基于保局部核RVM的说话人识别方法  

Speaker Recognition Method Based on RVM Using Locality Preserving Kernel

作　　者：郑泽萍[1] 王万良[1] 郑建炜[1] 

机构地区：[1]浙江工业大学计算机科学与技术学院,杭州310023

出　　处：《计算机工程》2011年第14期208-210,共3页Computer Engineering

基　　金：国家自然科学基金资助项目(61070043)

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。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.

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 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]