Migrate to uv: one-command install (uv sync) + CI smoke test

[bhadraagada]

Summary

This PR migrates the project from pip-based dependency management to uv, enabling a one-command install experience and adding CI with smoke tests. Users can now install all dependencies, including PyTorch, with a single uv sync --extra cpu command.

What's Changed

One-command install with uv

• Added pyproject.toml: Migrated all dependencies from requirements.txt to modern Python packaging standards
• PyTorch extras: CPU-only PyTorch available via --extra cpu flag
• Platform-aware resolution: Configured for Windows and Linux compatibility
• Locked dependencies: uv.lock ensures reproducible installs

CI with smoke tests

• Added .github/workflows/ci.yml: Uses official astral-sh/setup-uv action
• Smoke tests: tests/test_ci_smoke.py validates core imports (numpy, librosa, soundfile, torch) and project modules (encoder, synthesizer, vocoder)
• System dependencies: Installs libsndfile1 for audio library support
• Manual triggers: Added workflow_dispatch for on-demand CI runs

Updated documentation

• README: Added "Install with uv" section with simple CPU/GPU instructions
• Git ignore: Added .venv/ and .pytest_cache/ to avoid committing local artifacts

Installation comparison

Before (multi-step):

# Install PyTorch manually first
pip install torch --index-url https://download.pytorch.org/whl/cpu
pip install -r requirements.txt

After (one command):

# Install uv once
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install everything
uv sync --extra cpu

Technical details

• Python compatibility: Maintains support for Python ≥3.7 as specified in original README
• Dependency pins: Preserves exact versions from requirements.txt for stability
• Cross-platform: Works on Windows (dev) and Linux (CI) with platform-specific wheel selection
• PyTorch integration: Uses uv's index feature to fetch CPU wheels from PyTorch's official index

Testing

• Local Windows: uv sync --extra cpu + uv run pytest -q → 2 passed
• CI Ubuntu: All steps pass including system dependency installation and smoke tests

Files changed

• pyproject.toml - uv project configuration with dependencies and PyTorch extras
• .github/workflows/ci.yml - CI workflow using uv
• tests/test_ci_smoke.py - Import validation tests
• README.md - Added uv installation instructions
• .gitignore - Ignore local virtual environment

Migration path for users

Existing users can continue using pip install -r requirements.txt. New users get the streamlined uv sync experience. No breaking changes to functionality.

---

Addresses: Request for uv migration with trivial install and CI unit tests

[CorentinJ]

Thank you, I will be reviewing in the following days

[CorentinJ]

One thing: remove the requirements file, this needs to be a complete replacement. I don't want users to keep using pip or conda

[bhadraagada]

I've updated that

[CorentinJ]

• I believe you hadn't set a mechanism for selecting cuda for torch, I did that
• I updated the python, numpy and sci packages versions because on numpy 1.20.* there would not be any prebuilt binaries and I didn't want users to have to build on windows

lmk if all is good on your end. I'll also update the readme with the new instructions

[CorentinJ]

Haven't had a reply from you, will merge