Application of Cortical Processing Theory to Acoustical Analysis

The goal of this STTR program is to formulate a template-matching operation, with perception-related rules of integration over time and frequency at its core, in the context of human perception of degraded speech. In Phase I we focused on implementing signal processing principles realized by the auditory periphery in providing the observed graceful degradation of human performance in noise, a prerequisite to success in the overall goal of the program. We have implemented a closed-loop model of the auditory periphery with efferent-inspired feedback, which produces spectrograms of noisy speech that are more consistent with spectrograms of clean speech than are spectrograms produced by open-loop models of the auditory periphery. In Phase II we shall use this model as a front-end in a system designed to predict human consonant confusions, under a variety of acoustic distortions. A template-matching operation will serve as the back-end component, identifying an input pattern generated by the peripheral model by matching it against internal templates. We shall tune the auditory model first, by means of a psychophysical task designed to reduce errors due to the back-end to a minimum (phone discrimination). After the front end has been optimized, the parameters of the template-matching operation will be determined by means of a phone identification task.