kapynAI / Models

Nvidia drops a free 100M-parameter model that identifies up to eight speakers in real time

Nvidia releases Nemotron 3 Diarization, a 100‑million‑parameter model that detects up to eight speakers in real time. The free model can be integrated into live audio streams, meeting transcription, and accessibility tools, offering developers a lightweight, high‑accuracy speaker‑identification solution. Its modest size and open availability make it attractive for edge deployments and rapid prototyping.

The Decoder·Sep 27, 2026

Opening Kapyn…