A new method to train VQ codebook for HMM-based speaker identification

In this paper, we build a HMM-based speaker identification system by using a novel method to trained VQ codebook. The codebook is trained based on the criterion of making each codeword in the codebook to share training vectors equally and is implemented with genetic algorithm. An evaluation experiment has been conducted to compare the codebooks trained by the Linde-Buzo-Grey (LBG) and the new algorithm. It is showed that the codebook trained with the new algorithm can give a much higher speaker identification rate than the LBG trained codebook when used in HMM-based speaker identification, especially for text-independent speaker identification.