Nvidia released Nemotron 3.5 Transcribe, a 30B open-weight model for transcription and translation across 40 languages.
It handles recordings up to an hour, separates up to eight speakers, and runs on a single chip from Jetson to Blackwell, alongside Granary v2.
The new system can translate 31 languages to and from English, handle recordings of up to an hour, and distinguish up to eight different speakers. It works exclusively with recorded audio, not live streaming.
Where is it addressed and how is it adapted?
The new tools are aimed at developers who build apps for business meetings and medical records.
Developers can use the model, the company's updated Granary data collection, and other tools to customize apps to recognize specialized terminology.
Operational capabilities and competition
Despite its size of 30 billion parameters, the model can run on a single chip, from systems for robotics and wearable devices to powerful data center processors.
↳
READ ALSO
HACKS.GR / RELATED
AI Unlocked
Meta releases Waddle World, a retro virtual pet with artificial intelligence
↗
This announcement comes at a time of intense activity in the field of voice AI, as other companies have recently introduced similar products, such as Mistral with Voxtral 3, ElevenLabs with Realtime, and Salesforce with Launa-Omni.
G
HACKS.GR / GOOGLE SEARCH
Hacks.gr on Google
Set Hacks.gr as a preferred source.
Preferred source
↗
end of article
Top comments (0)