Hi Im buildind a speech recognitions system for isolated words for
independebt speaker.

I have a data base of 30 wave files for each word

1 I extract de MFCC of each word and I get a matrix of 171x12
2 I put the matrix in 1 columm
3 I do it all again for eache wave sample
4 i build de RBFNN

Code

file is the name of eache of the wave files, consider it just for info

For i=1:30

[s,fs]=wavread(file);
X=mfcc(s,fs);%extarct de features 171x12 matrix
X=X(:); % all in 1 columm
train=[train X] % all together in 1 matrix
target=[target 1] % target matrix

End

net=newrb(train, target);%build the network

to test the network

[s,fs]=wavread(file);
X=mfcc(s,fs);%extarct de features 171x12 matrix
X=X(:); % all in 1 columm
output=sim(net,X);

The rsults I get are pretty bad!

Can anyone help me?
Each word must have a RBFNN model right?

How can I get the correct word?

Thank you in advance

Pedro Start a New Thread