Sakhanda Wire
NVDA $154.51 -1.04% MSFT $506.49 -2.22% GOOGL $281.27 -0.86% META $621.51 -0.78% AMZN $256.90 -0.33%
← Back to the news

NVIDIA unveils eight-speaker Nemotron 3 Diarization model

NVIDIA has introduced Nemotron 3 Diarization, an open-weight model with 100 million parameters designed to identify when different speakers are active in live or recorded conversations. It supports up to eight speaker channels, including simultaneous speech, and can process audio in chunks with configurable streaming latency. The model produces anonymous speaker labels and timestamps rather than identifying people by name.

Nemotron 3 Diarization uses arrival-time ordering to keep speaker channels consistent between streaming audio chunks. Its output can be combined with automatic speech recognition to create transcripts that associate words with speakers; diarization alone does not generate text. The system accepts 16 kHz, single-channel audio and offers recommended input-buffer latencies of 30.4, 1.04, 0.64 and 0.32 seconds.

In VoiceArena’s initial Diarization-Bench evaluation, the model ranked first among 12 systems and 17 configurations tested on 139 English conversations. With overlapping speech included and no boundary collar, it recorded a 14.72% diarization error rate, compared with 19.3% for the next-ranked system. NVIDIA also reported lower error than its previous four-speaker Streaming Sortformer across eight evaluation conditions at 1.04-second latency, with an average relative reduction of 41.0% across those conditions.

Written by Sakhanda Wire AI from reporting by Hugging Face. Facts, quotes and full context are in the original article below.

Originally published by Hugging Face on

Read the original on Hugging Face ↗

The reporting and the image belong to Hugging Face and are credited and linked above; the text on this page is Sakhanda Wire's own summary.

← Back to the news

More stories

All the latest news