Handbook for Sound Engineers

Designing for Speech Intelligibility 1389

three bands therefore provide over 75% of the available
spectral intelligibility content.

Whereas the range 300–3000 Hz has been shown to
be adequate for telephone intelligibility, a wider range is
generally required for sound system use—particularly
under more difficult acoustic conditions. This effect is
shown in Fig. 36-6. This contrasts the results of tele-
phone (monophonic listening) with some recent
research carried out by the author in a reverberant space
(RT 60 = 1.5 s). The upper curve after Fletcher (1929)
shows that the contribution to intelligibility hardly
increases beyond 4 kHz, while the lower curve, made on
a system in a real space (binaurally) shows improve-
ments occurring up to 10 kHz. The need for an extended
bandwidth can therefore immediately be seen. Limited
bandwidth should not be a problem with modern sound
system equipment and loudspeakers. However, there are
some notable exceptions. These include:

1. Inexpensive poor-quality microphones.


2. Some re-entrant horn loudspeakers (or CD horn
   drivers used without equalization).


3. Some inexpensive digital message stores.


4. Miniature, special purpose loudspeakers.

Many potentially adequate sound systems are often

let down by employing a cheap or restricted bandwidth

microphone at the front end of the system. In the

author’s experience, even on a basic paging system

employing restricted bandwidth loudspeakers—e.g.,

re-entrant horns—the difference between a microphone

with a reasonably wide and well-controlled frequency

response can always be readily identified over one with

a restricted response, even if it exceeds the response of

the loudspeakers themselves. Rubbish in equals rubbish

out is certainly the case here. However, when operating

under high-background noise conditions, a compromise

may need to be reached between optimal frequency

response and optimal noise rejection, as the two parame-

ters are often divergent.

Apart from component equalization (or the lack of it)

by far the most common problems associated with sys-

tem frequency response stem from either loud-

speaker/boundary-room effects or interactions between

closely spaced (multiple) loudspeakers. Fig. 36-7 shows

the effect of positioning a high-quality monitor loud-

speaker with an impeccably flat response close to a

boundary wall. As can be seen the response is now far

from flat!

Equalization alone cannot correct for this problem.

Reduction of the peaks is possible but the notches in the

response cannot be equalized out as they are caused by

complex phase interactions that cannot be corrected by

means of frequency filtering. Interaction between loud-

speakers is a common problem in cluster design, where

the radiated wave fronts can suffer from missynchroni-

zation due to different acoustic path lengths occurring—

e.g., due to differences between acoustic centers. Fig.

36-8 shows a typical interaction problem (after Davis

and Davis).

Figure 36-5. Octave band percentage contributions to
speech intelligibility.

Figure 36-6. Effect of frequency bandwidth on speech intel-
ligibility. Upper curve—monophonic listening (after
Fletcher). Lower curve—binaural listening (after Mapp).

Octave Band Contributions to Speech Intelligibility

Frequency (Hz)

%

40

30

20

10

0

125 250 500 1k 2k 4k 8k

Speech Intelligibility versus Bandwidth–SNR >20 dB

1k 2k 4k 10k

Frequency–Hz

Inte lligibility–%

Fletcher

In room

50

60

70

80

90

100

Figure 36-7. Effect of local boundary interaction on loud-

speaker frequency response.

10 dB

20 50 100 200 500 1k 2k 5k 10k 20k

Frequency–Hz

5 dB

31 63 125 250 500 1k 2k 4k 8k 16k

(^1) / 3 Octave band center frequency–Hz