14th International Congress of Phonetic Sciences (ICPhS-14)

San Francisco, CA, USA
August 1-7, 1999

A Hierarchical Model of Speech and Voice Perception based on Time Resolution

Jean-Sylvain Liénard

LIMSI-CNRS, Orsay, France

The theoretical model presented aims primarily at explaining the large variability encountered in speech analysis. It implements a joint processing of speech and voice, and a hierarchy of 6 levels based on time resolution: cochlear output, acoustical indices, phonetic features, phonemes, words, concepts. Perception is viewed as the cooperation of two information flows, bottom-up and top-down. The hierarchy employs processing modules of the same type, one between any pair of adjacent levels. The function of the module is to associate two time-varying patterns. Then each processing step is presented, with its specific units and its time characteristics. Finally some phonetic notions and problems are examined in the light of the model: pitch, formants, vowel normalization, motor theory, relations between voice and speech.

Full Paper

Bibliographic reference.  Liénard, Jean-Sylvain (1999): "A hierarchical model of speech and voice perception based on time resolution", In ICPhS-14, 1839-1842.