基于局部峰值方差检测的改进DUET算法研究

Improved DUET algorithm based on subarea peak variance detecting

DUET(degenerate unmixing estimation technique)算法是一种典型的盲源分离方法,而混合参数估计是DUET算法的重要组成部分,其估计精度直接影响到源信号的分离效果。针对混合参数估计,提出了一种基于局部峰值方差检测的改进DUET算法。首先利用DUET算法模型中混合参数创建直方图,根据直方图上信号占优的特点,把直方图划分成若干个子区域,计算每个子区域的方差,然后利用排序算法检测出最大的P个方差(P为源信号的个数),这P个方差所在子区域峰值的横纵坐标即是混合参数。此算法改进了原有DUET算法中混合参数估计的智能性和精度,通过语音分离的仿真实验和实录实验表明,此算法简单有效,并且估计源信号的精度有了提高。

The DUET（degenerate unmixing estimation technique）algorithm is a typical blind source separation method,while the estimation of mixing parameters is an important part of DUET algorithm,which has a directly impact on the separation result of source signals.According to the parameter estimation,an improved DUET algorithm based on subarea peak variance detection is proposed in this paper.First,the histogram is created via the mixing parameters,and the histogram is partitioned into some subareas according to the character of sources dominating.Then each subarea＇s variance is calculated,subsequently the maximum P variances are found using the sorting algorithm,in which P is the number of source signals.The P peaks＇ coordinates represent the mixing parameters.The proposed algorithm improves the intelligence and accuracy of the original DUET in mixing parameter estimate.From the experimental results on audio mixtures,the proposed algorithm is simple and highly effective,and the accuracy of the estimated source signals is higher than that of the original DUET algorithm.