pyannote.audio
Neural building blocks for speaker diarization: speech activity, speaker change and overlap detection plus embeddings, in one pipeline.
About
pyannote.audio is the open-source toolkit behind a huge share of the world's speaker diarization: the 'who spoke when' step in meeting transcription, call analytics and subtitle pipelines. It provides neural building blocks for speech activity detection, speaker change detection, overlapped speech detection and speaker embeddings, composed into an end-to-end pipeline that segments an audio file by speaker.
Published at ICASSP 2020 and actively maintained by its creator, the project is MIT licensed with over ten thousand GitHub stars and commits as recent as October 2026, making it the de-facto standard in its niche. Pretrained pipelines are distributed through Hugging Face, though downloading them requires accepting usage terms via a short form. Benchmark claims on AMI, DIHARD and VoxConverse are author-reported, but the project's real credential is its long record of production use.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Papers report strong results on AMI, DIHARD and VoxConverse diarization benchmarks (author-reported)
- ICASSP 2020 publication; de-facto standard toolkit with ~10.6k GitHub stars
- No single headline number independently verified
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 9/10
Scored 9/10 (TIER verified, metrics-v1). Evidence: ICASSP 2020 paper (arXiv 1911.01255); MIT-licensed official repo (pyannote/pyannote-audio, ~10.6k stars, pushed Oct 2026); pretrained pipelines on Hugging Face. Strengths: de-facto standard, permissive license, actively maintained for a decade. Docked for: (1) pretrained pipelines are gated behind a usage-terms form; (2) benchmark claims are author-reported. License: MIT.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 10.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · ICASSP 2020 paper (arXiv 1911.01255); neural building blocks for speaker diarization
- · Official repo pyannote/pyannote-audio, MIT, ~10.6k stars, pushed Oct 2026 — actively maintained
- · Pretrained pipelines on Hugging Face (gated, accept terms); de-facto production standard