Nemotron Transcribe — Style-Aware Speech to Text
Drop in your audio and describe the transcript you need — verbatim, cleaned up, translated or speaker-labelled.
None

None

None

Long Story Video Skill

Turn your idea into 10s–10min video—AI writes scripts, prompts, and generates footage automatically.

Ads Video Skill

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

AI Video Prompt Generator

Feedback

freeTrialImage.bannerPity

freeTrialImage.upgradeUnlock

  • ✓freeTrialImage.benefitHd
  • ✓freeTrialImage.benefitWatermark
  • ✓freeTrialImage.benefitUnlimited

nemotron transcribe

Turn raw audio into clean, styled text with Nemotron Transcribe: describe the format you want and get speaker-aware output in 40 languages.

All Tools

Discover our comprehensive AI-powered animation toolkit

Why Teams Pick Nemotron Transcribe

A courtroom, a podcast studio and a call centre all need different transcripts. With Nemotron Transcribe you simply describe the convention and the model complies.

  • Request the Style You Need
    Pick verbatim, written-form, contextual or speaker-labelled output just by wording your request — no retraining and no templates.
  • Speech-Aware Model Stack
    An audio encoder, a locked speaker-diarization module and a 30B-A3B mixture-of-experts language model with roughly 4.3B active parameters.
  • Open Weights, Ready to Fine-Tune
    Download OpenMDW-1.1 checkpoints in BF16, FP8, NVFP4, GGUF or DFlash2 form and adapt them with the NeMo Speech toolkit.

Running Nemotron Transcribe in Three Moves

Load a checkpoint, spell out the instruction, then decode and align to end up with a speaker-labelled transcript.

Capabilities Inside Nemotron Transcribe

Benchmark-topping accuracy, clean speaker separation on overlapping talk and fast NIM inference — here is what the model delivers.

Leaderboard-Topping Accuracy

A mean WER of 4.71% across the nine English Open ASR test sets, and 4.62% on the multilingual board.

Speaker Labels Through Overlapping Speech

Whether it is a one-minute clip or a full meeting, this is the first SALM to outscore cascaded setups on speaker-attributed ASR.

Entity Lists Sharpen Contextual Output

Supplying the expected entities cuts named-entity errors six-fold on ContextASR-Bench, thanks to the parallel expert encoder.

Word Timings Tied to Each Speaker

A compact CTC adapter realigns the finished transcript against the audio, so long overlapping calls still get per-speaker timings.

Fast Serving and High Throughput

An MTP layer or a DFlash2 draft accelerates decoding; the NVFP4 NIM processes 2,238 audio seconds per second on a single RTX PRO 6000 Blackwell.

Adapt to New Tasks on One GPU

A supervision-balanced mixture lets a single GPU absorb new languages and domains in roughly 0.5–2.3 GPU-hours per task.

FAQ

Nemotron Transcribe: Frequently Asked Questions

Answers on benchmarks, language coverage, timestamps, licensing and available checkpoints for Nemotron Transcribe.

1

What exactly is Nemotron 3.5 Transcribe?

It is a speech-augmented language model that combines an audio encoder, a frozen diarization model and a 30B-A3B hybrid MoE LLM.

2

Which transcription styles can I request?

Verbatim, written-form, contextual, speaker-attributed and rich transcription, plus translation and language identification — each one selected through an instruction.

3

How does it score on public benchmarks?

A 4.71% mean WER across the nine English leaderboard sets, 4.62% multilingual, and six times fewer named-entity errors on ContextASR-Bench.

4

What recording length and language coverage should I expect?

A single Nemotron Transcribe checkpoint covers up to one hour of audio across 40 languages, code-switched speech included.

5

Does it identify who is speaking and when?

Yes. Overlapping multi-speaker audio gets attributed per speaker, and per-speaker word timestamps are returned on request.

6

What licence and checkpoints ship, and can I fine-tune?

OpenMDW-1.1 weights arrive as BF16, FP8, NVFP4, GGUF and DFlash2 files, and NeMo Speech fine-tunes them in about 0.5–2.3 GPU-hours per task.

Put Nemotron Transcribe to Work

Choose a checkpoint, serve it through the Nemotron Transcribe NIM and send your first request — styles, translation, language ID and speaker timestamps across 40 languages.