Research On The Speaker Recognition Based On VQ And HMM

The speaker recognition is a kind of biological certification, it is a technical process of automatic recognizing the identity of speakers from the voice who is speak. In many fields, speaking recognition has a well application prospect and huge market potential. By analyzing the general principles and the system structure of the speaker recognition and refer to subsistent technology of speaker recognition, and researched the feature of speech signal, extraction and recognition method. Processing of the speech feature extraction was improved, and had has got a good experimental results.According to the Chinese phonetics, single vowel, then extracts of each vowel press short-term energy extracted from low to high then the MFCC will be calculated by the group of short-term signals. In pattern recognition, the Vector Quantization and Hidden Markov Models for each vowel modeling, the results will be matched by a set logical judgment.This paper studies the principle and method of Vector Quantization and Hidden Markov Models in speaker recognition, and the main research work are as follows:(1) Analysis of the main characteristics of phonetic features studied, and the model based on VQ & HMM speech recognition of feature extraction method.(2) HMM based on the model with VQ identification method is analyzed, and the principle of the process for identifying, can simplify recognition model and improve recognition method.(3) Through the research and practice of vowels, based on the signal for modeling method respectively, MFCC extraction process were improved, and the application to the speech recognition.Finally analyzes and summarizes the results of the experiment.