GitHub radar
Transcribe audio on your machine: text with timestamps and speakers
Nvidia released an offline model that turns a voice recording into accurate text — with speaker labels and timestamps, on your own machine, no cloud needed.
01nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B 51424k downloads/mo34B paramsautomatic-speech-recognition
A speech recognition model from Nvidia: it listens to audio and returns text — it can't talk back.
When it helps
Useful for transcribing calls, interviews, and meetings in many languages offline; not for a regular laptop or real-time conversation.
Pros
- Understands 40 languages, including rare ones
- Tags who's speaking and adds precise timestamps
- Handles up to an hour of audio at once
- Runs locally, with no paid cloud service needed
Cons
- Needs a Mac with 64GB+ memory or a 48GB GPU
- Even the compressed version runs about 20GB
- Not built for real-time speech recognition
How to set it up — step by step
- 1Open your coding agent in your project folder.
- 2Send the agent the text from the block below.
- 3Wait while the agent picks a version for your hardware and deploys it.
- 4Give the agent a multi-speaker audio file to test.
- 5Check that the output includes text, speaker labels, and timestamps.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Set up Nvidia's Nemotron 3.5 Transcribe model for me: https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B — pick the right build for my hardware and show me how to feed it a multi-speaker recording and get a transcript with timestamps.
You need a Mac with 64 GB of memory or more, or a GPU with 48 GB — even the compressed version takes about 20 GB, and the original about 67 GB. A regular laptop will not handle it.
Open on Hugging Face▌ More finds