Windows desktop tool for sound designers to rename and re-tag audio assets to the UCS (Universal Category System) standard. See a video demonstration here.

Professional audio libraries often differ in metadata and naming convention across different libraries and projects. As a sound designer working with many audio libraries, this means that finding the right sounds can be very time consuming and tedious, especially if assets are poorly named or if certain metadata simply doesn't exist. The Universal Category System (UCS) helps solve this issue, but applying it to existing libraries is a very manual, file-by-file process.

Nomen Audio addresses this issue by providing an automated way to standardize audio assets to the UCS convention. The application imports audio assets (.wav) and their embedded metadata (iXML, BEXT, RIFF INFO chunks), and a machine learning backed pipeline infers each file's UCS category, subcategory, and description. Addiitonal matadata fields are also generated, and the user may review every suggestion, edit fields freely in-place, or write additional metadata tags. Nothing touches disk until the user explicitly saves - the assets are then atomically renamed and metadata written. See pipeline.md for more details.

Spreadsheet-style editor with an expandable column layout. The left sidebar shows an imported file tree across two test folders. The selected file — an Ethiopian food market ambience recording — has been analysed: UCS category AMBIENCE › MARKET, CatID AMBMrkt, and FX Name "Group People Are Talking Background" are all auto-populated, with colour-coded badges indicating generated fields. The bottom strip shows the waveform player mid-playback, with the old filename on the left and the proposed UCS-standard rename on the right.

System requirements: Windows 10 or 11 (64-bit), 8 GB RAM (16 GB recommended), ~4 GB free disk space. No Python, Node.js, or Rust required — everything is bundled.

Go to the Releases page and download one of:

Double-click the downloaded file. If Windows shows a "Windows protected your PC" SmartScreen warning, click More info → Run anyway. This appears because the app is not yet code-signed — it is not a virus alert.

Accept the UAC prompt and follow the installer wizard. Default path: C:\Program Files\Nomen Audio\.

The app opens immediately. On first launch:

• The Python backend starts in the background (takes a few seconds — normal).
• The status bar shows "Loading models…" — ML models load in the background. The app is fully usable as a manual metadata editor while this happens.

Model weights are not bundled in the installer and download automatically on first use:

Total on first full use: ~2.8 GB. Subsequent launches use cached weights — no download.

Weights are stored in the standard Hugging Face cache at %USERPROFILE%\.cache\huggingface\hub\.

For full troubleshooting steps (antivirus, proxy, slow first analysis), see docs/installation.md.

• Git
• uv (Python package manager)
• Node.js ≥ 20
• Rust stable + cargo
• Visual Studio Build Tools (C++ workload, required by Tauri)

git clone https://github.com/your-org/nomen-audio.git
cd nomen-audio

# Python backend
uv sync

# Frontend
cd frontend
npm install

UCS data: The data/UCS/ directory (committed in repo) must contain the UCS 8.2.1 spreadsheet, downloaded free from universalcategorysystem.com. Place the .xlsx file at data/UCS/UCS Master List V8.2.1.xlsx.

cd frontend
npm run tauri dev

This starts the Vite dev server, spawns the Python sidecar, and opens the Tauri window with hot-reload.

# Python backend
uv run pytest -q

# Frontend (Vitest)
cd frontend && npm run test

pwsh -NoProfile -File scripts/build.ps1

The script runs tests → lint → PyInstaller sidecar → Tauri installer and prints the output path on completion. Installers land in:

frontend/src-tauri/target/release/bundle/msi/
frontend/src-tauri/target/release/bundle/nsis/

See scripts/README.md for details on individual build scripts.

nomen-audio/
├── src/                  # Python FastAPI sidecar (business logic, I/O, ML)
│   └── app/              # Main package — routers, services, db, ml, ucs, metadata
├── frontend/             # Tauri + Svelte 5 application
│   ├── src/              # Svelte components, stores, TypeScript
│   └── src-tauri/        # Rust shell, Tauri config, sidecar binaries
├── data/                 # UCS spreadsheets
├── docs/                 # User-facing documentation
├── scripts/              # PowerShell build automation
├── tests/                # Python pytest suite

The frontend is a pure display layer — it never reads or writes files directly. All file I/O, metadata parsing, UCS lookups, and ML inference run inside the Python sidecar process. On startup the sidecar binds to a random localhost port, prints PORT=<n> to stdout, and Tauri reads that port to construct request URLs. HTTP/JSON over 127.0.0.1 is the only IPC channel. This keeps the Rust layer thin and makes the backend independently testable. See architecture.md for more details.