Identification of sonorant intervals in the speech signal using adaptive representations

GEORGINA FLESIA, ANTONIO GALVES, JESUS GARCIA AND MARCELA MORVIDONE

Campos de Jordao

Workshop; ICTP School and Workshop on Probability Theory and Applications ICTP/CNPq; 2007

ICTP - CNPQ

In this paper we describe a method to localize the sonorant andobstruency regions of the speech signal. It is based on a sparsedecomposition of the sonority trajectories  in terms of afamily of bases or frames. We also show how this decomposition maybe used to estimate the coefficients of a linear model thatrelates the mean sonority and the one step mean sonority of thesignal. We study the accuracy of the model using a corpus of 160phrases previously analyzed by Ramus et al.