Add support for WebRTC AEC3

[sauwming]

The current WebRTC source included in PJSIP is already quite old (from Oct 2015) and only supports a clock rate of 16 KHz (see #2404). So in this PR, we add the latest WebRTC AEC3 into PJSIP.

Requirement: C++17
Due to this requirement, and the complexity needed to build the AEC3 component (in my local Mac machine, PJSIP build time nearly doubles from ~35 sec to ~65 sec if AEC3 is included), by default AEC3 will be disabled and the old WebRTC AEC is still made available.

Licenses: check third_party/webrtc_aec3/PJSIP_NOTES about the licenses of WebRTC itself and external third party components required, such as abseil, rnnoise, and pffft.

Features:

• AEC3
• Noise Suppressor
  Pass the option PJMEDIA_ECHO_USE_NOISE_SUPPRESSOR to enable it.
• AGC2
  The next generation AGC functionality. AGC2 aims to automatically adjust levels by acting on the microphone gain and/or applying digital gain.
  Pass the option PJMEDIA_ECHO_USE_GAIN_CONTROLLER to enable it.

To support WebRTC AEC3, enable it when configuring PJSIP, i.e.:
./configure --enable-libwebrtc-aec3
And check if it can be built:
checking if WebRtc AEC3 can be compiled with C++17... yes

To use it, specify PJMEDIA_ECHO_WEBRTC_AEC3 in pjsua_media_config.ec_options or MediaConfig::ecOptions.
You can test with our sample app pjsua, by pass the number 4 as the EC option desired, i.e..:
./pjsua --ec-opt=4
You can check the EC statistics using the API pjsua_get_ec_stat() or from pjsua_call_dump(). It will output information like this:
EC stat: WebRTC AEC3 metrics: delay=12 ms, return loss=-7.30, return loss enh=1.09

[ismangil]

Also update detailed Licensing wiki

[nanangizz]

As briefly discussed before, there seem to be no microphone gain adjustment? So perhaps we should remove or add some notes about "acting on the microphone gain" in the PR desc?

[sauwming]

As briefly discussed before, there seem to be no microphone gain adjustment? So perhaps we should remove or add some notes about "acting on the microphone gain" in the PR desc?

It's the original wording in their documentation, taken from:
https://github.com/pjsip/pjproject/blob/webrtc-aec3/third_party/webrtc_aec3/src/modules/audio_processing/gain_controller2.h#L28

And the dictionary says that "acting on" means "to use (something, such as a feeling or suggestion) as a reason or basis for doing something", so what the doc says is that it uses microphone gain information and then apply digital gain based on that info.

[maneeshj]

I did the above steps to build AEC3, but my logs output AEC3 stats as all 0 numbers. I am not sure if AEC is actually kicked in or not. Does O in return loss, delay indicate it is not being used ?

[monocyte]

Having the same issue: followed build instructions on macOS and Windows, but every call stats show zeroes in AEC3 stats on both OS:
EC stat: WebRTC AEC3 metrics: delay=0 ms, return loss=0.00, return loss enh=0.00
I used opus codec and ec_options == PJMEDIA_ECHO_WEBRTC_AEC3 | PJMEDIA_ECHO_USE_NOISE_SUPPRESSOR.

After some research I discovered that echo cancellation is working (at least it's created and called in echo_webrtc_aec3.cpp), but stats (using pjsua_call_dump) are printed before webrtc_aec3_cancel_echo is called with echo->get_metrics == PJ_TRUE, so they are empty.

[fbreton06]

Hi guys, sorry it's probably not the better location to post that.
I've write a python program to establish a SIP audio communication on my Raspberry4.
It's based on pjsua2 module from PJSIP V1.13.1 release.
I'm trying to suppress echo with activating the PJMEDIA_ECHO_WEBRTC_AEC3, but result is disappointing.
Has anyone advices to improve that?
cheers,