Nvidia releases Nemotron 3.5 Transcribe, a free speech-to-text model

In short: Nvidia has released Nemotron 3.5 Transcribe, a free speech-to-text AI model for turning recorded audio into written text in up to 40 languages.

What happened

Nvidia released a new AI model called Nemotron 3.5 Transcribe. It is designed to transcribe audio, meaning it listens to a recording and writes out what was said, like an automatic note-taker.

Nvidia says the model supports up to 40 languages and can translate 31 languages to and from English. It can also process recordings up to an hour long and separate up to eight different speakers, which is useful for group conversations where voices overlap.

The company also released version 2 of “Granary,” an open source speech dataset. A dataset is a large collection of example audio used to train and test speech systems, like giving a student many practice recordings to learn from.

One key limit is that Nemotron 3.5 Transcribe is for recordings, not live audio. That means it is better for tasks like writing up meeting notes after a call, rather than creating captions instantly during a call.

Why it matters

Speech-to-text tools are already common at work, especially for meeting summaries and medical dictation. A free model that supports many languages could make these features cheaper to build into products, including tools that focus on workplace conversations like /observe-ai and /parloa. At the same time, people who need real-time transcription will still look to other options that focus on live speech.

Source: TechCrunch AI