计算机工程与应用 ›› 2009, Vol. 45 ›› Issue (34): 118-120.DOI: 10.3778/j.issn.1002-8331.2009.34.036

• 数据库、信号与信息处理 • 上一篇    下一篇

基于分布特征统计的说话人识别

李邵梅,郭云飞,卫红权   

  1. 国家数字交换系统工程技术研究中心,郑州 450002
  • 收稿日期:2009-06-12 修回日期:2009-07-13 出版日期:2009-12-01 发布日期:2009-12-01
  • 通讯作者: 李邵梅

Speaker recognition via statistics of distribution feature

LI Shao-mei,GUO Yun-fei,WEI Hong-quan   

  1. National Digital Switching System Research Center,Zhengzhou 450002,China
  • Received:2009-06-12 Revised:2009-07-13 Online:2009-12-01 Published:2009-12-01
  • Contact: LI Shao-mei

摘要: 给出了基于公共码书的说话人分布特征的定义。提出了基于分布特征统计的说话人识别算法,根据所有参考说话人的训练语音建立公共码书,实现对语音特征空间的分类,统计各参考说话人训练语音的在公共码字上的分布特征进行建模。识别中引入双序列比对方法进行识别语音的分布特征统计与参考说话人模型间的相似度匹配,实现对说话人的辨认。实验表明,该方法保证识别率的情况下,进一步提高了基于VQ的说话人识别的速度。

关键词: 说话人识别, 矢量量化技术(VQ), 分布特征, 公共码书, 双序列比对

Abstract: The concept of speaker distribution feature around common codebook is defined,and a speaker recognition algorithm is proposed based on it.A common codebook is generated via the training data from all the reference speakers,which is used to classify speech feature space,and the model of each reference speaker is derived by the statistics of speaker’s distribution feature measured by the common code book.In the recognition,pairwise sequence alignment is introduced to measure the distortion between the test speech distribution feature sequence and each reference model,and speaker recognition is realized by distortion compare.Experimental results show that it can save calculation and storage resource while shows better performance over VQ.

Key words: speaker recognition, Vector Quantization(VQ), distribution feature, common codebook, pairwise sequence alignment

中图分类号: