I am sameer, new member of this excellent group. Its very nice to see that student like me are getting very valuable advices from the experianced techonocrats. I am doing my masters digree in communications and doing my final year project in " Echo cancellation in Packet telephony".
From the theroy and ieee papers I came to know that apart from the adaptive filter a block called " Double Talk Detector" is very important in echo cancelllation. When both near end and far end signals are present the adaptive filter must be freezed to avoid the divergence of the filter taps. There are 'geigle type and cross correlation type methoda are there for this. For this purpose( ie: to know the presence of double talk) I am using the cross correlation method. In this method a threshold " T ", we have to setup, if the cross correlation coefficents are less than the " T ", we can declare that double talk is happening. But the problem is, here I am setting the threshold " T " by observing the speech wave form, which I am using for the matlab simulations.
And the threshold set for one speech signal is not matching when I am using differnet signal ( recorded in diffent network situations). So is there any concept, which can automatically set the threshold by its own ??(ie: it must be adaptive i think). I request to all group members to help me in this. Please write me if any such things is there?
waiting for your valuable reply,