A Text-Independent Speaker Recognition Algorithm And Its System Based On DSP

In this thesis, the methods of text independent speaker recognition are studied.In algorithm ways, Gaussian mixture model(GMM) is the most successful speaker recognition model at present. LBG algorithm is used to initial positions of the GMM centers usually before training the parameters of GMM by Expectation Maximum algorithm. But tend to get into local minimum and must specify the amount of the GMM centers is the disadvantage to this method. In the paper, Based on artificial immunology and Expectation Maximum algorithm, a hybrid algorithm to specify the parameters of the GMM is proposed. An artificial immune mechanism for data clustering is used to adaptively specify the amount and initial positions of the GMM centers according to input data set; then the GMM is trained by Expectation Maximum algorithm .The preliminary experimental results show that the algorithm have higher recognition rate than traditional algorithm. In system implement, We utilize DSK6713 board to put up a speaker identification system. The system can be applied to speaker identification experimentation. It verifies the feasibility of the system.