Diarize a recording first and then match each voice against a roster of known speakers to find who spoke when. Recordings and voiceprints are never sent to the 3rd party cloud.
On-device speaker identification across meetings combines Falcon Speaker Diarization and Eagle Speaker Recognition. Speaker diarization and speaker recognition solve different halves of the same problem. Falcon segments a recording into speaker turns without knowing who anyone is. Eagle scores each turn against stored voiceprints to attach a name. Running them in sequence turns anonymous recordings into named and timestamped transcripts.
Falcon Speaker Diarization takes a meeting recording and returns timestamped segments, each tagged with an anonymous speaker label, answering who spoke when before any name is attached. It is decoupled from transcription, so it slots in alongside any speech-to-text engine you already run, and it does the work on CPU without requiring GPU or cloud processing.
Eagle Speaker Recognition enrolls a speaker from a few seconds of natural speech and exports a voiceprint to a file, with no language or passphrase restriction. Load those voiceprints into later runs and the same people are recognized across meetings, calls, and interviews. Speakers not yet in the roster can be profiled and saved, so the next recording recognizes them automatically. Everything runs on-device, so Picovoice never accesses end-user audio.
Automatically identify and label every speaker in recorded standups, interviews, and board meetings. On-device speaker identification keeps audio private while improving meeting transcripts, notes, action items, and AI-generated recaps.
Use speaker identification to label attorneys, witnesses, investigators, and interview subjects by name across recordings. Create searchable legal transcripts, improve attribution accuracy, and reduce manual review time.
Track who said what across board meetings, executive reviews, and governance sessions. Named speaker attribution supports accurate meeting minutes, decision tracking, compliance workflows, and audit-ready records.
Identify agents and recurring callers across support conversations using on-device speaker recognition. Improve call center transcription, quality assurance, customer history tracking, and support analytics while keeping audio private.
A complete working recipe in Python. Open-source on GitHub. Runs 100% on-device.