using this sound file: http://www.ism.ac.jp/~shiro/research/sounds/RSM/X_rsm2.wav
I'm trying to recreate Andrew Ng's Machine Learning presentation(https://class.coursera.org/ml-005/lecture) from coursera in matlab
What I do is to read a .wav file (16khz, 7 sec, 2 channels)
[x,xfs] = wavread('track.wav')
Now I transpose x
x = x'
Now I proceed to use x on the cocktail party algorithm
[W,s,v] = svd((repmat(sum(x.*x,1),size(x,1),1).*x)*x')
MATLAB returns:
W =
-0.9233 -0.3841
-0.3841 0.9233
s =
265.4832 0
0 13.0768
v =
-0.9233 -0.3841
-0.3841 0.9233
Where is the separated audio?
EDIT: From further research, I found out that W is only the unmixing matrix. Meaning this algo is incomplete if my goal is to get the two output separated sound sources. What do I do with this unmixing matrix?