基于时频稀疏约束的多通道声源分离算法  

Multi-channel sound source separation based on time-frequency sparsity constraint

作　　者：黄镒东 应忍冬[1] 

机构地区：[1]上海交通大学电子工程系,上海200240

出　　处：《信息技术》2017年第1期72-75,共4页Information Technology

摘　　要：文中讨论基于多通道非负矩阵分解的声源分离算法,通过分解空间相关矩阵实现独立声源信号的分离。现提出了基于时频稀疏的声源分离算法,通过在非负矩阵分解的目标函数中增加分解基的频域稀疏性和分解稀疏的时域稀疏约束,改进了现有多声源分离的效果。实验结果表明算法具有较好的收敛性,同时对于不同的声源分离场景,其分离信扰比均得到了提高。This paper discusses multichannel sound source separation algorithm based on multichannel nonnegative matrix factorization（ NMF）. It proposes a new separation method employing both the frequency sparsity constraint of NMF bases and the sparsity constraint of activation coefficients of NMF bases in factorization of spatial covariance matrix（ SCM）. The experimental results show a good convergence performance as well as improved performance in signal-to-distortion ratio（ SDR） for separation of both musical and speech mixtures compared with methods based on frequency domain ICA and multichannel NMF.

关 键 词：声源分离 非负矩阵分解 时频稀疏约束 

分 类 号：TN912.3[电子电信—通信与信息系统]