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

Diarization is the job of working out who spoke when, and it is what lets a transcript label voices instead of blending them. Nvidia's Nemotron 3 Diarization is small enough to run in real time, separates up to eight speakers, and flags moments when people talk over one another. Nvidia is giving the weights away, and it says the model leads a strict public benchmark.

Summary written with AI by HowTechMoves, based on reporting by The Decoder.