From the NVIDIA edition of October 28, 2026
NVIDIA's Hugging Face post details Granary v2 and a medical fine-tuning case study, and advises keeping 100 real labelled clips and spot-checking output
NVIDIA (published on Hugging Face) · Speech AI platform (vendor technical blog) · Global
- Publisher
- Hugging Face Blog
- Original publication
- 2026-10-28
- Source retrieved
- 2026-10-29
What happened
NVIDIA reports the model translates 31 languages and attributes speakers in 13, handles up to eight speakers, and ranks first among open-weight models on the English Open ASR Leaderboard at 5.33% average WER (API models excluded). In a medical case study on Eka Care's Indian doctor benchmark, a better prompt cut WER from 30.8% to 23.8% and fine-tuning to 11.8%; on 2,764 held-out clips WER fell from 22.44% to 12.17% (45.8% relative), while LibriSpeech WER rose from 1.54% to 1.69% and a medical specialist model still led on brand-name drugs. Granary v2's public manifest covers 463,062 hours in 48 languages; source audio must be downloaded under each corpus's licence.
Why it matters
Shows the adaptation path an agency or university would follow to tune transcription for local place names, legal or clinical terms and accents, and NVIDIA's own advice to hold out real labelled clips and spot-check names, speaker changes and answers against the recording.
Evidence and measured results
Vendor technical blog with benchmark figures and a held-out-speaker case study on a public medical benchmark; NVIDIA-run evaluation.
Limitations and uncertainty
Case study is medical and Indian-English, not U.S. public-sector audio; general-speech accuracy declined slightly after specialisation; leaderboard figures exclude API models.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-29; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Generic transcription misses local names, legal or clinical terms and regional accents.
- Stakeholders
- Court or clerk operations lead
- University disability services
- IT data science team
- General counsel (licensing)
- Discovery questions
- Which terms and names does your current transcription get wrong?
- Do you have human-verified transcripts to build a test set?
- Can training audio legally be reused under your retention rules?
- Value hypothesis
- A measured tuning project could reduce domain-term errors on the customer's audio, validated on held-out speakers.
- Potential engagement
- Prompt-then-tune evaluation on a held-out local test set.
- Unsupported claims
- That NVIDIA's 45.8% medical gain transfers to public-sector audio
- That tuning removes the need for human review
- Cost of hosted endpoints
Pre-sales engineering
Role takeaway
- Fit
- Domains with stable vocabularies such as courts, councils and lectures.
- Architecture and integration
- NeMo Speech fine-tuning (encoder plus LoRA), vLLM or SGLang serving, optional hosted endpoints.
- Prerequisites
- Labelled local clips with speaker-level holdout
- GPU time for adaptation
- Licence review for training corpora
- Constraints
- Case study is medical Indian-English only
- General-speech accuracy can regress
- Granary v2 audio must be fetched under source licences
- Security
- Keep sensitive training audio on premises; avoid hosted endpoints without contractual controls.
- Proof of value
- Held-out-speaker WER before and after tuning
- General-speech regression check
- Entity accuracy for local names
Delivery
Role takeaway
- Work
- Build a local evaluation set, test prompts, tune if needed and release a versioned model.
- Dependencies
- Labelled clips
- GPU capacity
- Licence and records sign-off
- Ownership
- Shared IT data team with business-owner sign-off.
- Skills and adoption
- Train staff to label clips and review transcripts.
- Governance checkpoints
- Licence review before training
- Model release approval with evaluation record
- Acceptance criteria
- Proposed: held-out WER improves vs best-prompt baseline
- Proposed: no material regression on general speech
- Risks
- Overfitting to tuning clips
- Licence breach on training audio
- Sensitive audio exposure in pipelines
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Adaptation trains the speech encoder and uses LoRA on the LLM, about 14 GPU-hours on one GB300; serving via vLLM or SGLang, with hosted endpoints from Baseten and DeepInfra as alternatives.
Governance
Who approves, reviews and stays accountable for outcomes?
Fine-tuned variants become agency-specific models that need versioning, evaluation records and a rule for when they replace the base model.
Security and privacy
What data, permissions and controls need testing?
Fine-tuning on real recordings means sensitive audio enters training pipelines; hosted endpoints would move audio outside agency control.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Spot-checking guidance implies human reviewers remain necessary; agencies need staff who can build and label evaluation sets.
Procurement
What should contracts, pricing and exit terms secure?
Granary v2 audio comes under each source corpus's own licence, so any agency training effort needs licence review; hosted endpoints are separate commercial services.
Operating model
Which teams own the service once it runs?
A tuning workflow adds data curation, evaluation and model lifecycle tasks to IT or a shared-services team.
Publication history
- 2026-10-28NVIDIA · Issue 539 resources
Stable resource ID: nvidia-hugging-face-nemotron-transcribe-granary-v2-adaptation-case-study-20261028