Nemotron 3.5 Transcribe - Intelligent Transcription with a Speech LLM
Nemotron 3.5 Transcribe is an open speech LLM for intelligent transcription: a single model that transcribes 40 languages verbatim or in written form, with context, speaker attribution, and word timestamps, selected with a natural language instruction. It is the best open-weights model on the English and multilingual Open ASR Leaderboards.
Replies
Hi everyone!
An interview transcript might need every hesitation and repeated word. Meeting notes probably don't. But both need names and numbers written correctly.
@NVIDIA 's Nemotron 3.5 Transcribe lets you specify those preferences in plain language. You can ask for verbatim or cleaned-up text, provide a list of technical terms, and label speakers even when people talk over each other.
Supports 40 languages and recordings up to an hour, with word-level timestamps available when needed.
NVIDIA reports a 4.71% mean WER across nine public English Open ASR test sets, the best among open-weight models in its published comparison. Under the hood is a 34B-parameter MoE model with only about 4.3B active parameters.
Weights are available on@Hugging Face , along with a fine-tuning toolkit.