1408 Chapter 36most of the problems associated with RaSTI. However,
just as with STI, it is a far from perfect measure and there
are a number of limitations that need to be understood.
STIPA is vulnerable to some forms of digital signal
processing and in particular CD player errors. STIPA
test signals are generally distributed on CDs, and some
CD players can introduce significant errors. It is there-
fore essential to conduct a loop back measurement to
insure that a valid signal is being generated. More
recently, however, signals have been distributed as wav
files on solid state memory cards and can also can be
directly downloaded, which helps overcome the prob-
lem (see Mapp 2005 for further details). Most hardware
implementations include some form of error detec-
tion—particularly the detection of the occurrence of
impulsive noise during a test. Not all STIPA meters have
incorporated the level dependency relationship that
exists between speech level and intelligibil-
ity—although it is clearly defined within the IEC 268-16
standard. The same is also true of the masking function
that the standard requires.
One of the major shortcomings of STI and STIPA is
their inability to correctly assess the effect that equaliza-
tion can have on a sound system. For example, STI mea-
surements made on the system described earlier and
whose frequency response is depicted in Fig. 36-27
were exactly the same pre and post equaliza-
tion—although the word score intelligibility improved
significantly. Adapting STI to correctly account for such
occurrences is not a simple or straightforward matter
and it will be some time yet before we have a measure
that can accurately do this.
The STI/RaSTI scale ranges from 0 to 1. Zero repre-
sents total unintelligibility while 1 represents perfect
sound transmission. Good correlation exists between the
STI scale and subject-based word list tests. As with all
current objective electroacoustic measurement tech-
niques, STI does not actually measure the intelligibility
of speech, but just certain parameters that correlate
strongly with intelligibility. It also assumes that the
transmission channel is completely linear. For this rea-
son, an STI measurement can be fooled by certain sys-
tem nonlinearities or time-variant processing. STI is
also liable to corruption by the presence of late discrete
arrivals (echoes). These, however, can be readily spotted
by examination of the modulation reduction matrix.
The basic equation for the STI modulation reduction
factor m(f) is(36-5)Unfortunately, this equation cannot be directly
solved, making STI prediction a complex procedure
requiring detailed computer modeling and analysis of
the sound field. An approximate relationship exists
between STI (RASTI) and %Alcons. Fig. 36-34 shows
the two scales while Table 36-4 gives a numerical set of
equivalent values.The subjective scale adopted for STI (and
RaSTI/STIPA) has led to considerable confusion when
rating PA and sound systems. (For example a rating of
0.5 STI would normally be rated as good if heard in a
highly reverberant or difficult acoustic environment
rather than only fair. Also, in practice, there is usually a
marked difference in perception between 0.45 and 0.50
STI (and more particularly 0.55 STI)—although they
are all rated as fair. In an attempt to overcome the prob-
lem and also to add a degree of tolerance to the measure,
the author has proposed that a new rating scale be
employed for PA/sound systems (Mapp 2007). The pro-
posed scale is shown in Fig. 36-36 and is based on a
series of designated bands rather than absolute catego-
ries. While the bands will remain fixed, their application
can vary so that, for example, an emergency voicem f
11 2 FT
13.8----------2
+-----------------------------^1110– S/N
+ -----------------^10= u----------------------------Table 36-4. RaSTI and %Alcons Numerical Set of
Equivalent Values
Quality RASTI %Alcons Quality RASTI %Alcons0.20 57.7 0.62 6.0
0.22 51.8 0.64 5.3
0.24 46.5 0.66 4.8
0.26 41.7 0.68 4.3
BAD 0.28 37.4 0.70 3.8
0.30 33.6 0.72 3.4
0.32 30.1 0.74 3.1
0.34 27.0 GOOD 0.76 2.8
0.36 24.2 0.78 2.5
0.38 21.8 0.80 2.2
0.40 19.5 0.82 2.0
0.42 17.5 0.84 1.8
POOR 0.44 15.7 0.86 1.6
0.46 14.1 0.88 1.4
0.48 12.7 0.90 1.3
0.50 11.4 0.92 1.2
0.52 10.2 EXCELLENT 0.94 1.0
0.54 9.1 0.96 0.9
0.56 8.2 0.98 0.8
FAIR 0.58 7.4 1.0 0.0
0.60 6.6