NVIDIA Released Nemotron 3 Diarization Model

The new AI tool tracks up to eight speakers in real-time with high precision.

Updated on Sept. 23, 2026 in Artificial Intelligence

Bold flat-color editorial illustration of server cooling fins and optic cables, representing high-performance AI computational architecture.
NVIDIA has launched its Nemotron 3 Diarization model, an AI tool capable of identifying up to eight simultaneous speakers in real-time. AI Illustration. Upload story photo >

Live Poll

Do you trust AI models to accurately identify who is speaking in a recording?

NVIDIA has released the Nemotron 3 Diarization model, an artificial intelligence tool designed to track and identify up to eight speakers in real-time. The model aims to improve automated speech attribution in multi-party conversations.

Why it matters

Speaker diarization allows automated systems to accurately attribute speech to specific individuals in complex, multi-party conversations. This capability is essential for enhancing transcription accuracy and meeting documentation in professional settings.

The Nemotron 3 Diarization model features 100 million parameters and requires Linux with NVIDIA Ampere, Ada Lovelace, Hopper, or Blackwell architecture GPUs. It supports multiple audio formats including .wav, .flac, .opus, and .mp3 at 16 kHz.

The players

NVIDIA

A global technology company that specializes in the design and manufacture of graphics processing units, artificial intelligence software, and hardware infrastructure.

Hugging Face

A leading platform for the artificial intelligence community to collaborate on and distribute machine learning models and datasets.

The details

The architecture utilizes a 31-layer Transformer encoder with rotary positional embeddings and a Conv1D upsampling layer to maintain speaker context. It was trained on 10,000 hours of real conversations and 82,611 hours of simulated mixtures across 21 languages.

Timeline

  1. September 23, 2026: NVIDIA released the Nemotron 3 Diarization model.

The Tech Race

The release of this diarization model reflects a broader shift toward optimizing small-scale, high-efficiency language models for real-time edge applications. This positions the company to compete directly against other AI labs focusing on specialized audio-processing frameworks.

Users and developers can leverage this model to build more accurate automated transcription and meeting summary tools for professional use. The open license allows for broader commercial integration into consumer-facing applications that require voice attribution.

The takeaway

Advancements in diarization technology enable more efficient organization of audio data in collaborative environments. Developers looking to implement this should ensure their infrastructure meets the specific GPU architecture requirements for optimal performance.

Further reading

Learn more about the latest innovations in the field at Artificial Intelligence.

Live Poll

Do you trust AI models to accurately identify who is speaking in a recording?