基于时域去相关预处理的卷积混合语音盲分离算法  

Blind Convolution Mixed Speech Separation Algorithm Based on Preprocessing of Decorrelation in Time Domain

作　　者：夏秀渝[1] 何培宇[1] 周激流[1] 

机构地区：[1]四川大学电子信息学院,成都610064

出　　处：《数据采集与处理》2009年第2期140-143,共4页Journal of Data Acquisition and Processing

基　　金：国家自然科学基金(60472096)资助项目

摘　　要：因语音信号自身的相关性及不平稳性,使卷积混合语音信号的盲分离变得十分困难。本文提出了一种基于时域去相关预处理的卷积混合语音盲分离时域算法。该算法采用自适应格型预测误差滤波器对语音信号进行时域去相关处理,同时基于空域去相关算法完成卷积混合语音分离。该算法充分考虑了语音信号自身的相关性及不平稳性,具有计算量小、收敛速度快、稳定性好的优点。仿真实验验证了该算法在对卷积混合语音信号进行盲分离时的有效性。It is difficult to separate convolution mixed speech signals due to the auto-correlativity and the non-stationarity of speech signals. A blind convolution mixed speech separation algorithm is presented based on the decorreiation in time domain. The correlativity of speech signa s in time domain is reduced by using an adaptive lattice linear prediction filter, and the convolution mixed speech signals are separated based on the second order statistics. Because the correlativity and the nonstationarity of speech signals are fully considered, the algorithm has advantages of the lower computational complexity, the fast convergence and the stability. Experimental results show that the algorithm is effective in separating convolution mixed audio signals.

关 键 词：语音分离 卷积混合信号 去相关 格型滤波器 

分 类 号：TN912.3[电子电信—通信与信息系统]