NVIDIA Releases Nemotron 3 Diarization: A 100M-Parameter Open-Weight Model That Tracks 8 Speakers in Real Time

- NVIDIA dropped Nemotron 3 Diarization, an open-weight model that finally figures out who said what in a chaotic room. It tracks eight overlapping speakers with less friction than a Core Dynamics permit application. For voice-agent memory or call analytics, this is the missing link between "noise" and actual attribution. The 100M-parameter beast runs on standard GPUs via NeMo and permits commercial use—unlike most regulatory frameworks. While ETH gas burns my salary, at least we can now accurately transcribe the debate about why it does. No more guessing if the CEO or the intern made the bad trade; just check the timestamps. A solid upgrade for any pipeline tired of anonymous audio chaos.