关键词: 说话人识别, 概率序列核, 通用背景模型, 支持向量机

Abstract: This paper proposes a probabilistic sequence kernel based on the universal background model.The Gaussian components are used to construct the speaker character space，and the utterances with different length are mapped into the fixed size vectors after normalization with correlation matrix.Then four feature normalization methods is proposed for the mapped vectors：mean/variance normalization，weight scaling，WLOG scaling and spherical normalization.Finally the normalized vectors are inputted to the linear support vector machine for speaker recognition.A transfer probabilistic sequence kernel is also proposed，which adapts the transfer information between neighbor frames.The experiments on NIST 2001 show that the probabilistic sequence kernel is compared with the traditionally UBM-MAP model and the performance will be improved clearly after the linear fusion of the models.

Key words: speaker recognition, probabilistic sequence kernel, universal background model, support vector machine