希尔伯特黄改进算法在语音增强中的应用  

Speech Enhancement Algorithm Based on Improved Hilbert-Huang Transform

作　　者：李楠[1] 邓舒[1] 

机构地区：[1]东北电力大学信息工程学院,吉林吉林132012

出　　处：《河南科技大学学报（自然科学版）》2013年第5期50-54,63,共6页Journal of Henan University of Science And Technology:Natural Science

基　　金：国家自然科学基金项目(51077010)

摘　　要：语音信号在传输过程中常常受到传输系统或周围环境的噪声影响而无法识别说话者的内容。为了尽可能消除语音中的噪声,提高语音的信噪比及清晰度,提出了一种基于奇异值分解的希尔伯特黄消噪算法。该算法对带噪语音信号进行经验模态分解,得到各阶固有模态函数,在对各阶固有模态函数分量分析的基础上,利用软阈值和奇异值算法相结合的方法来达到语音消噪的目的。这种分阶消噪算法可以尽可能的除去噪声,提取出纯净的语音信号。通过大量的仿真结果可以证明:该算法在语音消噪应用具有可行性,而且优于传统语音消噪方法。Speech signal is often polluted by background noise during transmission,which is difficult to be recognized. For reducing the noise and improving signal-to-noise ratio,a new approach of speech denoising by using the Hilbert-Huang transform and singular value decomposition was proposed. The noise signal was decomposed into IMFs with the method of empirical mode composition. After analyzing feature of each IMF, IMFs were denoised by combining soft-threshold and singular value decomposition. This sub-order denoising algorithm can significantly remove noise. Extensive experiments show this new method can effectively extract pure voice,which is better than the traditional methods.

关 键 词：奇异值分解 希尔伯特黄变换 软阈值 语音消噪 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]