---
title: SmartFile Speaker Diarization
emoji: 🎙️
colorFrom: indigo
colorTo: blue
sdk: gradio
sdk_version: 5.9.1
python_version: "3.11"
app_file: app.py
pinned: false
---

# SmartFile Speaker Diarization

Detects **who speaks when** using `pyannote/speaker-diarization-3.1`. SmartFile
calls this before transcription so each speaker turn can be transcribed and
labeled (Locuteur 1, Locuteur 2, …).

Send base64 audio, get JSON: `{"segments":[{"speaker","start","end"}...], "num_speakers"}`.

**Gated models** — accept terms for BOTH on the same account as HF_TOKEN:
- pyannote/speaker-diarization-3.1
- pyannote/segmentation-3.0

Pinned to Python 3.11 + torch/torchaudio 2.1.2 (last version with
`torchaudio.AudioMetaData`, which pyannote 3.3.2 requires). Heavier than the STT
Space — use a paid CPU/GPU tier for usable speed.