Nvidia releases a free AI model that turns recordings into text

Nvidia has released a new speech-to-text model, Nemotron 3.5 Transcribe, and made it free for anyone to use. It is "open-weight", which means developers can download it and adapt it rather than paying to use a closed service.

According to Nvidia, it can transcribe up to 40 languages, translate 31 languages to and from English, cope with recordings up to an hour long and tell apart as many as eight speakers. It can run on a single chip, including the kind found in some new laptops. Nvidia says developers can train it to recognise the jargon of a particular trade, and its main targets are business meetings and medical transcription.

There are caveats. It only works on recorded audio, not live speech. Nvidia's claim that it beats rivals on industry leaderboards is its own, and TechCrunch gave no scores. The article also doesn't spell out the licence terms, which matter if a business wants to build on it.

Voice AI is busy at the moment, with several other new voice models released this month. This one is a tool for software makers rather than something you install and use yourself, but free, capable building blocks like this tend to end up inside the apps small firms use.

What this means for your business

You don't need to do anything with this directly. If you pay for a meeting-recording or transcription service, though, expect more competition and possibly lower prices. Whichever tool you use, check where your recordings are stored and whether they are used to train its AI before you feed it client conversations.