pyannoteAI

pyannoteAI builds speaker intelligence tools that identify who speaks when in audio, helping developers create meeting transcripts and customer service analytics. Its Speaker Intelligence Platform labels speakers, detects overlapping speech, combines diarization with transcription, and identifies individuals across sessions using voiceprints. The product family includes the open-source Community-1 model and commercial Precision-2, with API, on-premise, and Argmax-powered on-device deployment options. Streaming diarization adds speaker attribution to live conversations.

Founded in 2024, the company’s co-founders are CEO Vincent Molina, CSO Hervé Bredin, and CTO Juan Coria. Its research roots predate the business: pyannote.audio provides trainable PyTorch components for detecting speech activity, speaker changes, overlapping speech, and speaker embeddings, which developers can combine into diarization pipelines. Its subsequent streaming engineering moved beyond DIART’s batch-derived approach to models processing roughly 100-millisecond audio chunks, with infrastructure designed to preserve speaker identity across distributed inference instances.

In 2026, the company reported an ecosystem of more than 200,000 developers and over 1 billion Hugging Face model downloads. On the financing side, Crane Venture Partners announced a $9 million seed round in 2025, led alongside Serena.