计算机工程与应用 ›› 2009, Vol. 45 ›› Issue (10): 168-171.DOI: 10.3778/j.issn.1002-8331.2009.10.051

• 图形、图像、模式识别 • 上一篇    下一篇

基于类内方差归一化和SVM的说话人识别方法

高新建,李弼程,屈 丹   

  1. 解放军信息工程大学,郑州 450002

  • 收稿日期:2008-02-19 修回日期:2008-04-29 出版日期:2009-04-01 发布日期:2009-04-01
  • 通讯作者: 高新建

Method for speaker recognition based on With-in Class Covariance Normalization and SVM

GAO Xin-jian,LI Bi-cheng,QU Dan   

  1. PLA Information Engineering University,Zhengzhou 450002,China
  • Received:2008-02-19 Revised:2008-04-29 Online:2009-04-01 Published:2009-04-01
  • Contact: GAO Xin-jian

摘要: 支持向量机(SVM)由于其强大的分类能力,引起人们广泛的重视,并且成功地应用于说话人识别。其中基于GLDS核的SVM系统性能比较优异。引入类内方差归一化(WCCN)方法来处理SVM的输入特征向量,并和GLDS核相结合,提出一种基于类内方差归一化和SVM的说话人识别方法。该方法利用WCCN方法对SVM的输入特征向量进行变换,增强特征向量的类间区分能力,再采用GLDS核函数进行SVM的训练,以提高SVM的分类效果。实验表明,新方法是有效的,其性能优于基于GLDS核的SVM系统。

Abstract: SVM have shown its powerful ability to do pattern classification,so people pay more and more attention to it.Furthermore,SVM have proved to be an effective technique for speaker recognition.Recently,a new SVM kernel function GLDS,has shown better performance than conventional SVM kernel.This paper mainly introduces a method WCCN to process the SVM input vectors.And combined with GLDS a method based on WCCN and GLDS is proposed.It can boost the feature-vectors’ differentiability between classes and improve the SVM’s classification performance.This paper employs the method of WCCN combined with GLDS to conduct speaker recognition.Experiments show that WCCN effectively improve SVM system’s classification performance.