"Predicting the intelligibility of cochlear implant processed speech"
<br>
by Ray Goldsworthy
<p>
ABSTRACT
<p>
The Speech Transmission Index (STI) is a physical metric that is well
correlated to the intelligibility of speech degraded by additive noise and
reverberation for normal hearing listeners. STI is calculated from the
envelopes of original and degraded (or processed) signals in a number of
frequency bands spanning the speech spectrum. Since the majority of
cochlear implant (CI) stimulation strategies are based on the envelopes of
a number of frequency bands, we expect that STI is well suited to
predicting intelligibility scores of CI users. As a first step, we
evaluate STI's ability to predict the intelligibility of speech for normal
hearing listeners when listening to speech processed through noise-vocoder
simulations of cochlear implant speech processing.
<p>
We investigate the effects of acoustic degradation, spectral subtraction,
and binaural noise reduction on the intelligibility of noise-vocoder
processed speech.  Based on the results of these experiments, we conclude
that 1) both spectral subtraction and binaural noise reduction can
significantly improve intelligibility for CI users and that 2) only the NCM
is an accurate and useful tool for investigating acoustic degradation,
spectral subtraction, and binaural noise reduction on the intelligibility
of CI processed speech.