Model of Selection of Characteristic Parameters of a Voice Signal for Announcer Identification Tasks

Abstract

A method for isolating the parameters of the voice signal characteristic of the speaker by selecting the features of the extremums of the surface of the scalogram of the voice signal, built on the basis of wavelet coefficients. The structural model of the speech signal is described, the sequence of mathematical transformations of the information channel of speech communication is considered and generalized, on the basis of which the approach to selection of features of language structures is offered. An algorithm for selecting the characteristics of speech signal structures has been developed. The selection uses a complex Morlet wavelet, which most effectively describes the structure of the speech signal at the first level of decomposition. By synthesizing the speech signal with the specified characteristics, the rational parameters of the wavelet transform settings for the implementation of the method are selected.