It only works for the calls where audio of each speaker is already recorder into a separate track.

It uses faster-whisper and Silero VAD under the hood.