Speaker Recognition Using Composite Vector Stochastic Processes Model representation

Abstract

The authors presenting a study of the automatic speaker verification for short utterances. The verification method of the speaker using word's sound parametric spectrum factorization in composite vector stochastic process representation on the base of multiplicative autoregressive model is presented in the paper. The developed method enables to receive the words features with stable characteristics for the same speaker and differ for the different speakers. The results presented in the paper showed the high correct identification probability. The features analysis for different words showed that these features also can be used in the task of connected speech recognition.