稳健的欠定混合矩阵盲辨识

A Robust Underdetermined Mixing Matrix Estimation Algorithm

主要研究欠定盲源分离中的混合矩阵估计问题。提出了一种改进的欠定混合矩阵估计算法,首先采用短时傅里叶变换得到混合信号的时频表示并检测出时频单源点,然后检测出时频单源点中影响混合矩阵估计性能的时频点并将其去除,最后采用聚类的方法实现混合矩阵的估计。语音信号的仿真实验表明,与已有算法相比较,本文提出的混合矩阵估计算法有更高估计精度和更强的鲁棒性。

In this paper, an improved algorithm is proposed for mixing matrix estimation in underdetermined blind source separation （UBSS）. In the algorithm, the single source points in the time-frequency （TF） domain are detected first. Then, those single source points which deteriorate the mixing matrix estimation accuracy are removed. Finally, the mixing matrix is estimated by using a clustering method. It is experimentally shown that the proposed algorithm estimates the mixing matrix with high accuracy and robustness compared with existing algorithms.