(Dr. Peter Rogers, advisor)

"Human Capabilities and Neurological Models of Dereverberation"

Abstract

Limited research has addressed the auditory systems capabilities to process reverberant signals. This study hopes to answer two questions: 1. Can humans dereverberate sound? 2. Is the binaural cross correlation model capable of removing reverberation?

Word comprehensibility testing will be performed on human subjects using both realistic reverberation and random convolutional noise as parameters. A comparison of results will demonstrate if realistic reverberation is more understandable than random convolutional noise. In a similar fashion to the above tests, monaural and binaural capabilities will be investigated, looking at both binaural masking release and binaural reverberation.

In conjunction, binaural auditory models are being investigated to determine if they are feasible for the reconstruction of non reverberant signals. The primary focus will be on the Jeffress/Lindemann cross correlation model that is capable of reproducing many related psychoacoustic phenomena. Traditionally this model makes use of similarities in both input channels for determining source direction. Extensions to this model will explore its usefulness for reconstructing a less reverberant signal. The processing algorithms developed will be tested on two channel real room recordings.