I've seen this issue on github:
This is really exciting! Basically, it tells that active speaker detection and last-N features are now supported by Jitsi.
Could you elaborate a bit about the status and plans for it?
E.g. I have the following questions:
1) Is it only supported at the Colibri protocol level, i.e. new messages, etc? Or have you already implemented the server-side logic for really analyzing audio streams and filtering out the last-N streams?
2) If you have already implemented active speaker detection, which audio formats are supported?
3) How Jitsi signals to the client that certain streams are not sending at the moment?
4) How Jitsi signals to the client which participants are the last-N ones?
5) Have you considered the API/Colibri protocol extension which would allow a participant to explicitly pick which streams (i.e. from which participants) it would like to receive? What do you think of having such a feature? It would allow for scenarios where you want to have explicit floor control, etc...