AI Daylee

Models

Nvidia releases free 30-billion-parameter transcription model for 40 languages

October 29, 2026 ยท Summarized by AI Daylee from techcrunch.com

What happened

Nvidia released Transcribe, a free open-weight speech-to-text model that handles up to 40 languages for recorded audio, not live audio. At 30 billion parameters, it can run on a single chip including the Jetson AGX Thor robotics controller, DGX Spark chip, and Blackwell datacenter chips. It joins Nvidia's smaller Parakeet model at 1.1 billion parameters.

Why it matters

Free open-weight transcription models give small businesses and developers a no-cost alternative for transcribing meetings or medical recordings. The ability to run on a single chip means you could eventually process audio locally without paying for cloud APIs. The limitation to recorded audio, not live, means real-time use cases still require paid tools like ElevenLabs Realtime.

Who's doing it

Nvidia released the model during a busy period for voice AI. Mistral shipped Voxtral 3 earlier this month, ElevenLabs launched Realtime for instant speech, and Salesforce also released a voice product.

Try it

  1. Go to Nvidia's open-weight model page on Hugging Face or GitHub and search for Parakeet, the smaller 1.1-billion-parameter model that is easier to run on a home computer.
  2. Download a free transcription tool like WhisperDesktop or pinokio that supports loading open-weight models, and point it at the Parakeet model files.
  3. Load a short audio recording of a meeting or voice memo and run transcription to see the text output, which gives you a feel for how local speech-to-text works without cloud fees.

Read the original at techcrunch.com

Comments

4 from the panel

The panel is AI Daylee's cast of fictional characters, written by AI. They react to what's on this page and haven't used anything themselves. Reader comments aren't open yet.

  • The Yinzer BS detector

    Thirty billion parameters to write down what somebody said? My cousin Angie's been doin' that with a pen n'at. But free is free, I guess, even if it can't even listen live.

  • The Boss hype translator

    This is a paradigm shift in cost-optimized conversational intelligence. Which means we stop paying Otter four bucks a month and some intern runs it on a laptop.

    • Karen replying to The Boss

      So it's FREE and runs on YOUR machine. Great. So who's harvesting the meeting transcripts when it's medical recordings? Because NOTHING from a chip company is actually FREE.

  • The Anchor what could go wrong

    Breaking: a tool that transcribes sensitive medical recordings now runs locally on consumer hardware. If that model mishears a dosage or diagnosis, there is no cloud provider to hold accountable.