SP395 SoundPro Audio Integrator
Form7492
Operation Manual
63
RASTI Caveats
RaSTI can be used to evaluate intelligibility, and is considered superior to some other methods,
particularly ALCONS. However, it is possible to get false results. To get the best possible
results, watch out for these problems:
Out of band noise.
Since RaSTI looks only at the 500Hz and 2000Hz octave bands, it will not
take into account noise outside of these bands. For example, imagine a very loud signal at 4 kHz.
This could cause terrible intelligibility yet show a good RaSTI score.
Non-linear systems.
Driving a system into distortion, or running the signal through compressors
or other non-linear systems can cause false readings.
Bad “talker” level.
If the talker level is not within normal limits, very low room noise can cause
a bad score, or a very loud test signal can mask room noise. Set the talker (test signal) level at
expected speech or system signal levels.
If these cases are avoided, the results will correlate well with STI.
Speech Intelligibility – STI-PA
STIPA is a speech intelligibility measurement derived from the more extensive STI standard.
STI specifies 98 data points, each an octave band of pink noise modulated by a slow sine wave
envelope. Octave bands from 125Hz to 8000Hz are specified, and for each octave band, up to 9
different sine wave modulations. For a single data point, imagine generating octave band pink
noise and changing the volume from off to maximum to follow a slow sine wave. The combined
STI signal for a set of points sounds like a steam locomotive.
For each data point, the amount of the modulation that is lost in an acoustic space is measured,
either directly or indirectly, and from this “loss of modulation” the speech intelligibility is
derived. Things that can affect loss of modulation are primarily reverberation time in the room
and the noise level, in the particular octave band being analyzed.
After the modulation loss numbers have been computed for each data point, they are combined
mathematically to arrive at a single value. This value is referred to as “STI” and ranges from 0.0
(no speech intelligibility) to 1.0 (best possible speech intelligibility.
Talker
Listener
STI Signal
Generator
STI
Analyzer
Direct signal
Noise
Reverberation
