NVIDIA releases Nemotron 3.5 Transcribe, an open-weight speech model for recorded audio
Announced · Published · By Contractor AI Shop
NVIDIA released Nemotron 3.5 Transcribe, one promptable open-weight model for transcription in 40 languages, speaker labels, translation and questions about recorded audio. It uses about 4.3B active parameters. It works on recorded audio only, not live streams. Quantized versions for local hardware are "coming soon."
Key facts
- NVIDIA says one promptable model handles transcription, speaker labels and translation, plus questions about the audio, in 40 languages; it is available on Hugging Face. NVIDIA Blog (Shreevar Rastogi)
- NVIDIA says the model uses approximately 4.3 billion active parameters; quantized versions for RTX Spark laptops, DGX Spark and Jetson AGX Thor are "coming soon." NVIDIA Blog (Shreevar Rastogi)
- TechCrunch (Julie Bort) reports it translates 31 languages to and from English, handles recordings up to an hour and up to eight speakers, and works with recorded audio, not livestreams. TechCrunch (Julie Bort)
- The Decoder (Jonathan Kemper) reports Nvidia says the model is built only for recorded audio and that users should spot-check names, speaker changes and answers. The Decoder (Jonathan Kemper)
Our take: what it means for contractors
Turn a recorded OAC/coordination meeting or site walk into a speaker-labeled transcript plus "what was decided," on your own hardware — handy for Spanish/English crews. It's a developer model today, not an app; wait for the quantized local builds.
What we don't know yet
- Release date for quantized local builds
- Any Mac/MLX or Ollama build
- Real-world accuracy on jobsite noise
Sources
- NVIDIA Blog (Shreevar Rastogi) · Thin, Light and Built for Everything: NVIDIA RTX Spark Windows Laptops Gain New Models, Agent Tools and Games · 2026-10-28 · Primary
- TechCrunch (Julie Bort) · Voice AI in vogue: Nvidia releases new open transcription model · 2026-10-28 · Secondary
- The Decoder (Jonathan Kemper) · Nvidia's open Nemotron 3.5 Transcribe combines transcription, translation, and speaker recognition · 2026-10-28 · Secondary