Ableton Audio Separator

Stem separation inside Ableton Live — powered by python-audio-separator and the same AI models used in Ultimate Vocal Remover.

Right-click any audio clip or arrangement selection and split it into stems (vocals, instrumental, drums, bass, guitar, piano) without leaving Live.

Features

• 2-stem separation — Vocals + Instrumental (BS-Roformer, MelBand Roformer, MDX-NET)
• 4-stem separation — Vocals, Drums, Bass, Other (HTDemucs FT)
• 6-stem separation — Vocals, Drums, Bass, Guitar, Piano, Other (HTDemucs 6s)
• GPU accelerated — CUDA support for fast inference
• Auto-install — uses uv to create a local Python venv and install dependencies automatically (no system Python required)
• Progress feedback — single unified progress dialog from render through separation to track creation
• Single undo — all created tracks can be undone in one step

Requirements

• Ableton Live 12 Beta with Extensions SDK support
• NVIDIA GPU + CUDA (optional, for fast processing — falls back to CPU)

Note: Python is managed automatically by uv. You do not need Python on your PATH.

Setup

1. Clone/download this repository
2. Install Node.js dependencies:

   npm install

3. Set the Extension Host path in .env (pre-filled by the generator):

   EXTENSION_HOST_PATH=C:\ProgramData\Ableton\Live 12 Beta\Program\ExtensionHost\ExtensionHostNodeModule.node
   

4. Enable Developer Mode in Ableton Live: Settings → Extensions

Usage

npm start       # build + launch in Live's Extension Host

Once running in Live:

1. Right-click an audio clip → "Separate Clip Stems (UVR)"
2. Or select a time range on an audio track → right-click → "Separate Selection Stems (UVR)"
3. Choose separation mode, model, and options in the dialog
4. Click Separate — the extension handles the rest

On first use, uv will be downloaded and a local Python environment with audio-separator will be created in the extension's storage directory. Model files are downloaded on first use of each model (~100MB–1GB depending on model).

Scripts

npm start          # build + run in Live's Extension Host
npm run build      # production bundle
npm run build:dev  # dev bundle (sourcemaps, not minified)
npm run package    # build + create a .ablx archive for distribution

Project Structure

src/
  extension.ts    — Entry point: commands, context menus, orchestration
  separator.ts    — uv-managed local venv, audio-separator CLI wrapper
  tracks.ts       — Import stems and create audio tracks in the Live Set
ui/
  settings.html   — Modal dialog UI for separation settings

How It Works

1. User selects separation settings from a modal dialog
2. The extension renders audio from Live via renderPreFxAudio()
3. Calls audio-separator from the local .venv/ with the chosen model
4. Monitors stdout/stderr for progress updates (model download, tqdm bars)
5. Imports the resulting stem files back into the project
6. Creates new audio tracks per stem, named like "TrackName — Vocals"
7. All track creation is grouped in a transaction for single-undo

Troubleshooting

• "audio-separator not found" on every launch? Delete the .venv folder in the extension's storage directory and reinstall. On Windows this is at %LOCALAPPDATA%\Ableton\Extensions Data\carson-kompon.audio-separator\.venv.
• Slow separation? Delete the .venv folder and reinstall with the GPU option. Check that CUDA PyTorch was installed (look for "Installing PyTorch with CUDA support" in the progress dialog).
• Duplicate menu entries? Restart Live before running npm start.

License

MIT