Nvidia Releases Free 40-Language Model for Recorded Audio Transcription

What changed

Nvidia released a free speech-to-text model that can transcribe recorded audio in up to 40 languages. It does not handle live audio, leaving real-time transcription outside the reported capability and still in the field of rivals such as ElevenLabs.

Why it matters

This is immediately useful for meeting-transcription and medical-transcription tools, where the recording already exists and the main job is turning speech into searchable text. The multilingual range also makes the release more practical across teams working in different languages.

The important catch is timing. Nvidia’s model can process the conversation after it happens, but it cannot follow one as it unfolds. That keeps live captions and instant voice workflows dependent on other systems for now. As a product decision, this looks less like a complete voice assistant and more like a broad transcription layer that developers can build around.