The improved VQ-MAP and its combination with LS-SVM for speaker recognition

Abstract

Maximum a posteriori vector quantization (VQ-MAP) procedure adapts the mean vectors only and weights were not considered. To solve this problem,this paper proposes the improved VQ-MAP procedure which uses weighted mean vector to replace mean vector. Adaptive parameter sets in the improved VQ-MAP procedure are used as the training samples of least square support vector machines(LS-SVM) in speaker recognition system. According to the results of simulation using Matlab, speaker recognition system based on VQ-MAP and LS-SVM uses less training time of SVMs and it also has high recognition rate.