Dear sir

Please refer to this link

https://www.google.be/url?sa=t&rct=j&q=&esrc=s&source=web&cd=5&ved=0CFIQFjAE&url=http%3A%2F%2Fweb2py.iiit.ac.in%2Fpublications%2Fdefault%2Fdownload%2Farticle.pdf.b84633fdd6db8a2e.746573742e706466.pdf&ei=Way7U4OpBoSp0AWfwYGwDg&usg=AFQjCNGiO5MixD4F9bMQba7e46yPp6mlrA&sig2=AoUgchkjv3AEgaIjv9NlZA

Dear Djamel

I'm not able to view your response, can u send it again.

Best Regards

Mohammed Nasar.

why dont you try neural network methods

Dear Mohammed,

traditional methods for speech activity detection work reasonably well only in unrealistic conditions in which not too much noise affect the wanted signals. In a more realistic case, the speech is often combined with both stationary and non-stationary noises making traditional solution not adequate. For this reason an optimal solution which is reasonable in all the possible situations does not exist. Personally, I think that modern Deep Neural Networks fed by Mel-Cepstral Coefficients (MFCCs) and pitch features  could work pretty good in general. I suggest you to increase as much as possible both the context window (i.e., the number of frames to feed the neural network) and the material to train the network. In [1]  I have used a deep neural network based solution to perform classification of sounds with interesting and promising results.

Mirco Ravanelli

https://sites.google.com/site/mircoravanelli/

[1] M. Ravanelli, B. Elizalde,  K. Ni, G. Friedland, "Audio Concept Classification with Hierarchical Deep Neural Networks",  in Proceeding of the European Signal Processing Conference, EUSIPCO 2014, Lisbon, Portugal.