混合语音信号的盲分离  

Blind Source Separation for Mixed Speech Signal

作　　者：张倩蓉[1] 王新新[1] 

机构地区：[1]兰州交通大学电子与信息工程学院,甘肃兰州730070

出　　处：《山西电子技术》2008年第1期16-17,共2页Shanxi Electronic Technology

摘　　要：重点研究了卷积混合语音信号的盲分离方法。语音信号是非平稳信号,但是它在短时间上具有平稳性。因此,本文对语音信号进行加窗傅立叶变换(FFT)将卷积混合问题转换为频域上每个频点的瞬时性BSS(blind source separation)问题,采用定点(fixed-point)ICA(independent component analysis)算法对混合语音信号进行了分离,并用matlab进行了仿真。This paper focuses on the blind processing methods for the convolution of mixed speech signals. Speech signals are intrinsically non-stationary but quasi-stationary in a short time. This paper applies the windowed Fourier transform to convert the mixed problem of convolution into instantaneous BSS problem on every frequency component in frequency domain. It separates the audio signals by fixed-point ICA and simulates it with rnatlab.

关 键 词：盲分离 语音信号 ICA 

分 类 号：TN912.3[电子电信—通信与信息系统]