Designing for Speech Intelligibility 1403- Provide a quiet area or refuge for the announcement
microphone or use a good quality and effective noise
canceling microphone with good frequency response. - Insure that the microphone user is properly trained
and understands the need not to go off mic and to
speak clearly and slowly in reverberant environments. - Repeat important messages.
- In very difficult environments, use simple vocabulary
and message formats. Consider use of high-quality
specially annunciated prerecorded messages. - Consider making improvements to the acoustic envi-
ronment. Do not design the sound system in isolation.
Remember, the acoustical environment will impose
limitations on the performance of any sound system.
36.14 Intelligibility Criteria and MeasurementA number of intelligibility criteria and rating and
assessment methods have already been noted in earlier
sections. Here they are treated in a rather more compre-
hensive overview. However as each technique is quite
complex, readers are referred to the bibliography at the
end of this chapter to obtain more detailed information.
It is obviously important to be able to specify the
desired degree of intelligibility required either for a par-
ticular purpose or so that it can be objectively specified
for a given project or system. The need then also auto-
matically follows that there has to be a corresponding
method of measuring and assessing that a given criterion
has been met. Intelligibility measurement and assess-
ment techniques can be divided into two broad catego-
ries. These are:- Subject based measures—employing a panel of
listeners and using a variety of speech-based test
materials. - Objective acoustic measures of a parameter or
parameters that correlate with some aspect of
perception.
Subject-based measures include writing down word
scores, sentence recognition, modified rhyme tests, and
logotom recognition. Objective acoustic measures
include broadband and weighted SNR, Articulation
Index, Speech Interference Level (SIL and
PSIL),direct-to-reverberant measures (including TEF
%Alcons and C35/C50), and STI. There are also a num-
ber of subsets of these latter techniques.
It should not be forgotten that it is not just sound
reinforcement or public address systems where the
resultant intelligibility may require assessment. Other
related audio applications include telephone and inter-
com systems (telephone/headphone or loudspeaker
based) as well as teleconferencing systems and other
communication channels—e.g., radio. Hearing assis-
tance systems for the hard of hearing can also be
assessed and rated using a number of the techniques
described below as can the effectiveness of noise mask-
ing systems where conversely a reduction in intelligibil-
ity is deliberately sought. Measurements may also need
to be made in order to assess the natural intelligibility of
a space perhaps so that the potential benefits or need for
a speech reinforcement system can be evaluated and
objectively rated (e.g., churches, classrooms and lecture
rooms/ auditoria, etc.).
Not all of the techniques are applicable to every
application. The area of application is therefore noted at
the end of each section. The practical limitations of each
of the methods described are also briefly discussed.36.14.1 Subject-Based Measures and TechniquesThe fundamental measurement of intelligibility is of
course speech itself. A number of techniques have been
developed to rate speech intelligibility. The initial work
was carried out in the 1920s and 1930s and was associ-
ated with telephone and radio communication systems.
From this work the effects of noise, SNR, and bandwidth
were established and subjective test methods formulated.
(Much of this work was carried out at Bell Labs under
the direction of Harvey Fletcher.) The sensitivity of the
various test methods was also established and it was
found that tests involving sentences and simple words
were the least sensitive to corruption but often did not
provide sufficiently detailed information to enable firm
conclusions to be drawn regarding the effects and
parameters under study to be definitely made.
The need to insure that all speech sounds were
equally included led to the development of phonemi-
cally balanced (PB) word lists. Lists with 32, then 250,
and finally 1000 words were developed. Tests using syl-
lables (logatoms) were also developed. These latter tests
provide the most sensitive measure of speech informa-
tion loss but are complex and very time consuming and
costly in application.
The modified rhyme test (MRT) was developed as a
simpler alternative to PB word lists and is suitable for
use in the field with only a short training period. (The
more sensitive methods can require several hours of
training of the subjects before the actual tests can
begin.) The various methods and their interrelationships
are shown in Fig. 36-12 where the Articulation Index is
used as the common reference.