一种基于时域解相关的双通道语音分离算法

A co-channel speech separation arithmetic based on decorrelation in time domain

提出了一种基于时域解相关的卷积混合语音盲分离时域算法。该算法利用相关系数对语音信号进行时域去相关处理,同时基于二阶统计量完成双通道语音分离。该算法充分考虑了语音信号本身的相关性及不平稳性,收敛速度快,稳定性好,为减小计算量提出了该算法的递推改进算法。仿真实验表明,在对卷积混合双通道语音信号进行盲分离时,该算法是非常有效的。

A blind speech separation arithmetic based on decorrelation in the time domain for convolved mixed audio signals is presented in this paper. The arithmetic wipes off audio signal＇s relativity in time domain by correlation coefficient, in the same time it separates co-channel mixed audio signals based on second order statistics. For thought over audio signal＇s relativity and unsteadiness, the arithmetic have virtues of fast convergence and good stability. In order to deduce calculate quantity, an improving arithmetic is put forward. Experimental results show that the proposed method realizes fast and robust separation for co-channel convoluted mixed audio signals.