Choose model, diarization mode, upload voice print and full audio, and label the known speaker. Target speaker → chosen model; other speakers → Whisper Large v3.