ITRW on Non-Linear Speech Processing (NOLISP 03) May 20-23, 2003 Le Croisic, France

Double-vowel Segregation Through Temporal Correlation: A Bio-inspired Neural Network Paradigm

Ramin Pichevar (1,2), Jean Rouat (1,2)

(1) Department of Electrical and Computer Engineering, Université de Sherbrooke, QC, Canada
(2) ERMETIS, Universit du Québec, Chicoutimi, QC, Canada

A two-layer spiking neural network is used to segregate double vowels. The first layer is a partially connected spiking neurons of relaxation oscillatory type, while the second layer consists of fully connected relaxation oscillators. A twodimensional auditory image generated by the enhanced spectrum of cochlear filter bank envelopes is computed. The segregation is based on a channel selection strategy. At each instant of time each channel is assigned to one of the sources present in the auditory scene, i.e. speakers. No prior estimation of pitch for the underlying sources is necessary.

Full Paper

Bibliographic reference.  Pichevar, Ramin / Rouat, Jean (2003): "Double-vowel segregation through temporal correlation: a bio-inspired neural network paradigm", In NOLISP-2003, paper 006.