Metody automatické identifikace řečníka z řečového signálu

Abstract

In my bachelor thesis I´m devoting to the problem of speaker recognition from speech signal. Basic theory to understand this problem is included. Theory focuses to analysis and processing of speech signal, description of feature extraction methods of speech signal for speaker identification, description of speech signal preprocessing and description of methods used for speaker identification. In practice part I use knowledge from theory in practice and implement one of methods used for speaker identification in Matlab. I use vector quantization method, which uses features obtained from Mel-frequency cepstral coefficients. Success rate of this method is about 70%. My goal is functional program for speaker recognition with success rate going to 70%. Next use of program is demonstration for teaching.