Research On Technology Of Speaker Recognition Based On VQ And HMM

As one of the biometrics technologies, speaker recognition is the process of automatically recognizing who is speaking on the basis of individual information included in speech signal. This technology can be applied to a number of areas, such as banking by telephone, database access services, remote access to computers, security control for confidential information and automatic control system, so it is paid attention to by people more and more in the biometrics technologies realm in recent years.This paper mainly researches how to gain effective individual characteristic parameters of the speaker and select reasonable recognition method to improve the recognition rate and reliability of speaker recognition system. By analyzing the general principles and system architecture of speaker recognition and considering subsistent technologies of speaker recognition, research on the method of extracting the secondary characteristic parameter based on Linear Prediction Cepstrum Coefficient (LPCC) and Mel Frequency Cepstrum Coefficient (MFCC) and combining Vector Quantization (VQ) with Hidden Markov Model (HMM) to apply to the recognition method of speaker recognition.This paper designs a text-dependent speaker recognition system. For getting useful speech section and extracting the characteristic parameter of speaker well, it needs pretreat to speech signal of the speaker, such as denoising, pre-emphasis, framing, windowing and endpoint detection, and get the secondary characteristic parameter based on effective measures to LPCC and MFCC, such as weight, differential, combination and filter. Then use VQ to design codebook for every speaker before HMM to avoid causing domino effect of accumulating errors. At last, use Baum-Welch algorithm and Viterbi algorithm to train and recognize speakers.The experiments in this paper show the speaker recognition system that uses the secondary characteristic parameter based on LPCC and MFCC and adopts the combination of VQ and HMM has the virtue of high recognition rate and low mistake rate.