一种改进的FastICA算法在语音信号盲源分离中的应用  

An improved FastICA algorithm for blind source separation of speech signals

作　　者：朱立娟 赵风海[1] ZHU Lijuan;ZHAO Fenghai(College of Electronic Information and Optical Engineering,Nankai University,Tianjin 300350,China)

机构地区：[1]南开大学电子信息与光学工程学院

出　　处：《电声技术》2019年第5期64-68,共5页Audio Engineering

摘　　要：为了克服快速不动点算法(FastICA)在语音信号盲源分离中由于计算量较大时造成的运算速率明显降低的弊端,提出了一种矩阵联合对角化和FastICA算法相结合的改进算法。首先对观测信号进行特征矩阵近似联合对角化处理,得到初步分离的信号,进而利用FastICA算法实现语音信号的二次分离。仿真结果表明,和传统的基于负熵极大化的FastICA算法相比,改进的FastICA算法能够在保证分离效果的前提下,大幅度降低了运算的迭代次数,进一步加快了运算的收敛速度。In order to overcome the disadvantage of fast fixed point algorithm(FastICA)in blind source separation(BSS)of speech signals,an improved algorithm combining matrix joint diagonalization and FastICA algorithm is proposed.Firstly,the observed signal is approximated and diagonalized by eigenvalue matrix,and the preliminary separated signal is obtained.Then,FastICA algorithm is used to realize the second separation of speech signals.The simulation results show that,compared with the traditional FastICA algorithm based on negative entropy maximization,the improved FastICA algorithm can reduce the number of iterations and further accelerate the convergence speed of the operation while guaranteeing the separation effect.

关 键 词：盲源分离 联合对角化 FASTICA 负熵极大化 

分 类 号：TN912.35[电子电信—通信与信息系统]