Speaker diarization / who-said-what transcription. Whisper v3 + speaker labels. Returns utterances grouped by speaker, plus per-speaker stats (count, seconds, words). 60 min max.