NVIDIA Unveils Nemotron 3.5 Transcribe for Speech Recognition
NVIDIA has introduced the new Nemotron 3.5 Transcribe model, designed for speech recognition and voice data processing.
The tool can transcribe audio into text even when multiple speakers are present in a single recording.
The model supports work with 40 languages, including Russian, for which all main system functions are available.
Key capabilities of the technology include:
- Speaker diarization during dialogue with timestamping for each phrase
- Speech translation into other languages and removal of filler words from transcripts
- Ability to answer questions based on the content of the recorded material
- Processing of audio files up to one hour long without quality loss
Compared to competitors, the new development demonstrates significantly higher speed and accuracy metrics in data processing.
You can test the model on the Hugging Face platform.









