Nvidia releases open speech-to-text model for recorded audio
Nvidia has released a free, open speech-to-text model capable of transcribing recordings in up to 40 languages.
The model is designed for prerecorded audio rather than live streams. That makes it potentially useful for workflows such as meeting and medical transcription, where recordings can be processed after the fact, but limits its applicability to real-time voice applications.
Real-time transcription remains an area where rivals such as ElevenLabs are already active. Nvidia’s release therefore broadens access to multilingual transcription while leaving it outside the most immediate live-audio use case.