Nemotron Transcribe — Directable Speech-to-Text Workbench
Add your audio, describe the transcript conventions you need, and let the model apply them across long recordings and multiple languages.
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

Direct the transcript format in everyday language, and Nemotron Transcribe applies it to recordings up to an hour in 40 languages, with speaker labels included.

All Tools

Discover our comprehensive AI-powered animation toolkit

Nemotron Transcribe: Output Conventions You Control

Every team reads transcripts differently. Nemotron Transcribe lets you spell out the format you need—verbatim or polished, with or without speaker tags—and applies it to the recording.

  • Choose the Transcript Style per Request
    Ask for verbatim, readable, context-aware or speaker-tagged output; the model switches conventions without retraining.
  • Speech Encoder Meets a Sparse Language Model
    A speech encoder, a locked diarization module and a 30B-A3B hybrid MoE language model with roughly 4.3B active parameters work together.
  • Open Weights Plus a Fine-Tuning Toolkit
    Download OpenMDW-1.1 checkpoints in BF16, FP8, NVFP4, GGUF or DFlash2 formats, and adapt them with the NeMo Speech toolkit for Nemotron Transcribe.

Deploy Nemotron Transcribe in Three Steps

Serve a checkpoint, phrase the output you want, then decode and align the audio—three moves toward a speaker-aware transcript.

What Nemotron Transcribe Delivers

Top-tier accuracy, speaker attribution for overlapping voices and fast NIM inference—these are the strengths Nemotron Transcribe brings to production audio.

Strong Open ASR Leaderboard Scores

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

Speaker Attribution for Overlapping Talk

From short clips to full meetings, it is the first SALM to outperform cascaded systems on attributed ASR.

Entity Lists Reduce Contextual Errors

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

Per-Speaker Word Timestamps

A compact CTC adapter realigns the finished transcript with the audio, so long overlapping calls still receive timings for each speaker.

Fast Inference and High Throughput

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

Single-GPU Adaptation for New Domains

A supervision-balanced mixture lets one GPU learn new languages and tasks in roughly 0.5–2.3 GPU-hours per task.

FAQ

Nemotron Transcribe: Common Questions

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

1

What Exactly Is Nemotron 3.5 Transcribe?

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

2

Which Output Styles Can I Ask For?

You can request verbatim, written-form, contextual, speaker-attributed or rich transcription, along with translation and language identification, by stating the convention you want.

3

How Does It Score on Public Benchmarks?

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

4

What Recording Length and Language Coverage?

A single Nemotron Transcribe checkpoint supports recordings up to one hour across 40 languages, including code-switched speech.

5

Can It Identify Speakers and Timings?

Yes. Overlapping multi-speaker audio is attributed by speaker, and per-speaker word timestamps are available when requested.

6

What License, Checkpoints and Fine-Tuning Options?

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

Start Using Nemotron Transcribe Today

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