Tha.Les
bf561a6c81
* Add on-demand lead/backing vocal split, fix stems relocation bugs, and eager model pre-download (#275, #403) Lead/backing vocal split: - New on-demand POST /api/jobs/{id}/vocal-split endpoint, running UVR-MDX-NET Karaoke 2 (audio-separator) as a second pass over Demucs's vocals.wav - Desktop and mobile UI toggle to request the split, auto-chained once the base separation finishes, for both foreground and background jobs - Mixer shows Lead Vocals / Backing Vocals lanes in place of Vocals once split Stems relocation fixes (#403): - user-data.json (library metadata) now lives inside the jobs folder so it follows a Settings relocation instead of staying behind in Documents - The relocation endpoint's settings persist step was silently swallowing write failures and reporting false success; it now reports persisted: false and the Settings UI shows a clear warning instead Desktop setup wizard: - Demucs, beat-this, and the karaoke model now download eagerly during first-boot setup instead of lazily on first use Also: - Credit audio-separator / Ultimate Vocal Remover in the README per its license's attribution requirement, plus a license audit in docs/models.md - Add models/ to .gitignore * ci: install build-essential so diffq (audio-separator's dependency) can compile diffq has no prebuilt wheel for Python 3.11+ on Linux, its last release only ever shipped cp310 wheels, so uv sync must compile it from source, which needs gcc. Docker and the Linux desktop release build already install build-essential for the same reason; the plain lint/test CI container never needed it before audio-separator (#275) pulled diffq in. * chore: pin Unraid template to 0.12.0 This PR ships as v0.12.0, per the user's decision given it introduces the new lead/backing vocal split feature. --------- Co-authored-by: Thales <>
126 lines
6.0 KiB
TOML
126 lines
6.0 KiB
TOML
[project]
|
|
name = "stemdeck"
|
|
# Version is derived from the git tag by hatch-vcs (see [tool.hatch.version]),
|
|
# so it is never hand-edited — `git tag vX.Y.Z` is the single source of truth.
|
|
# Builds between tags report a dev version (e.g. 0.7.0a5.dev3+g<sha>). (#169)
|
|
dynamic = ["version"]
|
|
description = "Paste a YouTube URL, get audio stems split into a DAW-style player."
|
|
requires-python = ">=3.10,<3.14"
|
|
dependencies = [
|
|
"fastapi>=0.115,!=0.136.3",
|
|
"uvicorn[standard]>=0.30",
|
|
# Floor raised for CVE-2026-55404 (fixed in 2026.7.4). The lock pin alone is
|
|
# not enough: build/Dockerfile installs from these constraints, not uv.lock.
|
|
"yt-dlp>=2026.7.4",
|
|
# Pin below 4.1: demucs 4.1.0 (2026-07-11) added a dependency on `sphn`, a
|
|
# Rust/CMake native package with no Intel-macOS wheel, so the x64 macOS
|
|
# build compiles it from source and fails (CMake dropped <3.5 support).
|
|
# 4.0.1 is what every prior release shipped and what the torch/torchaudio
|
|
# pins below were written for.
|
|
"demucs>=4.0.1,<4.1",
|
|
# Pin torch/torchaudio to 2.6.x where available — torchaudio 2.7+ removed its built-in
|
|
# audio writer and now requires the separate `torchcodec` package, which
|
|
# has ABI issues with torch 2.11. Demucs 4.0.1 calls torchaudio.save() to
|
|
# write stems, so it breaks on 2.7+.
|
|
"torch>=2.6,<2.7; sys_platform != 'darwin' or platform_machine != 'x86_64'",
|
|
"torchaudio>=2.6,<2.7; sys_platform != 'darwin' or platform_machine != 'x86_64'",
|
|
# PyTorch does not publish macOS x86_64 wheels for 2.6.x. Keep Intel macOS
|
|
# on the last compatible line so the x64 runtime pack can still be built.
|
|
"torch>=2.2,<2.3; sys_platform == 'darwin' and platform_machine == 'x86_64'",
|
|
"torchaudio>=2.2,<2.3; sys_platform == 'darwin' and platform_machine == 'x86_64'",
