Double order hybrid optimum codebook design for speaker recognition

Codebook design is important part of speaker recognition, and it is important research direction how to fast realize precise codebook design. Currently, there are two research methods. They are codebook design methods based on local optimization algorithms and global optimization algorithms. They have drawbacks to some extent for different speakers. Based on it, this paper proposed a double order hybrid optimum codebook design algorithm. This algorithm combines the two representative research methods, realize two codebook design algorithm based on GA (genetic algorithm) and LBG (linde-buzo-gray algorithm) +GA in parallel and obtain final optimum codebook by combination judgment. The experimental results show that, this algorithm can obtain better recognition accuracy and more stable recognition rate.