🎛️ Track Extract

Local-first desktop stem separation for producers, engineers, DJs, remixers, and creators.

Track Extract wraps AI stem-separation models in a practical desktop workflow: import audio, choose a curated workflow, run an offline render, preview the generated stems, and export DAW-ready files with predictable names and folders.

The current prototype uses a Tauri + React desktop shell with a Python-owned ML engine. Rust stays intentionally thin: it handles desktop plumbing, command/event forwarding, local media serving, path resolution, and process cancellation. Python owns projects, sessions, jobs, model registry state, installs, catalog sync, and separation providers.

🎬 Product Preview

From audio import to six previewable, mix-ready stems in one local desktop workflow.
Explore the interactive Track Extract showcase on Phlosion →

✨ Current Status

Track Extract is usable as a development prototype, not a polished end-user release.

Note

Track Extract is a personal portfolio project and is not accepting external contributions.

What works today:

• Desktop app shell with React, TypeScript, Tauri 2, and a thin Rust bridge.
• Browser-based development view on the same local engine state as the desktop window.
• Drag/drop and file-picker audio import.
• Project/session folder creation under TrackExtract Projects.
• Job queue with queued, preparing, running, complete, failed, and cancelled states.
• Python providers for Demucs, audio-separator, and stub/test renders.
• Model registry with Demucs presets, downloadable public UVR entries, MVSEP references, source links, and license notes.
• audio-separator catalog sync for discovering local model filenames.
• Stem preview and export flows through stable Tauri commands.
• Multi-step workflow plumbing for curated vocal cleanup chains.

Not included yet:

• No DAW plugin.
• No real-time separation.
• No cloud processing, account system, payments, or subscription logic.
• No fully packaged Python runtime for non-developer installs yet.

🧭 Architecture

The canonical engine source lives in engine/src, packaged as trackextract_engine. It exposes python -m trackextract_engine for the Rust bridge and future CLI/service work.

Layer	Role
src/	React + TypeScript UI for workspace, models, workflows, jobs, preview, and export.
src-tauri/	Tauri desktop shell, Rust command bridge, event forwarding, cancellation, media URLs.
engine/	Python ML/workflow engine for projects, sessions, jobs, providers, installs.
resources/	Bundled model and workflow registries copied into app data.
schemas/	JSON schemas for model, workflow, and session formats.
scripts/	Setup, validation, formatting, linting, testing, and registry-generation helpers.
docs/	Architecture, development, and model-registry notes.

See docs/architecture.md for command/event flow and backend boundaries.

🚀 Quick Start

Install JavaScript dependencies:

npm install

Create the managed Python engine environment:

scripts/setup-trackextract-engine.sh

On Windows PowerShell:

scripts/setup-trackextract-engine.ps1

Run the desktop app:

npm run tauri:dev

Track Extract auto-detects .venv-trackextract-engine/bin/python when launched from this repo. To force a specific Python environment, set TRACKEXTRACT_ENGINE_PYTHON.

For NVIDIA CUDA support through audio-separator, build the engine environment with the GPU extra:

TRACKEXTRACT_AUDIO_SEPARATOR_EXTRA=gpu scripts/setup-trackextract-engine.sh

scripts/setup-trackextract-engine.ps1 -Runtime gpu

The GPU setup installs aligned PyTorch CUDA 11.8 packages and runs real CUDA, ONNX Runtime, and bundled FFmpeg probes before reporting success.

For Windows DirectML:

TRACKEXTRACT_AUDIO_SEPARATOR_EXTRA=dml scripts/setup-trackextract-engine.sh

scripts/setup-trackextract-engine.ps1 -Runtime dml

🐧 Linux Tauri Prerequisites

On Ubuntu-like Linux systems, Tauri needs WebKit/RSVG/AppIndicator development packages:

sudo apt update
sudo apt install pkg-config libdbus-1-dev libwebkit2gtk-4.1-dev build-essential curl wget file libxdo-dev libssl-dev libayatana-appindicator3-dev librsvg2-dev

The npm Tauri scripts sanitize Snap VS Code's GTK/GIO environment automatically. Raw tauri dev commands launched from Snap VS Code terminals may still hit a libpthread symbol lookup error.

🌐 Browser Development

For browser UI iteration:

npm run dev:browser

When the Vite server is running on http://localhost:1420, both the Tauri desktop window and a normal browser tab use the same local Python engine bridge. Imports, current project/session state, jobs, generated stems, model installs, and workflow changes are shared during development.

🧠 Models and Workflows

The generated bundled registry lives at resources/models.json. Source fragments live under resources/models/.

npm run models:build
npm run test:models

Model entries can represent:

• Installed local providers, such as Demucs runtime presets.
• Direct-download public model files.
• audio-separator installable filenames.
• Source references for models that need licensing/artifact follow-up before Track Extract can install them directly.

Workflow presets live in resources/workflows.json. The UI also supports custom named workflows.

See docs/model-registry.md for registry structure, source notes, and install-state semantics.

✅ Quality Checks

Run the full local suite:

npm run check

This runs formatting checks, frontend linting, Python linting, Rust clippy, model/workflow validation, Python engine tests, frontend tests, production frontend build, and Rust tests.

Network checks for model download URLs are opt-in so local tests are not flaky:

TRACKEXTRACT_TEST_NETWORK=1 npm run test:all

The Linux CI lane also runs the frontend with coverage thresholds and exercises the production build through a Playwright browser smoke test. Production JavaScript and CSS gzip sizes are checked against explicit budgets.

🗺️ Roadmap

1. Package the Python engine cleanly for end users.
2. Expand audio-separator catalog sync into curated installable workflows.
3. Improve multi-step vocal cleanup execution.
4. Add native ONNX Runtime only where it clearly improves packaging or acceleration.
5. Improve batch processing, cancellation, and resumable jobs.
6. Add DAW export templates for Ableton, Logic, Pro Tools, Reaper, and FL Studio.
7. Explore a future VST3/AU bridge plugin that talks to the same offline Track Extract engine.

📄 License

Track Extract is released under the MIT License.