基于后级滤波处理的频域语音盲信号分离  

Blind speech separation in frequency domain with post-processing

作　　者：潘帆[1] 何培宇[1] 吴景田[1] 邓方[1] 徐自励[1] 

机构地区：[1]四川大学电子信息学院,成都610064

出　　处：《四川大学学报（自然科学版）》2009年第2期413-416,共4页Journal of Sichuan University(Natural Science Edition)

基　　金：国家自然科学基金(60472096)

摘　　要：针对现有盲信号分离算法在真实环境下性能会降低的局限性,提出了一种将频域盲信号分离算法与后级滤波处理相结合的方法。在后级滤波处理阶段用自适应对消算法来去除分离信号中的残留干扰分量,根据语音信号的能量随频点分布起伏较大的特点,后级滤波算法引入了变步长的频域泄漏LMS(leaky LMs)算法。仿真实验表明,作者提出的方法使得分离效果有明显改善,而且比已提出的后级滤波处理方法效果要好。To solve the problem of speech blind separation in real-world, we propose a method of blind signal separation in frequency domain combining with post-processing. In the post-processing, we use adaptive cancellation algorithm to cancel the residual crosstalk component in separated signal. We introduce a modified leaky LMS algorithm in post-processing based on that the speech signal exhibits large power fluctuations at different frequency bins. The simulation experimental results show that the proposed method improves the separation performance obviously, and it is superior to a similar method with post-processing.

关 键 词：盲信号分离 自适应对消 leaky LMS 

分 类 号：TN912[电子电信—通信与信息系统]