[feat] Audiofeedback prevention

[theunafraid]

The purpose of this PR is to make use of a RNN to prevent the audiofeedback loop and be able to act on it BEFORE "howl" without affecting the user experience. More info here https://github.com/theunafraid/audiofeedback-prevention .

This needs much more work and did not have time for it, model must be trained using much more data and did not have time to find or sort it(used TCD VoIP Dataset and MLCommons . At the time i began, did not find anything similar(trained model) that works in a similar way to this(classifier).

https://github.com/jitsi/lib-jitsi-meet/pull/1694

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[sampaccoud]

@theunafraid Hello. We are interested by this feature and could decide to work on it in the short term. Could you tell us a little bit more about the todo and whether you still think the method is relevant?

[theunafraid]

@theunafraid Hello. We are interested by this feature and could decide to work on it in the short term. Could you tell us a little bit more about the todo and whether you still think the method is relevant?

Hi @sampaccoud method is not relevant, if you have a better way to prevent the feedback from happening(by lowering the volume for example, or just a notification to the user) or detect it right before the howl happens, then please share. The Jitsi Team might still be interested in this feature. Keep in mind that the solution has to be lightweight for web... I currently do not have time to resume work on it, but will maybe make time for this in the not so near future.

[sampaccoud]

@theunafraid thank you for your feedback. Will keep you posted if we make progress :pray:

[github-actions[bot]]

This PR has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.