1404 Chapter 36
36.14.2 Objective Measures and Techniques
36.14.2.1 Articulation Index
The Articulation Index (AI) was one of the first criteria
and assessment methods developed to use acoustic
measurements and relate these to potential intelligi-
bility. AI is concerned with rating the effects of noise on
intelligibility and was primarily developed for assessing
telephone communication channels. Later corrections
were added in attempt to take account of room reverber-
ation but these methods are not considered sufficiently
accurate for sound system use. AI is a very accurate and
useful method of assessing and rating the effects of
noise on speech. ANSI Standard S3.5 1969 (subse-
quently revised in 1988 and 1997) specifies the methods
of calculation based on measurements of the spectrum
of the interfering noise and desired speech signal.
(Either in terms of -octave or -octave bands.)
The Index ranges from 0 to 1 with 0 representing no
intelligibility and 1 representing 100% intelligibility.
The Index is still very good for assessing the effects of
noise on speech in range of applications where room
reverberation effects are negligible—e.g., communica-
tions channels or aircraft cabins, etc.
Another important application relates to the assess-
ment of speech privacy in offices and commercial envi-
ronments. Here a very low AI score is required in order
to insure that neighboring speech is not intelligible. This
is extremely useful when setting up and adjusting sound
masking systems and a speech privacy scale has been
developed for this purpose. Unfortunately, few commer-
cial analyzers incorporate the measurement, which
would be an extremely simple matter to do if a
-octave real-time spectrum display and data are
available. Currently, most users of AI in this application
either have to compute the result manually or by a sim-
ple spreadsheet procedure.
36.14.2.2 Articulation Loss of Consonants
This method was developed by Peutz during the 1970s
and further refined during the 1980s. The original equa-
tion is simple to use and is in fact based on a calculation
of the D/R ratio, although this is not immediately
obvious from the equation. The long form of the equa-
tion takes into account both noise and reverberation—
but unfortunately does not give exactly similar values to
the simpler form—which is regarded by many to be
overly optimistic. The original work was based on
human talkers and not sound systems. (The original
prediction equation was modified by Klein in 1971 to
its now familiar form in order to do this.)
During 1986, a series of speech intelligibility tests
were run that enabled a correlation to be found between
MRT word scores carried out under reverberant condi-
tions and a D/R measurement carried out on the TEF
analyzer. For the first time this allowed the widely used
predictive and design rating technique to be measured in
the field. However the correlation does have a number
of limitations which need to be considered when apply-
ing the method. The measurement bandwidth used at the
time was equivalent to approximately octave cen-
tered at 2 kHz. Although three very different venues
were employed, each with three significantly different
loudspeakers and directivities, the correlation and hence
method is only valid for a single source sound system.
The measurement requires considerable skill on behalf
of the operator in setting up the ETC measurement
parameters and divisor cursors, so a range of apparently
correct answers can be obtained. Nonetheless the mea-
surement does provide a very useful method of assess-
ment and analysis. In 1989 Mapp and Doany proposed a
method for extending the technique to distributed and
multiple source sound systems by extending the dura-
tion of the measurement window out to around 40 ms.
A major limitation of the method is that it only uses
the 2 kHz band. For natural speech where there is essen-
tially uniform directivity between different talkers, sin-
gle band measurements can be acceptably accurate.
However, the directivity of patterns of many if not the
majority of loudspeakers used in sound systems is far
from constant and can vary significantly with fre-
quency—even over relatively narrow frequency ranges.
Equally, by only measuring over just one narrow fre-
quency band, no knowledge is obtained regarding the
overall response of the system. The accuracy of the
measurement correlation can therefore become
extremely questionable and any apparent %Alcons val-
ues extracted must be viewed with caution.36.14.2.3 Direct-to-Reverberant and Early-to-Late RatiosDirect-to-reverberant measurements or more accurately
direct and early reflected sound energy-to-late reflected
and reverberant energy ratios have been used as predic-
tors of potential intelligibility in architectural and audi-
torium acoustics for many years. A number of split
times have been employed as delineators for the direct
or direct and early reflected sounds and the late energy.
The most common measure is C50, which takes as its
ratio the total energy occurring within the first 50 ms to
the total sound energy of the impulse response. Other(^1) » 1 1 » 3
(^1) » 3
(^1) » 3