Study On Method Of Speaker Recognition

As one of the biometrics techniques, speaker recognition is the technology of automatically recognizing who is speaking on the basis of individual information included in speech waves. Because of its advantages on convenience, economy and extensibility, this technique can be applied to a number of areas, such as telephone banking, database access services, remote computer login, security verification and control. Because of that, lots of scientific researchers at home and abroad are involved in the research.The impulse response of the vocal track is an important feature of a speaker . A speech signal is a convolution of glottis excitation signal with the impulse response of the vocal track. This paper introduces two methods to get cepstrum coefficients by deconvolution: linear prediction analysis and homomorphic transformation. After deconvolution, we can extract the cepstrun coefficients related to the impulse response of the vocal track and form the feature vectors.At present GMM is one of the best speaker recognition arithmetic because of its good performance, simpleness and lower complexity. This paper introduces the concept, the parameter estimates and the recognition algorithm of GMM model. The order of GMM model related to speaker recognition performance is discussed and some experimental results are also given.This paper elaborates the principles of applying wavelet analysis theory to speech enhancement, wavelet de-noising algorithm with adaptive threshold value is applied to the preprocessing of speaker recognition system. The results show that this method not only has excellent speech enhancement effects but also can improve robustness of the speaker recognition system in noisy environments.In order to increase the recognition rate of speaker recognition, some features that be used usually in present speaker recognition system is analyzed. A speaker recognition method that combining efficiently more feature such as MFCC, PLCC and their first order difference and pitch period is put forward in this paper. Pitch period is extracted by autocorrelation method. The results show that the method of combining multi-feature is better than the method of using single feature and the ability of speaker recognition is improved.