计算机工程 ›› 2008, Vol. 34 ›› Issue (14): 14-16.doi: 10.3969/j.issn.1000-3428.2008.14.006

• 博士论文 • 上一篇    下一篇

基于背景噪声估计的说话人识别算法

钱 博,唐振民,李燕萍,徐利敏   

  1. (南京理工大学计算机学院模式识别与人工智能实验室,南京 210094)
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2008-07-20 发布日期:2008-07-20

Speaker Recognition Algorithm Based onEstimation of Background Noise

QIAN Bo, TANG Zhen-min, LI Yan-ping, XU Li-min   

  1. (Lab of Pattern Recognition and Intelligent System, Department of Computer, Nanjing University of Science and Technology, Nanjing 210094)
  • Received:1900-01-01 Revised:1900-01-01 Online:2008-07-20 Published:2008-07-20

摘要: 基于同段语音中有声和无声部分具有相同的背景噪声特征的假设,该文提出基于背景噪声模型估计的说话人识别算法。该算法从无声信号中提取背景噪声的HMM参数,并结合含噪音频特征估计说话人特征参数,进行识别处理。实验表明,该方法具有比传统去噪技术更好的识别能力,混合噪声下的说话人识别正确率达到了90%以上。

关键词: 说话人识别, 背景噪声估计, 高斯混合模型, 自适应

Abstract: This paper proposes a new method by modeling the background noise to statistically estimate the pure speaker information. The substantial similarity of noise exists between the voiced and unvoiced regions in the same audio data. Experimental result shows that, the system can adapt the background automatically and acquire better accurate rate, and the robustness can be improved with the method. The speaker recognition accurate rate is higher than 90% in mixture background noise environment.

Key words: speaker recognition, estimation of background noise, Gaussian mixed model, self-adaptive

中图分类号: