten out above
25 dB SNR. Under high noise conditions, such a SNR
could demand excessively high SPLs and caution must
be exercised.
Where noise is a particular problem, a full spectral
analysis should be carried out. Ideally this should be in
terms of octaves but for many applications octave
band analysis will be adequate and certainly more infor-
mative than a single dBA value. Fig. 36-14 shows such
an analysis.
In the upper curve, which depicts a positive SNR, it
can be seen that the speech signal is greater than the
noise over each of the octave band frequencies. How-
ever, in the lower curve, it can be seen that at high fre-
quencies the noise exceeds the desired speech signal.
The overall effect on potential intelligibility can be cal-
culated by looking at the individual octave band SNRs
and then weighting and summing them in accordance to
their relative contributions as shown earlier in Fig. 36-5.
This is the basis of the Articulation Index (AI), which
is a good measure for determining the effects of noise on
speech—either over single channel transmission lines
such as telephone or radio communications or over PAFigure 36-10. ETC curves of the two loudspeakers shown in
Fig. 36-9.
Vertical: 6 dB/division
Horizontal: 4000 - 16,639 Ms
A. In synchronization.Vertical: 6 dB/division
Horizontal: 4000 - 16,639 Ms
B. Out of synchronization.Figure 36-11. Effect of SNR on speech intelligibility.020406080100120(^15
5 0 5 1015 20)
SNR—dB
Intelligibility—%
(^1) » 3 1 » 1