Nvidia released Nemotron 3.5 Transcribe on Wednesday, giving developers a free, open-weight voice ai model that processes audio in up to 40 languages. According to TechCrunch, the company also dropped version 2 of its open-source Speech dataset named Granary during a busy period for audio technology releases.
Key takeaways
- Nvidia released Nemotron 3.5 Transcribe as a free, open-weight audio model.
- The system handles recordings up to an hour long and separates up to eight speakers.
- Nvidia also rolled out version 2 of its open-source Speech dataset Granary.
- The tools target business meeting and medical transcription software developers.
What can the new nvidia Open transcription model do?
TechCrunch reports that Nemotron 3.5 can translate 31 languages to and from English while beating rivals on benchmarks like Hugging Face’s Open ASR Leaderboard. Developers building apps for workplace meetings and medical transcription can combine the model and updated Granary dataset with tools like the NeMo Speech model. This setup helps custom apps recognize specific industry jargon across different languages, though Transcribe works exclusively with recorded audio rather than live streams.
Hardware compatibility and market context
Packing 30 billion parameters, the model is built to run on a single chip. TechCrunch notes that hardware targets range from the Jetson AGX Thor robotics controller and DGX Spark chip powering new laptops to Blackwell datacenter chips. This release adds to Nvidia's existing open-weight lineup, which includes the smaller Parakeet model featuring 1.1 billion parameters.
What it means for the industry
The new software arrives amid a crowded field of audio releases. TechCrunch points out that Mistral shipped Voxtral 3 earlier in the month, ElevenLabs launched Realtime for instant speech in voice agents, and Salesforce introduced its Launa-Omni voice model. These launches show how competitive the sector has become for workplace applications.
Frequently asked questions
Nemotron 3.5 Transcribe is a free, open-weight model released by Nvidia that transcribes audio in up to 40 languages and translates 31 languages to and from English.
TechCrunch reports that the model can tell up to eight speakers apart in audio recordings up to an hour long.
No, the model only works with recorded audio and does not support live streams according to TechCrunch.
The 30-billion-parameter model runs on a single chip, including the Jetson AGX Thor robotics controller, DGX Spark chip for new laptops, and Blackwell datacenter chips.
Nvidia also released version 2 of Granary, its open-source Speech dataset, to help developers customize applications for specific industries.
Based on reporting by TechCrunch. Read the original article
