Nvidia releases an open model to transcribe recorded meetings
In brief
Nvidia released Nemotron 3.5 Transcribe, a free open-weight model that transcribes audio in up to 40 languages, and version 2 of Granary, an open source speech dataset. According to Nvidia, the model translates 31 languages to and from English, handles recordings up to an hour long, and distinguishes up to eight speakers. It works on recorded audio only, not livestreams.
Source : TechCrunch - AI - « Voice AI in vogue: Nvidia releases new open transcription model »
What it changes for an SME
For an SME of 10 to 250 people, this is not a product you install without technical help. Nvidia is targeting developers of apps for business meetings and medical transcription. An open-weight model that Nvidia says can run on a single chip is an option to transcribe recordings in-house instead of sending audio to a cloud service, which can affect cost and where voice data is processed. It does not handle live audio, and Nvidia does not state hardware prices.
What you can do
- Check whether your meeting notes or dictation tools send audio to an outside provider.
- Separate the need to transcribe recordings from the need for live captions: according to Nvidia this model covers only recordings.
- If industry-specific transcription matters, ask your IT team whether an open-weight model can be tested on your hardware.
Let's talk about the time you could save.
In 20 minutes, we find where AI could save you time or money. Free, no commitment.