pyproject.toml

[project]
name = "funspeech"
version = "1.0.0"
description = "基于FunASR的语音识别和CosyVoice的语音合成API服务"
readme = "README.md"
requires-python = ">=3.11,<3.12"
authors = [
    {name = "Nexa Team"}
]
license = {text = "MIT"}

dependencies = [
    "fastapi==0.115.6",
    "uvicorn[standard]==0.30.0",
    "python-multipart==0.0.6",
    "funasr==1.2.6",
    "librosa==0.9.2",
    "requests==2.32.4",
    "modelscope==1.20.0",
    "soundfile==0.12.1",
    "pydub==0.25.1",
    "websockets==11.0",
    "addict==2.4.0",
    "datasets==4.0.0",
    "wetext",
    "numpy>=1.24.0,<2.0.0",  # Updated for Python 3.12 compatibility
    "scipy>=1.9.0,<2.0.0",  # Updated for Python 3.12 compatibility
    "httpx==0.28.1",
    "torch==2.3.1",  # Required for ASR (FunASR uses PyTorch)
    "torchaudio==2.3.1",  # Required for ASR audio processing
    "setuptools",  # Required for librosa (provides pkg_resources)
]


# TTS dependencies (not needed if only using ASR)
tts = [
    "torch==2.3.1",
    "torchaudio==2.3.1",
    "diffusers>=0.25.0",
    "transformers>=4.30.0",
    "lightning>=2.0.0",
    "conformer==0.3.2",
    "onnx>=1.16.0",
    "onnxruntime-gpu>=1.18.0",
]

[tool.hatch.build.targets.wheel]
packages = ["app"]

[dependency-groups]
dev = []

[tool.uv]
# Index resolution strategy: consider all indexes for best match
index-strategy = "unsafe-best-match"

[tool.uv.sources]
# Use CPU version of PyTorch from PyTorch's CPU wheel index
torch = { index = "pytorch-cpu" }
torchaudio = { index = "pytorch-cpu" }

[[tool.uv.index]]
name = "tsinghua"
url = "https://pypi.tuna.tsinghua.edu.cn/simple"
default = true
explicit = false

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

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"