Research On Text-independent Speaker Rre Cognition System

Speaker recognition is a kind of biological certification technology and it makes use of the speech coefficients which represent the speaker's physiological and physical feature to identify speaker. Based on the context of speech signal, speaker recognition can be divided into text-dependent and text-independent. This thesis select text-independent speaker recognition for study due to more flexible and widely application.In the text-independent speaker recognition, the GMM shifts the problem of speaker recognition to the problem of the estimation of distribution of training data. Thus, it divides more complex problems of data training and pattern matching into some simple problems, such as parameter estimation and computation of probability. Also, GMM has characteristics of simple, flexible and robust. So it is the-state-of-art in text-independent speaker recognition.In aspect of system construction, this paper describes the implementation of a full speaker recognition system by visual C++, including speech signal processing, feature extracting, model training and recognition. It uses Mel frequency cepstral coefficients (MFCC) as feature parameter. It also uses GMM for speaker modeling.The study work of this thesis has several aspects:1. This paper has studied that the performance of GMM relies on training data and testing data, especially training data. And it is verified by experiment that under limited training data, the GMM can reach the best performance with a number of Gaussians and will reduce the performance by continuously increasing a number of Gaussians.2. This thesis has tested different training methods, such as maximum likelihood training, discriminative training. The discriminative training was quite efficient to improve the performance of speaker recognition by experiment.3. A new method which optimize the GMM with clustering algorithm for greatly reducing number of Gaussians was proposed by this paper. The experiment demonstrate that this method speed up recognition rate without excessively degrading the recognition accuracy.