Vector Quantization is an important tool of speaker identification system. In this paper, we introduce a new speaker identification system based on improved VQ algorithmu00026;#65292;which takes the code vectors in traditional algorithm and those standard deviation as a new codebook to represent the particular speaker model in training phase.Based on the standard deviation.we redefine a weigthed distortion which are able to reflect the sample distribution character of each code vector .Experimental results show that the identification rate and robustness of this method is better than the one of the reference method.
展开▼
