readme

来自「用于盲信号分离的独立分量分析ICA算法」· 代码 · 共 38 行

ica_f.m bsepf2 is the main program and it needs 

correlation.m
decorrelation.m
permutation.m
sepfilter.m.

You also need to visit

ftp://sig.enst.fr/pub/jfc/Algo/Joint_Diag/

and get a matlab code "joint_diag.m". this is by Jean-Fran\c{c}ois
Cardoso. You have to put the code under this directory "ica".

usage is 

function [Y1,Y2] = ica_f(X,NFFT,FS,OVERLAP,N)

where 

X       : is a ".wav" data (You should use 16kHz sampling rate). 
NFFT    : number of the FFT points.
FS      : sampling rate (it should be 16000 Again..)
OVERLAP : overlap of the window function, if sampling rate 16kHz, it
	should be NFFT-20.
N	: number of the matrixes to be simultaneous diagonalized. 40
	is my recommendation.

I use this ica_f in the following way. 

 >> X=wavread('***.wav')

 >> [Y1,Y2] = ica_f(X,512,16000,492,40); 

X_linear.wav, X_room.wav are for test use. X_linear is an instantaneous
mixture, and X_room is a convolutive mixture. 

Shiro Ikeda, [e-mail]