pyannoteAI is a speech AI platform for speaker diarization, the task of finding who spoke when in a recording. The company grew out of pyannote.audio, an open-source toolkit with more than 1 billion downloads and more than 10 years of academic research; its co-founders are Hervé Bredin, Vincent Molina and Juan Coria.
A developer sends a recorded audio file or a live audio stream to the API and receives time segments labeled by speaker. Speaker identification matches a voice to a known person through a stored voiceprint, across files and sessions. Speaker separation splits overlapping voices into clean individual audio streams. STT orchestration combines the diarization with the output of a speech-to-text (STT) model in one call, which gives a transcript with each line assigned to a speaker. pyannoteAI states up to 52% lower error than other diarization services. The models run in its cloud, on premises or at the edge.
The open-source library is free. The hosted API is a monthly subscription that includes usage credit, spent at a rate per hour of processed audio; Enterprise customers get volume pricing and on-premise deployment.