Date of Award

Spring 1-1-2018

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

First Advisor

Kathryn H. Arehart

Second Advisor

James M. Kates

Third Advisor

Lewis O. Harvey

Fourth Advisor

Neeraja Sadagopan

Fifth Advisor

Brenda Schick

Abstract

We routinely hear speech that is degraded by both noise and reverberation due to the characteristics of our listening environments. These intrusions alter the spectral and temporal envelope cues that support speech perception. The effect of these perturbations is more intrusive for people with hearing impairments.

This work systematically analyzed the potential benefits of restoring the speech envelope information in reverberant speech through modulation spectral analysis, through objective predictions of intelligibility from quantifying the changes in spectrotemporal modulations due to reverberation using cepstral correlation and through listener tests of intelligibility. We examined the benefits of envelope restoration on the intelligibility of reverberant speech using 1. an ideal restoration that used the clean envelope from anechoic speech and 2. Several processing strategies that restored the envelope by expanding the reverberant envelope in multiple bands in both listeners with normal hearing and in listeners with hearing loss. Intelligibility changes in reverberation were shown to occur largely from the changes to the low-rate modulations in the speech envelope. Cepstral correlation was found to be a better descriptor of average performance with and without hearing-loss and individual performance in most of our listeners. Envelope restoration through reinstating the low-rate modulations (< 30 Hz) was found to effectively restore speech intelligibility in all the reverberation conditions tested here. None of the expansion processing schemes provided significant benefits in reverberation. However, the success of the ideal envelope restoration and the usability of low-rate cues indicated the possibility of using similar strategies in signal processing aimed at improving intelligibility in hearing aids.

Share

COinS