• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

J4 ›› 2014, Vol. 36 ›› Issue (08): 1599-1603.

• 论文 • 上一篇    下一篇

基于共同向量的非常态语音说话人识别算法

何俊1,贺前华2,张清华1,孙国玺1,肖明1,左敬龙1   

  1. (1.广东石油化工学院,广东 茂名 525000;2.华南理工大学电子与信息学院,广东 广州 510641)
  • 收稿日期:2012-11-01 修回日期:2013-03-14 出版日期:2014-08-25 发布日期:2014-08-25
  • 基金资助:

    国家自然科学基金资助项目(60972132,61174113,61101160);广东省自然科学基金资助项目(8152500002000011)

Speaker recognition of abnormal
voice based on common vector          

HE Jun1,HE Qianhua2,ZHANG Qinghua1,SUN Guoxi1,XIAO Ming1,ZUO Jinglong1   

  1. (1.Guangdong University of Petrochemical Technology,Maoming 525000;
    2.School of Electronic and Information Engineering,South China University of Technology,Guangzhuo 510641,China)
  • Received:2012-11-01 Revised:2013-03-14 Online:2014-08-25 Published:2014-08-25

摘要:

针对预先给定参数求解共同向量所存在的不足,提出了一种基于共同向量的非常态语音说话人识别算法,首先,通过系统识别率自适应调整求解共同向量的参数;然后,将系统识别率最高的参数视为最优参数,为测试语音提取共同向量,并用SVM分类器进行非常态语音说话人分类。实验结果表明:该算法所提取的共同向量,对轻微感冒语音说话人识别率为85.4%,比对特征不进行处理的GMM算法、SVM和结合共同向量的GMM算法的识别率分别提高了16.9%、15.2%和3.2%。

关键词: 非常态语音, 说话人识别, 共同向量, 支持向量机

Abstract:

A speaker recognition algorithm of abnormal voice based on common vector is proposed to overcome the drawback that exists when the traditional common vector is calculated by using the predefined parameters. The proposed algorithm uses the system recognition ratio to adaptively adjust the parameters of calculating the common vector, takes the parameter with the highest system recognition ratio as the optimal parameter so as to extract the common vector, and uses the SVM Classifier to categorize the speakers of abnormal voice. Experimental results show that, by using the common vector extracted by the proposed algorithm, the speaker recognition ratio of slight cold is 85.4%, which has the improvement of 16.9%,15.2% and 3.2% respectively in comparison to the methods of GMM,SVM and ref[6].

Key words: abnormal voice;speaker recognition;common vector;SVM