摘要： 针对说话人语音特征随音量、情绪、健康等因素变化呈现出的复杂分布结构，提出一种基于保局部核相关向量机(RVM)的说话人识别方法。在RVM模型所采用的高斯核函数中引入相似度因子，以保留数据局部结构，构成保局部核RVM模型。在模型训练过程中采用快速算法以避免大型矩阵逆操作，减少计算量，可适用于大样本场合。应用结果表明，该方法能加快测试速度，提高分类精度。

关键词: 说话人识别, 保局部核, 相关向量机, 高斯核函数, 类内相似度

Abstract: Taking account of the complex structure of the speech features, which is affected by the change of volume, emotion, health and other factors, a new method for Speaker Recognition(SR) based on Relevance Vector Machine(RVM) using locality preserving kernel is proposed. RVM using locality preserving kernel introduces intra-class similarity into Gaussian kernel function to keep the data set’s neighborhood structure, and is applied into SR. For the purpose of avoiding the inverse matrix operation and applying to a larger sample, the new method uses a fast algorithm for training. Experimental results show that the new classifier model speeds up the test speed and improves the classification accuracy.

Key words: Speaker Recognition(SR), locality preserving kernel, Relevance Vector Machine(RVM), Gaussian kernel function, intra-class similarity

中图分类号: 

• TP18