What is Diarisation? Automatic speaker segmentation and identification. It splits audio into Speaker-00, Speaker-01, etc., based on voice characteristics.
How to Get Diarisation:
pannous/PyAnnote for speaker diarisation https://huggingface.co/pyannote/speaker-diarization-3.0diarisation-tool (CNNDiarization, pyannote.audio-silero)AssemblyAI Async API, Deepgram Live APIvits-gnu + whisperx
Hugging Face: Search "diarisation" or "speaker embeddings" — models like pyannote/embedding are lightweight and can process speaker segments.
Starting a transcription downloads the selected Whisper, diarization and alignment models from Hugging Face on first use — roughly 140 MB–3 GB depending on model size. Later runs reuse the cached copy.
| File | Status | Progress | Started | |
|---|---|---|---|---|
| No active jobs yet. | ||||
| File | Status | Created | Interviewee | |
|---|---|---|---|---|
| Loading… | ||||
No transcript selected.
Choose a completed job from the table above.
No enrolled speakers.
Note: BYOM connects to any OpenAI-compatible server (Ollama, LM Studio, vLLM, OpenAI API) for post-transcription tasks like summarization or cleanup. It does not affect core transcription or diarization.