NVIDIA Announces Free Transcription AI "Nemotron 3.5 Transcribe" Supporting 40 Languages
On October 28, NVIDIA announced "Nemotron 3.5 Transcribe," an open-weight AI model supporting transcription, translation, and speaker diarization. It supports transcription in 40 languages, mutual translation with English in 31 languages, and speaker diarization in 13 languages, processing up to 1 hour of audio.
On October 28, NVIDIA announced "Nemotron 3.5 Transcribe," an open-weight AI model supporting transcription, translation, and speaker diarization. This 30B MoE model has 4.3B active parameters. It supports transcription in 40 languages, mutual translation with English in 31 languages, and speaker diarization in 13 languages, with the ability to add speaker labels and word timestamps. It can process up to 1 hour of audio.
The model posted a 4.71% mean WER on the English Open ASR Leaderboard. The base model is released on Hugging Face, and the company states that FP8 and NVFP4 quantized models are scheduled to be released soon. The license is "OpenMDW-1.1," which allows for free commercial use.
This item is a summary. The full, checkable version is at the original source:
Read the original at PC Watch ↗Read the original at MarkTechPost ↗