
Digital health clinical asr finetune
Fine-tunes Parakeet TDT ASR on clinical vocabulary and measures the gain offline
- Scripts
- None
- Platforms
- Claude Code, Codex, Cursor
- GitHub stars
- 4K
- Last commit
- 3 days ago
- License
- Apache-2.0
- Listing updated
- 1 day ago
Similar skills
About Digital health clinical asr finetune
Digital health clinical asr finetune is an open-source agent skill for health and science. It is available on Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Windsurf and Goose.
This skill guides an agent through a full-model supervised fine-tune of NVIDIA's Parakeet TDT v2 speech recognition model on a clinical transcription manifest. It covers a term-aware train and validation split stratified by entity category, a stock NeMo training script run inside the NeMo container, and a cycle N+1 offline re-evaluation that compares error rates against the previous cycle.
The workflow includes gating rules: fine-tuning is only recommended when the priority category error rate is above a set threshold and the manifest has enough rows. The skill also warns against fine-tuning the streaming Nemotron base and against adapter-style training on TDT and RNNT decoders, and lists hyperparameters that were verified on a small reference manifest.
It expects a CUDA GPU host with roughly 16 to 24 GB of VRAM, the NeMo container, and an NVIDIA API key for certain evaluation and deployment steps. It can optionally hand the resulting model to a companion skill for Riva NIM deployment, and includes a cost-confirmation step before any cloud GPU is provisioned.
When your agent uses it
Stage 4 of the Clinical ASR Flywheel. Use when priority KER is above 0.3 to run stock NeMo SFT on Parakeet TDT v2 and offline cycle N+1 re-eval. NOT for generic word boosting (use /finetune-asr).
From the skill's SKILL.md. Your agent reads this description to decide when to load the skill.
What it can access
From a static check of its 7 files (47 KB) on GitHub on 17 November 2026. Nothing was run. Read the files before you install a skill: it can tell your agent to run anything its tools allow.
Check another skill before you install it.
Install
Copy the skill folder to ~/.claude/skills/digital-health-clinical-asr-finetune to use it in all your projects:
tmp=$(mktemp -d) &&
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp" &&
mkdir -p ~/.claude/skills &&
rm -rf ~/.claude/skills/digital-health-clinical-asr-finetune &&
cp -R "$tmp/skills/digital-health-clinical-asr-finetune" ~/.claude/skills/digital-health-clinical-asr-finetune &&
rm -rf "$tmp"For one project only, put the folder in .claude/skills/digital-health-clinical-asr-finetune inside it instead.
Copy the skill folder to ~/.agents/skills/digital-health-clinical-asr-finetune to use it in every repository:
tmp=$(mktemp -d) &&
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp" &&
mkdir -p ~/.agents/skills &&
rm -rf ~/.agents/skills/digital-health-clinical-asr-finetune &&
cp -R "$tmp/skills/digital-health-clinical-asr-finetune" ~/.agents/skills/digital-health-clinical-asr-finetune &&
rm -rf "$tmp"For one repository only, put the folder in .agents/skills/digital-health-clinical-asr-finetune inside it instead.
Copy the skill folder to ~/.cursor/skills/digital-health-clinical-asr-finetune to use it in every project:
tmp=$(mktemp -d) &&
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp" &&
mkdir -p ~/.cursor/skills &&
rm -rf ~/.cursor/skills/digital-health-clinical-asr-finetune &&
cp -R "$tmp/skills/digital-health-clinical-asr-finetune" ~/.cursor/skills/digital-health-clinical-asr-finetune &&
rm -rf "$tmp"For one project only, put the folder in .cursor/skills/digital-health-clinical-asr-finetune inside it instead.
Run in a terminal (add --scope workspace to install it for the current workspace only):
gemini skills install https://github.com/NVIDIA/skills.git --path skills/digital-health-clinical-asr-finetuneOr copy the folder to ~/.gemini/skills/digital-health-clinical-asr-finetune yourself.
For GitHub Copilot in VS Code, copy the skill folder to ~/.copilot/skills/digital-health-clinical-asr-finetune:
tmp=$(mktemp -d) &&
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp" &&
mkdir -p ~/.copilot/skills &&
rm -rf ~/.copilot/skills/digital-health-clinical-asr-finetune &&
cp -R "$tmp/skills/digital-health-clinical-asr-finetune" ~/.copilot/skills/digital-health-clinical-asr-finetune &&
rm -rf "$tmp"For one project only, put the folder in .github/skills/digital-health-clinical-asr-finetune inside it instead.
