BUSINESSNew tool · · Source: TechCrunch
Nvidia releases a free AI transcription model for 40 languages
On October 28, 2026, Nvidia released Nemotron 3.5 Transcribe, a free, open speech recognition model. It transcribes recordings in up to 40 languages and distinguishes up to eight speakers.
What the model does
- Transcription of recordings in up to 40 languages (the coverage doesn't list them – the model page does).
- Speaker recognition: up to eight people per recording.
- Translation: 31 languages to and from English.
- Length: recordings up to about an hour.
Limitations
The model only works with finished recordings, not live during a conversation. At around 30 billion parameters it needs capable hardware – Nvidia lists devices from Jetson AGX Thor and DGX Spark up to data-center chips. It's a building block for developers and IT teams, not a ready-made app.
Availability
The model is free on Hugging Face. Nvidia also released version 2 of Granary, its open speech dataset.
What this means for you
- Transcription is becoming cheap, standard technology. Over time that lowers prices and improves meeting, dictation and note-taking tools.
- Companies with their own IT can run transcription themselves instead of sending every recording to an outside service.
- For individuals, a ready-made tool is still the easier route – an open model needs technical setup and upkeep.
Luigi's take
For most professionals, this isn't a reason to install a model yourself. It's still interesting, because it shows where things are heading: good speech recognition is becoming basic equipment. The value moves from “can the tool transcribe?” to “what does the tool do with what was said?” – summaries, tasks, decisions.
My advice: if you still type notes after every meeting, you're losing time you could easily save. Pick a ready-made meeting-notes tool, test it for two weeks and measure how much time you get back. Open models are more of a topic for IT once volume and requirements grow.
Source
TechCrunch: Voice AI in vogue: Nvidia releases new open transcription model
More sources
- Hugging Face – Modellseite
This article was researched and written with AI assistance. Responsible: Luigi Lugmayr.
