Research Of Speaker Recognition System Based On VQ And HMM

Speaker recognition is the process of automatically recognizing who is speaking on the basis of individual information include in speech signals. It has well application prospects in many fields. By analyzing the general principles and system structure of speaker recognition and considerating subsistent technology of speaker recognition. Linear prediction cepstrum coefficient (LPCC) and Mel cepstrum coefficient(MFCC) characteristic parameter are combined together, the vector quantization (VQ ) is combined with Hidden Markov's model (HMM) and applied to the discernment method that the speaker recognition, set up speaker's recognition system.This dissertation has analyzed since pretreatment of the signal of the pronunciation, measure the extreme point to the signal of the pronunciation, filter except silent section and noise of the signal of the pronunciation , has offered the effective pronunciation section for abstraction of the characteristic parameter of the pronunciation. Traditional pronunciation extreme point detection method and on the basis of LPCMFCC extreme point performance of detection method have relatively also in the article, draw the conclusion: On the basis of LPCMFCC extreme point detection method can finely appear pronunciation come by extreme point to measure under the environment of high noising.This dissertation uses the limit model completely, draw LPCC of the signal of the pronunciation, was it appear cepstrum coefficient, got linear prediction cepstrum coefficient and the difference divide to derive, used to describe the dynamic change of speaker's sound channel. Selected for use MFCC and the difference and divided the characteristic parameter as phonetic recognition, to describe the non-linear characteristic of frequency of sense of hearing of ears of people.This dissertation drew the characteristic parameter of the introduction pronunciation through MATLAB speech processing toolbox , adopted LPCC and the difference and MFCC and the difference combine to discern, improved systematic discernment performance , introduced LBG algorithm going on yards of book design , Baum-Welch algorithm trained , Viterbi algorithm was discerned, and turned every speaker and design one yard of books in the amount of vector adopted of model front of Hidden Markov, avoided the quantization error that the speaker usesthe same yard of books to bring, then the error went into the next class HMM discern, caused the error to accumulate the effect, gotten very high discernment rate. The advantage was Fast operation, the calculating amount, the low mistake rate.