Voice Focus Leak : WebRTC audio stream not released

What happened and what did you expect to happen?

After intializing Voice Focus, the webRTC audio stream is never released.

I’m creating a VoiceFocusTransformDevice through the VoiceFocusDeviceTransformer and using the default voicefocus-inline-processor. When Stopping the device through the stop() method and ending the meeting, we can still see active webRTC audio streams through chrome debugger (chrome://webrtc-internals).

I noticed that when stopping the worklet-inline-node we send a ‘stop’ message through the port, yet we dont stop the worker that seems to stay alive. In the meetingV2 demo app, you dont get the issue because you seem to redirect to another page straight after ending the meeting.

Have you reviewed our existing documentation?

• GitHub issues
• FAQs page
• README page
• Feature-specific guides

Reproduction steps

Supposing you have a started audioVideo meeting :

• Create a voice focus transformer with VoiceFocusDeviceTransformer.create. It will create in turn a VoiceFocus and its associated worker. How do you I destroy it even without creating a device ?

• Create a voice focus device with .createTransformDevice(innerDevice) and startAudioInput on deviceController of the meetingSession’s audioVideo

• Call stop on the device, when stopping the audioVideo and destroying the deviceController

You still have running webRTC audio streams.

Amazon Chime SDK for JavaScript version

3.2.0

What browsers are you seeing the problem on?

Mac OS Chrome

Browser version

Version 102.0.5005.61 (Official Build) (x86_64)

Meeting and Attendee ID Information.

No response

Browser console logs

Remaining webRTC audio streams :

[richhx]

Hey @chdeps, thanks for raising this issue. After giving it a look, the worker is definitely not being released. I’ve submitted a PR at https://github.com/aws/amazon-chime-sdk-js/pull/2320 to fix this issue.

However, I believe the worker leak is unrelated to the WebRTC stream leak. This can be tested out through the following steps in the demo application (on the release branch).

1. Deploy the demo application on the release branch (or main, though the additional options wasn’t working for me when I tried)
2. Open chrome://webrtc-internals/ in a tab.
3. Open the demo site in a new tab
4. Enter stuff into “Meeting Title” and “Your Name”
5. Make sure “Web Audio” is unselected
6. Click “Continue”
7. Click “Join” (notice that there’s no “Enable Voice Focus” option since Web Audio is disabled)
8. Hold the “Shift” key and click the “Leave meeting” green icon in the top-right. The “Shift” key is important here because it will not redirect the page, thereby allowing us to debug the contents of the page.
9. Notice that it says “Gone” on the page.
10. Visit the chrome://webrtc-internals/ tab that was previously opened.
11. Notice that even though the “Connection state” is “closed”, ICE connection state is still “connected”, Signaling state is “stable”, and WebRTC audio streams and audio sources are still running

Logs from my testing: https://pastebin.com/1WaxaC8m

Screenshot from my testing:

Based on this testing, my guess is it’s somewhere in the controller and not the Voice Focus library, since the Voice Focus code isn’t even executed in this test path and the issue is still reproduced.

@michhyun1 or @devalevenkatesh are you able to look into this further?

[chdeps]

@nainkunal933 Thanks for investigating! I’ll follow up on Chromium issue 😃