Today for AI
HOT RADAR
llmHEAT 57.2°

AI CURATED HOT STORY · 10/28/2026

NVIDIA Open-Sources Nemotron 3.5 Voice Model and Data Pipeline

4 reports archived4 independent sourcesupdated 10/28/2026, 23:49:39
Synthesis & Latest Updates
4 Sources Cross-Validated

NVIDIA has officially open-sourced the Nemotron 3.5 Transcribe multimodal voice model and the Granary v2 data pipeline, significantly lowering the barrier to high-quality voice AI development. The model supports transcription, translation, speaker diarization, and audio Q&A across 40 languages, achieving leading results in industry benchmarks while handling recordings up to one hour long with support for up to 8 speakers.

Technically, Nemotron 3.5 utilizes a Mixture-of-Experts (MoE) architecture with 30B total parameters but only 4B active, enabling efficient inference on single GPUs or edge devices like the Jetson AGX Thor. The accompanying Granary v2 provides an 844,000-hour audio dataset, fine-tuning recipes, and data processing tools, allowing developers to quickly adapt the model to vertical scenarios such as healthcare and meetings via prompts.

This full-stack open-source strategy offers end-to-end capabilities from raw audio to customized applications, breaking previous limitations of closed-source or costly deployment. It accelerates enterprise-grade voice interaction adoption and lays the groundwork for smarter, privacy-preserving local voice solutions.

LATEST/NVIDIA recently released the open-source Nemotron 3.5 voice model, integrating transcription, translation, and speaker diarization for 40 languages. Alongside it, the Granary v2 data pipeline and fine-tuning recipes were launched, utilizing an MoE architecture for efficient single-GPU operation to lower development barriers for vertical voice AI applications.

Hourly heat curve

3 fully observed hours
Now
57.2
Peak
57.210-29
24h change
–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 4 reports · Latest first
  1. The DecoderT2·78 pts

    Nvidia Releases Open Nemotron 3.5 Transcribe for Multilingual Speech Tasks

    • Nemotron 3.5 Transcribe is an open-source instruction-following speech model covering transcription, translation, and speaker recognition.
    • Uses sparse activation architecture (30B total/4B active) enabling efficient inference on a single GPU.
    • Released alongside the Granary v2 dataset (844k hours of audio) and tools for vertical domain adaptation.
  2. Hacker News AIT2·78 pts

    Nvidia Releases Nemotron 3.5 Transcribe: Open Multilingual ASR Model

    • Nemotron 3.5 Transcribe is an open-source ASR model supporting 40 languages with speaker attribution, translation, and audio Q&A capabilities.
    • The release includes the Granary v2 open-source data pipeline and fine-tuning recipes, lowering the barrier for model customization.
    • It achieved top leaderboard results in relevant benchmarks, demonstrating competitive performance.
  4. Hugging Face BlogT1·78 pts

    NVIDIA Open-Sources Nemotron 3.5 Transcription Model and Granary v2 Pipeline

    • Nemotron 3.5 integrates transcription, speaker diarization, translation, and audio Q&A with state-of-the-art performance.
    • Granary v2 provides a complete data pipeline and fine-tuning recipes, enabling rapid adaptation to vertical domains via prompting.
    • The full-stack open-source strategy (model + pipeline) offers developers end-to-end capabilities from raw audio to customized applications.