Speech intelligibility in complex environments: Modeling and possible applications

Abstract

A summary of recent modeling work from our lab on the topic of speech intelligibility in complex environments will be presented, including some results from experiments designed to evaluate the models. The primary focus will be models of binaural processing, with the dual goals of (1) understanding the processing within the brain and (2) suggesting possible strategies for external processing that could provide more useful acoustic inputs for listeners with impaired hearing. Several processing algorithms based on allocation of individual time-frequency slices will be considered, including one based on EC processing and one based on local (in time-frequency) estimates of interaural time and intensity differences and interaural coherence. Performance of these algorithms will be evaluated using statistics of source-separation accuracy (with the ideal binary mask as the “golden standard”) and also using human listening experiments. In the listening experiments, the waveforms generated by combining the time-fre...