Predicting confusions and intelligibility of noisy speech

Abstract

Current predictors of speech intelligibility are inadequate for making predictions of speech confusions caused by acoustic interference. This thesis is inspired by the need for a capability to understand and predict speech confusions caused by acoustic interference. The goal of this thesis is to develop models of auditory speech processing capable of predicting phonetic confusions by normally-hearing listeners, under a variety of acoustic distortions. In particular, we focus on modeling the Medial Olivocochlear efferent pathway (which provides feedback from the brain stem to the peripheral auditory system) and demonstrate its potential for speech identification in noise. Our results produced representations and performance that were robust to varying levels of additive noise and which mimicked human performance as measured by the Chi-squared test.