--- In , "andrew_xiang" <ahgu@a...> wrote: > Anyone implemented a system that get you 30dB of reduction for AEC? > > In my system, I only get 4X for voice signal. I will give it a try > with white noise. But seems 30X is far away. But I don't expect users > to utter white noise when they speak. > > My calculation: > > 4X=SUM(abs(microphone))/SUM(abs(error)) > > 4X leaves me a lot of residual. > > Are there other tricks in increasing the reduction? > > thanks > Andrew It depends on the excitation signal (ie how white it is), also what adaptation algorithm are you using? How long is the echo tail of your system? I have seen AECs in practical systems go up to 20dB for 128msec tail length, but not 30dB. We were using NLMS and the excitation signal was speech. Hope the benchmark helps. |