No credit card required
Accuracy for official records
Precision-3 offer the lowest diarization error rate in all ten domains of our public benchmark, including the interruptions, crosstalk, and three-second interventions typical of council chambers.
Human review built in
Confidence scores on a 0 to 100 scale, per speaker turn and per voiceprint match. Set a threshold and staff review only the segments the model flags, instead of re-listening to the whole session.
Sovereign by deployment
Run the models on-premise or in your own environment, where pyannoteAI receives no audio at all. Customer audio is never used to train models, on any tier.
Fits your pipeline, not the other way round
Pair diarization with the STT model that fits your language through STT Orchestration. Document formats, review steps, and archival rules stay yours to define.
Use cases
Council and assembly minutes: Speaker-attributed drafts of multi-hour sessions, with members recognized across sessions through voiceprints
Citizen service lines: Agent and caller separation on varied phone systems, background noise, and diverse accents
Hearings and administrative proceedings: Timestamped, attributed records of interviews, consultations, and internal proceedings
Justice and public safety: Attributed transcripts for law enforcement interviews and judicial proceedings, processed on-premise
Cross-government meetings: Transcription and speaker separation inside sovereign videoconferencing platforms
Features
Streaming Diarization
Who spoke, when, in order, across sessions of any length
Speaker identification with voiceprints
Recognize known participants by name, from a clean sample of up to 30 seconds
Confidence scores
Per-turn and per-match reliability, the basis of human-in-the-loop review
Overlapping speech detection
Explicit handling of crosstalk and interruptions
STT Orchestration
Diarization reconciled with the STT model you choose into one attributed output
Writes formal minutes for town halls and councils in Germany, Austria, and Switzerland, on-premise
The French state's sovereign videoconferencing platform, uses pyannoteAI for transcription and speaker separation
Prรฉsidence de la Rรฉpublique
and other French public institutions work with pyannoteAI models
Justus Feron
Co-founder at SpeechMind
benchmark domains with the lowest DER
received by pyannoteAI in on-premise deployments
of publicly funded research