None
None
Long Story Video Skill
Turn your idea into 10s–10min video—AI writes scripts, prompts, and generates footage automatically.
Ads Video Skill
Generate professional ads and sales videos—AI auto-generates scripts, prompts, and footage.
3D Science Explainer Video Skill
Convert scientific concepts into stunning 3D explain animations
Feedback
😢 Every free AI generation costs us real GPU money. We pay out of pocket to keep this free for you.
💎 Upgrade to Pro and unlock:
- ✓HD Download
- ✓No Watermark
- ✓Unlimited Generations
nemotron transcribe
Nemotron transcribe is an open-weights speech model: instruction-set styles, 40 languages, one-hour audio, speaker-attributed timestamps.
All Tools
Discover our comprehensive AI-powered animation toolkit
Why Use Nemotron Transcribe
No single transcript fits every reader. Nemotron Transcribe lets you name the convention you want in plain language and follows it.
- Instruction-Steered Transcription StylesVerbatim, written-form, contextual and speaker-attributed styles, chosen per request by instruction.
- Speech-Augmented Language Model ArchitectureA speech encoder, a frozen diarization model and a 30B-A3B hybrid MoE LLM at about 4.3B active parameters.
- Open Weights and Fine-Tuning ToolkitOpenMDW-1.1 weights in BF16, FP8, NVFP4, GGUF and DFlash2 form, plus the NeMo Speech toolkit for Nemotron Transcribe.
How to Run Nemotron Transcribe
Serve a checkpoint, write the instruction, then decode and align — three steps to an attributed transcript.
Nemotron Transcribe Features
Leaderboard-leading accuracy, speaker attribution for overlapping speech and fast NIM inference — what Nemotron Transcribe brings.
Open ASR Leaderboard Accuracy
Mean WER of 4.71% over the 9 English Open ASR test sets and 4.62% on the multilingual board.
Speaker-Attributed Transcription for Overlapping Speech
From one-minute segments to full meetings, it is the first SALM to beat cascaded systems on attributed ASR.
Contextual Transcription with Entity Lists
Expected entities cut named-entity errors six-fold on ContextASR-Bench, driven by the parallel expert encoder.
Speaker-Attributed Word Timestamps
A small CTC adapter maps the finished transcript back onto the audio, so long overlapping calls still get per-speaker timings.
Efficient Inference and Throughput
The model's MTP layer or a DFlash2 draft speeds decoding; the NVFP4 NIM does 2,238 audio seconds per second on one RTX PRO 6000 Blackwell.
Single-GPU Adaptation to New Tasks
A supervision-balanced mixture lets one GPU pick up new languages and domains in about 0.5–2.3 GPU-hours per task.
Nemotron Transcribe — FAQ
Common questions about Nemotron Transcribe: benchmarks, languages, timestamps, licensing and checkpoints.
What is Nemotron 3.5 Transcribe?
A speech-augmented language model joining a speech encoder, a frozen diarization model and a 30B-A3B hybrid MoE LLM.
Which transcription styles can I request?
Verbatim, written-form, contextual, speaker-attributed and rich transcription, plus translation and language ID, each set by an instruction.
How accurate is it on public benchmarks?
4.71% mean WER on the 9 English leaderboard sets, 4.62% multilingual, and six-fold fewer named-entity errors on ContextASR-Bench.
How long can recordings be, and how many languages are supported?
One Nemotron Transcribe checkpoint handles recordings up to an hour across 40 languages, code-switched speech included.
Can it tell me who spoke and when?
Yes — overlapping multi-speaker audio is attributed per speaker, and per-speaker word timestamps come back on request.
What license and checkpoints are released, and can I fine-tune?
OpenMDW-1.1 weights ship as BF16, FP8, NVFP4, GGUF and DFlash2 files; NeMo Speech fine-tunes them in about 0.5–2.3 GPU-hours per task.
Start Transcribing with Nemotron Transcribe
Pick a checkpoint, serve it with the Nemotron Transcribe NIM, and send your first instruction-driven request — styles, translation, language ID and speaker timestamps in 40 languages.