|
|
# torchaudio 2.6 dispatches WAV writes through libsndfile via soundfile.
|
|
# Without this, demucs crashes on save with "Couldn't find appropriate
|
|
# backend to handle uri ... drums.wav".
|
|
"soundfile>=0.12",
|
|
# BPM + key analysis on the downloaded source. Pulls numpy/scipy/numba.
|
|
"librosa>=0.10",
|
|
# Beat/downbeat tracker for the click track. MIT for both code and the
|
|
# published weights. librosa's tracker resolves fast music to half tempo
|
|
# (180 BPM punk -> 90) because of its 120 BPM lognormal prior, which is not
|
|
# tunable away and which no confidence metric detects. Weights (81 MB) are
|
|
# fetched on first use through torch.hub, so they land under TORCH_HOME
|
|
# alongside the demucs models; librosa remains the offline fallback.
|
|
"beat-this>=1.1",
|
|
# Current llvmlite releases no longer publish macOS x86_64 wheels, which
|
|
# makes Intel runtime builds require a full LLVM/CMake toolchain. Keep Intel
|
|
# macOS on a wheel-backed Numba line.
|
|
"numba>=0.61,<0.62; sys_platform == 'darwin' and platform_machine == 'x86_64'",
|
|
# Torch 2.2's Intel macOS wheel was built against NumPy 1.x.
|
|
"numpy<2; sys_platform == 'darwin' and platform_machine == 'x86_64'",
|
|
# ITU-R BS.1770 integrated-loudness (LUFS) measurement for the
|
|
# post-analysis loudness card. Pure Python, depends on scipy which
|
|
# librosa already pulls in.
|
|
"pyloudnorm>=0.1.1",
|
|
# FastAPI multipart/form-data (file uploads) unconditionally require
|
|
# python-multipart — it is not an optional extra.
|
|
"python-multipart>=0.0.9",
|
|
# CVE-2026-44431 / CVE-2026-44432: sensitive header forwarding and
|
|
# decompression-bomb bypass fixed in 2.7.0. urllib3 is a transitive
|
|
# dep via requests; pin floor to pull in the fix.
|
|
"urllib3>=2.7.0",
|
|
# Pure-Python QR code generator used by the /api/qr endpoint (server
|
|
# access QR codes in the desktop settings panel).
|
|
"segno>=1.6",
|
|
# On-demand lead/backing vocal split (#275). Ships with UVR-MDX-NET
|
|
# Karaoke 2 (ONNX) as the default model -- see docs/models.md for why the
|
|
# torch-native alternative (a community roformer checkpoint) was rejected
|
|
# on licensing grounds. Excludes Intel macOS: requires numpy>=2/torch>=2.3,
|
|
# which conflicts with the numpy<2/torch<2.3 pins that platform is stuck on
|
|
# above; the feature gates itself off there rather than fighting the pins.
|
|
# CPU-only onnxruntime for now -- onnxruntime-gpu is a separate, mutually
|
|
# exclusive package requiring the same kind of CPU/GPU build-time split
|
|
# torch already gets via make-portable.{sh,ps1} (see versioning.md); wiring
|
|
# that up for this occasional, best-effort action is a follow-up, not a
|
|
# blocker, since the main separation already gets full GPU accel via torch.
|
|
"audio-separator>=0.24; sys_platform != 'darwin' or platform_machine != 'x86_64'",
|
|
"onnxruntime>=1.19; sys_platform != 'darwin' or platform_machine != 'x86_64'",
|
|
]
|
|
|
|
[project.optional-dependencies]
|
|
dev = [
|
|
"ruff>=0.6",
|
|
"pytest>=8",
|
|
"pytest-asyncio>=0.24",
|
|
"httpx>=0.27",
|
|
]
|
|
|
|
[build-system]
|
|
requires = ["hatchling", "hatch-vcs"]
|
|
build-backend = "hatchling.build"
|
|
|
|
# Derive the version from git tags. Writes app/_version.py at build time as a
|
|
# runtime fallback for environments without installed package metadata.
|
|
[tool.hatch.version]
|
|
source = "vcs"
|
|
# Used when the tag can't be reached (e.g. CI's shallow/tagless clone) so the
|
|
# build never hard-fails. Real builds derive from the tag or the pinned
|
|
# SETUPTOOLS_SCM_PRETEND_VERSION. (#169)
|
|
fallback-version = "0.0.0"
|
|
|
|
[tool.hatch.build.hooks.vcs]
|
|
version-file = "app/_version.py"
|
|
|
|
[tool.hatch.build.targets.wheel]
|
|
packages = ["app"]
|
|
|
|
[tool.ruff]
|
|
target-version = "py310"
|
|
line-length = 100
|
|
# app/_version.py is generated by hatch-vcs at build/sync time — don't lint it.
|
|
exclude = ["jobs", ".run", "static/vendor", "app/_version.py"]
|
|
|
|
[tool.ruff.lint]
|
|
select = ["E", "F", "W", "I", "UP", "B", "SIM"]
|
|
ignore = [
|
|
"E501", # line too long -- format-check is the gate, not lint
|
|
"B008", # FastAPI uses Depends(...) in defaults
|
|
"SIM105", # try/except/pass is fine; contextlib.suppress is heavier for one-shot warm-up imports
|
|
]
|
|
|
|
[tool.ruff.lint.isort]
|
|
split-on-trailing-comma = false
|
|
|
|
[tool.pytest.ini_options]
|
|
testpaths = ["tests"]
|
|
asyncio_mode = "auto"
|