一种新的频域盲语音分离排序算法

A new algorithm for solving the permutation problem in the frequency-domain blind speech source separation

针对频域盲源分离过程中存在的排序模糊性问题,提出了一种新的解决方法。该方法将整个频段分为低频、中频和高频三个部分。在低频段采用比较不同频率点间幅度相关系数大小的排序算法,在中频段采用基于波束形成方位估计的排序算法,在高频段采用比较幅度相关系数大小和波束形成相结合的排序方法。实验中采用评价盲分离算法性能的工具箱BSS_EVAL作为评价标准,仿真结果表明,该排序算法的分离性能大大优于单独采用比较幅度相关系数大小的排序算法和单独基于波束形成方位估计的排序算法。

This paper presents a new method for solving the permutation problem in the frequency-domain BSS. This new method divides the whole frequency-domain into three sections including low-frequency section, mid-frequency section and high-frequency section, and applies different permutation algorithms for different sections. The inter-frequency correlation coefficient of signal amplitudes is used for the low-frequency section, the direction of arrival estimation with beamforming for the mid-frequency section, and the combination of these two methods for the high-frequency section, respectively. BSS_EVAL is used to compare this algorithm＇ s performance with other common permutation methods. Experimental results show that the new method provides a better solution to the permutation problem than the algorithm only with inter-frequency correlation coefficient of signal amplitudes or with direction of arrival estimation for sources.