计算机工程与应用 ›› 2011, Vol. 47 ›› Issue (5): 141-145.

• 数据库、信号与信息处理 • 上一篇    下一篇

Symlets小波和子空间联合增强下的语音识别

吴 昊,鲁周迅   

  1. 南京工业大学 电子与信息工程学院,南京 210009
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2011-02-11 发布日期:2011-02-11

Speech recognition based on combined enhancement of Symlets wavelet and subspace

WU Hao,LU Zhouxun   

  1. College of Electronics and Information Engineering,Nanjing University of Technology,Nanjing 210009,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2011-02-11 Published:2011-02-11

摘要: 针对小波阈值选择的多样性,主要研究了小波自适应阈值消噪联合子空间增强对特定人汉语孤立词识别系统的鲁棒性提升。采用Mel倒谱系数,在基于矢量量化(VQ)和高斯混合模型(GMM)的两个系统上,检验采用联合Symlets小波多阈值消噪和子空间增强算法在互为先后顺序作用下系统的识别率,给出一个先Symlets小波阈值消噪再子空间增强的语音增强方法。人耳感官和Matlab实验证实该方法结合了两者的优点,不但平衡了语音失真和噪声抑制,亦可提高VQ系统的顽健性,而对于GMM系统作用有限。

关键词: Symlets小波, 子空间, 语音增强, 语音识别

Abstract: A combined method of Symlets wavelet and subspace is proposed based on experimental robustness comparison in Chinese speaker-dependent isolated word recognition against wavelet threshold diversification.The experiment built on Matlab utilizes VQ(Vector Quantization) and GMM(Gaussian Mixture Models) respectively based on MFCC(Mel Frequency Cepstrum Coefficient).The way Symlets wavelet before subspace performs much better than the other order.Robustness of VQ system is strengthened distinctly after utilizing advantages of both methods.Aesthesis of auditory sense is improved,while the effect on GMM system is not so well as expectation.

Key words: Symlets wavelet, subspace, speech enhancement, speech recognition