Get the folder:
git clone --depth 1 https://github.com/NVIDIA/skills.gitThen copy skills/skills/digital-health-clinical-asr-finetune into your agent's skills folder as digital-health-clinical-asr-finetune. Agents that support SKILL.md read one folder per skill.
What it does
- Stratified train-validation split
- Splits a transcription manifest into row-disjoint train and validation sets balanced by clinical entity category.
- Stock NeMo fine-tuning run
- Runs NeMo's standard speech-to-text fine-tuning script in the NeMo container with recommended hyperparameters.
- Offline cycle re-evaluation
- Transcribes audio with the fine-tuned model offline and scores error rates to compare against the previous cycle.
- Base model guidance
- Recommends a supported Parakeet base and warns against bases whose fine-tuning path is unreliable.
- Cloud GPU provisioning steps
- Describes renting a GPU host, with a confirmation gate and reminders to stop billing when finished.
- Troubleshooting guidance
- Lists common failures such as precision errors, out-of-memory issues and manifest path mismatches with fixes.
Good for
- Fine-tune a speech model on medication and condition names from clinical dictation
- Split a transcription manifest into training and validation sets by entity category
- Re-run offline transcription with a fine-tuned model and compare error rates by category
- Decide whether a fine-tuned speech model is worth keeping after a new evaluation cycle
Digital health clinical asr finetune FAQ
What does this skill need before it runs?
A cycle-N manifest and evaluation result, a CUDA GPU host with about 16 to 24 GB of VRAM, the NeMo container nvcr.io/nvidia/nemo:25.11.01, and the companion finetune-asr and riva-asr-custom skills installed. An NVIDIA API key is needed for the offline re-evaluation round trip and any NIM deployment.
Which base model does it recommend for fine-tuning?
It recommends nvidia/parakeet-tdt-0.6b-v2 and says the Nemotron speech streaming base should not be fine-tuned because its training path is unreliable.
When should the fine-tuning step be used?
Only when the priority category error rate is above 0.3 and the manifest has at least 100 rows, with at least 5 rows per priority category. Otherwise the skill routes back to the manifest build step.
Does it deploy the model?
Deployment is optional and delegated to the riva-asr-custom skill. The core workflow covers training and offline evaluation.
Does it provision cloud GPUs automatically?
No. It walks through renting a Brev GPU host but requires the user to type an explicit confirmation before anything is created.
How do I install Digital health clinical asr finetune in Claude Code?
Copy the skill folder from GitHub to ~/.claude/skills/digital-health-clinical-asr-finetune to use it in all your projects, or to .claude/skills/digital-health-clinical-asr-finetune in one project.
Does Digital health clinical asr finetune run scripts?
No. It is instructions only, with no scripts. This comes from a static check of the files on GitHub; nothing was run. Read the files before you install it.
Which AI agents can use Digital health clinical asr finetune?
Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Windsurf and Goose. Each reads skills in the SKILL.md format from its own folder; the install steps on this page list them.
What are the best alternatives to Digital health clinical asr finetune?
The closest alternatives to Digital health clinical asr finetune are Nemotron asr finetune, Train sentence transformers, Huggingface LLM trainer, Digital health clinical asr build and Huggingface vision trainer.
Where this comes from
- Description: from the skill's own SKILL.md.
More from NVIDIA/skills

Guides an agent in converting cuTile GPU kernels to Triton and debugging the result
- Open source
- 4K GitHub stars
- Developer tools

Adds autotuning with cached config selection to CuTile GPU kernels
- Open source
- 4K GitHub stars
- Developer tools

Guides an agent to write, debug and validate cuTile GPU kernels in Python
- Open source
- 4K GitHub stars
- Developer tools

Guides an agent through iterative profiling and tuning of cuTile GPU kernels in TileGym
- Open source
- 4K GitHub stars
- Developer tools

Guides an agent to swap TileGym cuTile kernels into Hugging Face transformers models
- Open source
- 4K GitHub stars
- Developer tools

Composes and deploys an NVIDIA VSS vision stack with captioning, detection, search
- Open source
- 4K GitHub stars
- Design and media




