Speaker Recognition Based On The Improvement Of LBG Code Generation

As an important part of speech signal processing technology, speaker recognition has a very bright future for both research and applications. Many kinds of speaker recognition with their own characteristics have already come into being. This paper is based on the study of the speech signal processing, and then we adopt the theories in speaker recognition design.At first, this paper clarifies the principle of speaker recognition, then through the comparison of different acoustic characteristic parameters, we choose the parameter MFCC at last. Combined with vector quantization technology, the successful realization of the speaker recognition system has been developed.Because of the rigor of theory, convenience and better design effect in LBG algorithm, LBG algorithm is adopted in clustering feature codebooks. But the traditional LBG algorithm has some issue on the splitting of codebooks. We finally solve it in two ways, the first method use a variable as the splitting parameter, which is to make a variable step to improve the generation speed; The second is to change the original codeword, which was one in traditional, we add a new codeword in the distortion direction, so that the splitting codebook could get a better quality. At last we adopt the improved code generation LBG algorithm, and make a better recognition rate.At last, based on the speaker recognition, we make some discuss on the Language identification and point out what we should improve on this field.