speaker classification via neural network

hello,

im doing speaker recognition project. I already processed the input signal coming from my microphone,and done it with 13th coefficient of MFCC(mel-frequency cepstrum coeeficient). For recognition part, i decide to use neural network but i had problems with that part which is:

1)What the input should be in my neural network?

Please help me..

This is the same doubt I am having. How should the training data set be formed from the MFCC matrix. I have w1,w2,w3 as three matrices of order (100*10) where, 10 is the number of MFCC coefficients and 100 is because of the number of frames. How should I form the training data set for nntool box to recognize three speakers.

can u mail me the code u hav prepared for extracting MFCC thnks
mail id pgdutta1990@yahoo.co.in

Simple. Use the neural network toolbox. First you should use a matrice to save the input data.

For eg: If you have 3 classes say A,B and C and if you have 5 samples for every class then you should do the following

Create two matrices input and output.

Each column of the input matrice correspond to a single class.

Let the 5 samples of class A be a1, a2 and similarly for class B and C.

input=[ a1 a2 a3 a4 a5 b1 b2 b3 b4 b5 c1 c2 c3 c4 c5];
The target matice should be as follows

As there are 3 classes we should have 3 rows. For the corresponding input you should define the corresponding class in the target matrix

target=[ 1 0 0 ; 1 0 0 ; 1 0 0 ; 1 0 0 ; 1 0 0; 0 1 0; 0 1 0; and so on]

Transpose the above target matrix so that each column corresponds to a single sample

Here for the sample a1 we have set 1 0 0 which means that it belongs to class A.
Similarly for the sample b1 we have set 0 1 0 which means that it belongs to class B.

Then you open the Pattern Recognition Toolbox of the Matlab. It's easy from there..


Note: Increasing the no of samples of the input always increases the accuracy.


Thanks


---------- Post added at 19:40 ---------- Previous post was at 19:38 ----------

If you aren't good at the basis of the neural networks then I recommend you to go to Matlab Central. There you can learn a lot about neural networks using Matlab.

---------- Post added at 19:41 ---------- Previous post was at 19:40 ----------

fikrul said:

hello,

im doing speaker recognition project. I already processed the input signal coming from my microphone,and done it with 13th coefficient of MFCC(mel-frequency cepstrum coeeficient). For recognition part, i decide to use neural network but i had problems with that part which is:

1)What the input should be in my neural network?

Please help me..

Click to expand...

What are you doing? Speaker recognition or Speech recognition? I have a code for Speaker recognition.

hey, m doing Speaker recognition project plz help me out with the code!!! thnks a lot

Thanks a lot iVenky. I did it. Also, I have my code working. I also simulated it using the MFCC generated through the speech samples obtained through a linear predictive filter and estimated the norm between the two MFCC matrices. I visited your blog. Very nice of it. I follow your blog.I am much interested to get a contact. So, am I much fascinated to link mathematics with Communication. I have a drive towards Signal Processing and Communication systems.

Thank you.