Re: webRTC echo canceller calibration

[jequalizer]

I tried both.

The echo tail length is defined in aec_core.h by NR_PART*64

The aec seems to work fine when echo is not too loud or the true tail is too long, i.e. conference room aec.

The complexity is not low.

The suppression level of nlp can be set by a parameter among conservative, moderate and aggressive.

aecm seems to be a embedded version.

The complexity it low, and works fine when echo is low.
Considerably distort near end signal in moderate or strong double-talk.

在 2012年9月10日星期一UTC+8下午10时14分15秒，john_conor写道：

Hi,

 

I am trying to benchmark the AEC performance in webRTC and have built a standalone Visual studio project encompassing the audio_processing module. With standard test vectors, I find that the AEC is essentially suppressing the echo and adding confort noise. The performance is not great mainly due to level fluctuations when the coupled echo is strong.

 

Is there a way to control say the tail length of the canceller or the suppression levels of the nlp?

 

Also, has anybody tried out the 2nd canceller included in webRTC Echo control mobile(aecm)?

 

Regards,

john