Label different speakers

[savchenko]

Might be a stretch, but would it be possible to label different speakers if audio has >1 person talking?

This would come handy for conference recordings with multiple presenters, etc.

[savchenko]

Thinking about possible implementation, the simplest one might be to label based on the audio channel.

Say we have a stereo recording:

1. Recognize L/R separately
2. Label accordingly
3. Done

[ggerganov]

Stereo-diarization is already implemented - see #64 I have some other ideas in mind for general diarization, but low-priority for the moment.