How to use Neural network to distinguish between noise and speech signals?

I want to use the nftool to input a random noise signal, N = wgn(100000,1,0) and I want to add a speech signal to this. From this how can I use the toolbox to recognise speech from noise? I am quite new to the neural network field, any help would be greatly appreciated.

Thanks

Answer 1

[Greg Heath]

You have to train a classifier on a sufficient number of speech and noise examples.

I recommend (1) a literature search (2) a command line (not gui) approach

[Kenny]

How might I be able to train the system using command line approach?

[Greg Heath]

Each of N I-dimensional examples is a column of the input matrix, x. For c classes, the corresponding examples from the target matrix, t, are columns from the c-dimensional unit matrix eye(c);

 [ I N ] = size(x)
 [ O N ] = size(t) % O = c

% Use the classifier function patternnet. First read the documentation and corresponding examples

 help patternnet
 doc patternnet

% Then try one or more of the MATLAB classification datasets

 help nndatasets
 doc  nndatasets

% You can see my posted examples by searching

 greg patternnet

Hope this helps

Thank you for formally accepting my answer

Greg

[Kenny]

Thanks Greg, have you got any examples regarding training of the neural network with noise. What I am aiming to do is to find a pattern between noise and speech, hence if I input a noise speech signal the system will be able to cancel the noise from the trained data and the output will be left with just the speech signal. I am quite new to this, so any literature with simple detailed explanation would be greatly appreciated (for example a flowchart of how the system might work).

[Greg Heath]

Removing noise from noisy speech is a basic signal processing topic that, in general, involves frequency dependent filters which have nothing to do with NNs.

The original post regards identifying different classes of signals. That is the field of classification/pattern-recognition in which NNs are very effective.