Exploiting redundancy to construct listening systems

Abstract

This report highlighted some of the intricate relationships between the statistics of natural sounds and auditory processing. We have argued that many of the common steps that we often take to perform computational audition can be seen as processes driven by the nature of sound, and not so as steps inspired by human physiology or engineering. We have shown how different aspects of hearing can be explained using a simple and common rule exploiting the statistical structure of sound. Although the methods we employed are very simple, the results are just as promising as using any other more complex approach. We hope that the simplicity and the elegance of this approach will inspire further work along these lines, and give rise to more investigations in the field of computationally evolving audition. © 2005 Springer Science + Business Media, Inc.