稀疏分量分析在欠定语音信号盲分离中的应用  

Sparse Component Analysis and Application for Underdetermined Blind Source Separation of Speech Signals

作　　者：赵卫杰[1,2] 任明荣[1,2] 张亚庭[1,2] 

机构地区：[1]北京工业大学电子信息与控制工程学院,北京100124 [2]国家教育部数字社区工程研究中心,北京100124

出　　处：《电声技术》2010年第3期46-49,63,共5页Audio Engineering

摘　　要：研究了基于两步法的欠定语音信号盲分离。针对混合信号散点图在原点中心混叠程度过高的缺点,提出了弭灭圆K均值聚类算法,提高了混叠矩阵的估计精度。结合时频分析算法实现了欠定瞬时线性混叠语音信号的盲分离,取得了较好的分离效果。Sparse component analysis is a signal processing method based on sparse representation. In order to overcome the shortcoming that the aliasing level of the mixed-signal scatter at the origin center is too high, based on sparse component analysis method, a new K means clustering algorithm which can estimate mixing matrix more effectively is proposed. Combined with time-frequency analysis, the proposed algorithm can achieve the instantaneous linear aliasing blind separation of speech signals and get a good separation effect.

关 键 词：稀疏分量分析 K均值聚类 欠定 盲分离 

分 类 号：TN912[电子电信—通信与信息系统]