The overall goal of the proposed research is to investigate the internal spectral temporal space within which speech perception takes place. A guiding premise of the proposed work is the assumption that many of the most important phenomena of speech perception can be explained in terms of general auditory processes. This in mind, experiments are designed to exploit mechanisms listeners use to maintain continuity of speech and other complex sounds under normal environmental conditions which frequently are not optimal. Many studies have demonstrated that listeners are most likely to maintain continuity of a sound source that does not undergo extreme spectral transformation over too short a period of time, but little is known about what sorts of transformation are tolerable. Given that transformations that traverse relatively little space are more likely to be perceived as continuous, the proposed research is designed to map out a spectral temporal space and to describe the and model principles governing perceptual continuity of speech. Along the way, a number of fundamental questions will be addressed concerning the structure of phonetic inventories. The Principle Investigator is co-author of a somewhat controversial theoretical position that holds that many of the most widespread phonetic regularities can be explained by efforts of language communities to arrange their phonetic inventories in a way that exploits auditory predispositions of listeners. This """"""""auditory enhancement hypothesis"""""""" is put to a number of critical tests in attempts to explain near-universal tendencies to use certain types of vowels. It is argued that, for example, general tendencies of languages to have higher pitch for high vowels, and more nasalization of low vowels can be explained in terms of making these vowels auditorily maximally distinctive. By exploiting mechanisms that aid the listener in maintaining perceptual continuity, one can empirically assess the perceptual distance between vowel sounds that are in accordance with these phonetic regularities and vowel sounds that are not. Finally, an extensive modeling effort is described that investigates whether an unsupervised neural network model can be developed that can account for the behavior of human listeners in the proposed experiments.
