Abstract: The traditional Support Vector Machine（SVM） in speaker recognition has high computational complexity.To solve this problem，this paper proposes a kind of speaker recognition system based on VQ-MAP and SVM which formulates Maximum A Posteriori Vector Quantization（VQ-MAP） procedure that adapts the mean vectors only.The result is the adapted speaker model and the parameter vectors of this model are used as the support vectors of SVM for speaker recognition.According to the results of simulation using Matlab，speaker recognition system based on VQ-MAP and SVM has significantly reduced computational complexity and the training time of SVM is short and it also has high recognition rate.

Key words: Vector Quantization（VQ）, Maximum A Posteriori（MAP） adaptation, Support Vector Machine（SVM）, speaker recognition

摘要： 针对传统支持向量机（SVM）在说话人识别中运算量过大的问题，提出了VQ-MAP和SVM融合的说话人识别系统。它应用仅自适应均值向量的最大后验概率矢量量化过程（VQ-MAP），来得到自适应的说话人模型，用此模型中的参数向量作为支持向量应用于SVM来进行说话人识别。用Matlab进行仿真实验，结果表明，基于VQ-MAP和SVM融合的说话人识别系统大大降低了运算量，SVM训练时间短，且具有较高的识别率。

关键词: 矢量量化（VQ）, 最大后验概率（MAP）自适应, 支持向量机（SVM）, 说话人识别