基于影响因子的频域盲源分离排序算法

摘要

Blind source separation (BSS) for convolutive mixtures can be solved efficiently in the frequency domain ,howev-er ,the permutation problem must be solved in the frequency domain BSS .This paper studies the influence of distance between fre-quency bins and separation quality at each frequency on permutation algorithm based on the amplitude correlation between neighbor bins ,and proposes an improved method for permutation .The influence factor is utilized to control the influence of separation quality and distance between frequency bins on permutation .Lager influence factor is set for frequency bins of short distance and good sep-aration quality ;otherwise a smaller factor is set .Functions for setting influence factor are discussed in detail in this paper .Experi-ments of BSS for instantaneous mixtures ,convolutive mixtures and real-room recorded data are made at last .The results demonstrate the performance of the proposed algorithm is good .%卷积混合信号盲源分离可以在频域得到有效解决，但频域盲源分离必须解决排序问题。本文研究了频点距离和各频点分离质量对基于相邻频点幅度相关性的排序算法的影响，提出了改进的频域盲源分离排序算法。改进算法通过影响因子来控制频点距离和各频点分离质量对排序的影响，距离小且分离质量好的频点设置较大影响因子，距离大或分离质量不好的频点则设置较小影响因子。文中详细讨论了影响因子的设定函数。最后对瞬时混合信号、卷积混合信号、实际房间采集信号分别进行盲源分离实验。实验结果表明了本文算法的有效性。