# CosyVoice2-0.5B:本地详细部署教程(Windows)

0 阅读3分钟

CosyVoice2-0.5B:本地详细部署教程(Windows)

1.克隆项目

确保你的设备上安装了Git Git - Downloads

git clone --recursive https://github.com/FunAudioLLM/CosyVoice.git
cd CosyVoice
git submodule update --init --recursive

或者直接在github上面下载压缩包解压
FunAudioLLM/CosyVoice: Multi-lingual large voice generation model, providing inference, training and deployment full-stack ability.

2. 创建虚拟环境

安装 Conda  Download Now | Anaconda

安装后,在win开始菜单里面找到 Anaconda Prompt 并打开,此时你的终端应该是 :

Pasted image 20260329085558.png 注意此时是base

conda create -n cosyvoice python=3.10
conda activate cosyvoice
conda install -y -c conda-forge pynini==2.1.5

注意在执行 conda activate cosyvoice 后,确保你的终端 (base)变成 (cosyvoice)再执行下一步安装,否则会安装在系统Python中。

3.安装依赖库

进入你下载CosyVoice的文件夹,然后输入以下命令:

pip install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/ --trusted-host=mirrors.aliyun.com

4.下载模型

CosyVoice语音生成大模型2.0-0.5B · 模型库 如下图:

Pasted image 20260329085930.png

5. 运行模型

在这个CosyVoice文件夹下面运行python webui.py即可

6.参考资料

CosyVoice2-0.5B:本地详细部署教程(Windows)_cosyvoice2本地部署-CSDN博客

CosyVoice2-0.5B本地部署!语音生成,3秒极速语音复刻!支持消费级显卡!新手友好!含可视化页面_哔哩哔哩_bilibili

7. 安装出现的问题

6.1

Q:

ModuleNotFoundError: No module named 'pkg_resources' [end of output] note: This error originates from a subprocess, and is likely not a problem with pip. ERROR: Failed to build 'openai-whisper' when getting requirements to build wheel

这个报错的根因不是你命令写错了,而是旧版 openai-whisper==20231117 的打包脚本和新版 setuptools 冲突了。参考的那篇博客发布于 2025-05-22,但 setuptools 82.0.0 在 2026-02-08 已移除了 pkg_resources;而 CosyVoice 官方 requirements.txt 现在仍固定了 openai-whisper==20231117,这个版本在 PyPI 上又只有源码包,所以 pip 会在本地构建时卡在 Getting requirements to build wheel,最后报 No module named 'pkg_resources'

改为运行以下指令:

conda activate cosyvoice
python -m pip install --upgrade pip
python -m pip install "setuptools==80.9.0" wheel
python -m pip install --no-build-isolation "openai-whisper==20231117" -i https://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com
python -m pip install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com

如果最后一条还重新去构建 openai-whisper,就改成:

python -m pip install --no-build-isolation -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com
6.2

Q:
安装模型时,可以写一个download_model.py文件,然后运行文件进行下载模型。

from modelscope import snapshot_download

snapshot_download(
    'iic/CosyVoice2-0.5B',
    local_dir='pretrained_models/CosyVoice2-0.5B'
)

要进入CosyVoice目录下

conda activate cosyvoice
cd "C:\Program Files\CosyVoice"
python download_model.py

或者直接运行指令也行

conda activate cosyvoice
# 进入你自己的CosyVoice目录下
cd "C:\Program Files\CosyVoice"
python -c "from modelscope import snapshot_download; snapshot_download('iic/CosyVoice2-0.5B', local_dir='pretrained_models/CosyVoice2-0.5B')"

然后再运行

python webui.py --port 50000 --model_dir pretrained_models/CosyVoice2-0.5B

在浏览器里面打开 http://127.0.0.1:50000 就可以了

一定要选择流式输出,要不然生成语音太慢了

webui.py第 132 行附近把:
这个

stream = gr.Radio(choices=stream_mode_list, label='是否流式推理', value=stream_mode_list[0][1])

改为

stream = gr.Radio(choices=stream_mode_list, label='是否流式推理', value=stream_mode_list[1][1])

就可以改为默认流式输出

6.3

Q:
我的电脑显卡是NVIDIA 5070,好像PyTorch 有些低版本不适配,那就使用以下指令禁用GPU,用CPU跑

set CUDA_VISIBLE_DEVICES=-1
python -u -c "import torch, runpy, sys; sys.argv=['webui.py','--port','50001','--model_dir',r'C:\Users\15831\CosyVoiceModels\CosyVoice2-0.5B']; runpy.run_path('webui.py', run_name='__main__')"

 set CUDA_VISIBLE_DEVICES=-1,这会直接禁用 GPU
注意 这里 C:\Users\15831\CosyVoiceModels\CosyVoice2-0.5B'];用你自己下载模型的位置