Workshop on the Auditory Basis of Speech Perception Keele University, UK July 15-19, 1996

Temporal and Rate Aspects of Speech Encoding in the Auditory System: Simulation Results on TIMIT Data Using a Layered Neural Network Interfaced With a Cochlear Model

Li Deng, H. Sheikhzadeh

Department of Electrical and Computer Engineering, University of Waterloo, Waterloo, Ontario, Canada

A study on temporal and rate aspects of the auditory representation for major manner classes of speech sounds in American English, using fluent speech examples excised from TIMIT database, is reported in this paper. A modeling approach is taken in which a cochlear model is used to generate parallel sets of auditory-nerve (AN) instantaneous firing rates in response to the TIMIT utterances. These temporal responses at ANs are further fed to a layered neural network (NN) which includes such neural mechanisms as lateral inhibition, coincidence detection, and short-term temporal integration of post synaptic potentials for action potential generation. Correspondence between a temporal-nonplace code at ANs and a rate-place code at the NN model output is shown and discussed.

Full Paper

Bibliographic reference.  Deng, Li / Sheikhzadeh, H. (1996): "Temporal and rate aspects of speech encoding in the auditory system: simulation results on TIMIT data using a layered neural network interfaced with a cochlear model", In ABSP-1996, 75-78.