Mel Frequency Subband Coefficients Based Text Independent Speaker Recognition

Speaker recognition is the task that is used to identify or verify who is speaking by analyzing and recognizing specific information extracted from speakers'speech. As an important branch research field of speaker recognition, text-independent speaker recognition has carried out all over the world and plays a more and more important role because of its ease-to-use and highly popular application in the information technology and it will have a significant future. This paper focused on the problems of the codebook formation, feature extraction and the noise robustness, and experiments proved that the results obtained were good.It is an important part to design the codebook for both VQ based and GMM based speaker recognition system. In the VQ based system, the codebook affects the performance of the whole system. The codebook formation is also a part of the initialization process of the GMM based system. According to the algorithm of LBG and Kohonen, a modified algorithm of codebook formation was proposed to improve the performance of the codebook.Finding new speech feature and combination of existing speech features are hotspots of research on speech feature extraction. Through modifying and combining the Mel Frequency Subband Coefficients,the paper enhanced the difference of the frequency between the people in the system, improved the ability of separating the classes in the feature space. And the parameters this paper got could emphasize people's individuality and increase the average accuracy of recognition.In the end, based on the analysis of the noise robustness of each subband, a robust recognition algorithm was presented and it could increase the accuracy of recognition in...