pyannote.audio
pyannoteAn open-source PyTorch toolkit for speaker diarization: who spoke when. It bundles pretrained segmentation, embedding and clustering pipelines, the open community-1 pipeline among them, and a client for pyannoteAI's paid premium pipeline.
The community-1 pipeline weights on the Hub are CC-BY-4.0 behind a user-conditions gate.
Openness
4 medium confidence- license
- MIT(OSI)
- source
- public
- core features withheld
- yes
pyannote.audio is MIT-licensed, but its most accurate pipeline, precision-2, is a paid pyannoteAI service reached through the same library with an API key. The open community-1 pipeline is complete on its own; the premium tier is withheld from the published source, which makes this open core.
- https://raw.githubusercontent.com/pyannote/pyannote-audio/HEAD/LICENSE recorded 2026-09-27
LICENSE body: "MIT License" / "Copyright (c) 2020 CNRS".
- https://raw.githubusercontent.com/pyannote/pyannote-audio/HEAD/README.md recorded 2026-09-27
README: "built-in support for [pyannoteAI](https://pyannote.ai) premium speaker diarization"; the paid precision-2 pipeline also exists as a self-hosted product the open repository cannot build: "premium pipeline further improves accuracy as well as processing speed (in its self-hosted version)."
Adoption
4 medium confidencePyPI downloads of pyannote.audio, only just above one million a month.
- https://pypistats.org/api/packages/pyannote.audio/recent recorded 2026-09-27
"last_month":1000046 for pyannote.audio.
Capability
3 medium confidenceThe reference open diarization toolkit, but diarization is its single task.
- https://raw.githubusercontent.com/pyannote/pyannote-audio/HEAD/README.md recorded 2026-09-27
"is an open-source toolkit written in Python for speaker diarization"; the Benchmark section reports "[Diarization error rate]".
Verified 2026-09-27