pyproject.toml

[project]
name = "yet-another-youdub-webui"
version = "0.1.0"
description = "Translate and dub YouTube videos to Chinese using AI"
readme = "README.md"
requires-python = ">=3.10,<3.12"
license = { text = "MIT" }
authors = [{ name = "StarDuster" }, { name = "liuzhao1225" }]
keywords = ["youtube", "translation", "dubbing", "ai", "tts", "whisper"]

dependencies = [
    # Web UI
    "gradio",
    "typer>=0.12,<1.0.0",

    # Logging
    "loguru>=0.7.0,<1.0.0",

    # Configuration
    "pydantic>=2.0.0,<3.0.0",
    "pydantic-settings>=2.0.0,<3.0.0",

    # Video downloading (使用 nightly 版本以获取最新的 YouTube 反爬虫修复)
    "yt-dlp[default,curl-cffi,secretstorage]>=2026.1.29",
    "curl_cffi",
    "yt-dlp-invidious",
    "bgutil-ytdlp-pot-provider",
    "google-genai>=1.50.0",

    # Audio processing
    "scipy>=1.14.0,<2.0.0",
    "librosa>=0.11.0,<0.12.0",
    "audiostretchy>=1.3.5",
    "soundfile",
    "numpy>=2.0.0,<3.0.0",
    "cmudict>=1.1.2",
    "pypinyin",

    # Vocal separation (offline / GPU-friendly)
    # Use demucs-infer (actively maintained inference-only Demucs fork).
    "demucs-infer>=4.1.2",

    # Speech recognition (offline / GPU-friendly)
    "faster-whisper>=1.2.1",
    # Use CTranslate2 >= 4.5.0 to align with cuDNN 9 (avoid cuDNN 8 split libs like libcudnn_ops_infer.so.8).
    "ctranslate2>=4.5.0,<5.0.0",
    # Qwen3-ASR (offline, transformers backend)
    "qwen-asr>=0.0.4",
    # faster-whisper VAD uses onnxruntime. Keep it in cpu/gpu extras so you can choose the build.

    # Speaker diarization / speaker embedding (used by diarization + bytedance voice matching; optional at runtime)
    "pyannote.audio>=4.0.0,<5.0.0",

    # Local TTS (Qwen3-TTS / Qwen-Audio worker)
    "qwen-tts==0.0.5",

    # Translation API
    "openai>=1.0.0,<2.0.0",
    "python-dotenv>=1.0.0,<2.0.0",
    "huggingface-hub",

    # Image processing
    "Pillow>=10.0.0,<11.0.0",

    # HTTP requests (for bytedance TTS)
    "requests>=2.31.0,<3.0.0",

    # Bilibili upload

    # PyTorch ecosystem (CUDA 12.8 on Linux via uv index)
    # Keep as a range so you can upgrade without editing this file again.
    "torch>=2.10.0,<3.0.0",
    "torchaudio>=2.10.0,<3.0.0",
    "torchcodec>=0.10.0",
]

[project.optional-dependencies]
# Development dependencies
dev = ["pytest>=7.0.0", "black>=23.0.0", "ruff>=0.1.0"]

# CPU runtime stack (default runtime dependencies but kept optional so you can swap to GPU builds cleanly)
# Use: `uv sync --extra cpu`
cpu = ["onnxruntime>=1.17.0,<1.24.0"]

# Optional GPU acceleration stack (Linux).
# Use: `uv sync --extra gpu`
gpu = [
    # Used by our cuDNN preloader (and helps onnxruntime/ctranslate2 GPU providers).
    "nvidia-cudnn-cu12; sys_platform == 'linux'",
    # GPU build of onnxruntime (avoid co-install with onnxruntime)
    "onnxruntime-gpu>=1.17.0,<1.24.0; sys_platform == 'linux'",
]

# Flash Attention (optional, significantly speeds up Transformer inference but takes a long time to compile).
# Use: `uv pip install flash-attn --no-build-isolation`
flash = ["flash-attn; sys_platform == 'linux'"]

[project.scripts]
youdub = "youdub.app:main"

[build-system]
requires = ["setuptools>=61.0", "wheel"]
build-backend = "setuptools.build_meta"

[tool.setuptools.packages.find]
where = ["src"]
include = ["youdub*"]

[tool.ruff]
line-length = 120
target-version = "py310"

[tool.black]
line-length = 120
target-version = ["py310"]

# --- UV Package Manager Configuration ---

[[tool.uv.index]]
name = "pytorch-cpu"
url = "https://download.pytorch.org/whl/cpu"
explicit = true

[[tool.uv.index]]
name = "pytorch-cu128"
url = "https://download.pytorch.org/whl/cu128"
explicit = true

[tool.uv]
# 允许使用预发布版本 (nightly)，以获取最新的 YouTube 反爬虫修复
prerelease = "allow"

# faster-whisper / qwen-tts depend on `onnxruntime`. We declare the runtime explicitly in
# [project.optional-dependencies] (cpu/gpu extras), and patch upstream metadata to avoid forcing a specific onnxruntime wheel.
dependency-metadata = [
    # Patch faster-whisper to avoid forcing a bundled onnxruntime dependency.
    { name = "faster-whisper", version = "1.2.1", provides-extra = ["conversion", "dev"], requires-dist = [
        "ctranslate2<5,>=4.0",
        "huggingface-hub>=0.21",
        "tokenizers<1,>=0.13",
        "av>=11",
        "tqdm",
        "transformers[torch]>=4.23; extra == \"conversion\"",
        "black==23.*; extra == \"dev\"",
        "flake8==6.*; extra == \"dev\"",
        "isort==5.*; extra == \"dev\"",
        "pytest==7.*; extra == \"dev\"",
    ] },
    # Patch qwen-tts to avoid forcing a bundled onnxruntime dependency.
    { name = "qwen-tts", version = "0.0.5", requires-dist = [
        "transformers==4.57.6",
        "accelerate==1.12.0",
        "gradio",
        "librosa",
        "torchaudio",
        "soundfile",
        "sox",
        "einops",
    ] },
]

[tool.uv.sources]
torch = [
    { index = "pytorch-cu128", marker = "sys_platform == 'linux' and extra == 'gpu'" },
    { index = "pytorch-cpu", marker = "sys_platform == 'linux' and extra != 'gpu'" },
]
torchaudio = [
    { index = "pytorch-cu128", marker = "sys_platform == 'linux' and extra == 'gpu'" },
    { index = "pytorch-cpu", marker = "sys_platform == 'linux' and extra != 'gpu'" },
]