Experimentation Research Of Speaker Recognition System Based On Vq And Dtw

As a technology to identify persons' status, speaker recognition picks up persons' individual characters which can distinguish people from others. These characters don't depend on speech. With the advent of information times, communication technology and signal process technology develop quickly, and speaker recognition, fingerprint recognition and iris recognition have been used widely in justice investigation, network bank and code protection area. Speaker recognition will be divided into speaker identification and speaker verification in application way, and it can be divided into text-dependent and text-independent speaker recognition. In this thesis, we do some researches on speaker recognition with VQ and DTW.Speaker identification is not different from speaker verification essentially, both of them identify persons' status by picking up persons' individual characters. We introduce MFCC and△MFCC in this thesis. Studying and analyzing two methods is the emphases of this thesis.In this thesis, we establish a speech database of 100 persons, analyze speaker recognition system based on VQ and DTW, and compare the performance of VQ system with DTW system. We study LPCC, MFCC and△MFCC coefficient etc, analyze the recognition rates by changing rank of MFCC and the change of recognition rates after introducing△MFCC, and adopt abruption method to get original code in VQ system, the code design is optimized by LBG arithmetic. In the end, we compare performance of text-independent speaker recognition system with text-dependent speaker recognition system.