嘴型融合wav2lip基础版和高清版安装

#### 1.下载安装conda，https://www.anaconda.com/

#### 2.创建运行环境

```
conda create -n wav2lip python=3.7
```

Java

```
conda activate wav2lip
```

Python

#### 3.在电脑磁盘为wav2lip新建一个目录

#### 4.[下载](https://ffmpeg.org/ "下载")ffmpeg并配置变量

#### 5.[下载](https://developer.nvidia.com/cuda-downloads "下载")并安装CUDA，[安装教程](https://blog.csdn.net/qq_43497966/article/details/131220381 "安装教程")

![](https://www.sz126.net/upload/attach/202310/1_GE8SXZ4BWHH4UMB.png)

#### 6.安装pytorch(根据显示查询对应的CUDA版本进行下载)

```
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
```

#### 7.下载运行程序

```
git clone https://github.com/Rudrabha/Wav2Lip.git
```

#### 8.安装环境依赖

```
pip install -r requirements.txt
```

#### 9.准备模型，如wav2lip\_gan.pth或wav2lip.pth

#### 10.运行命令

```
python inference.py --checkpoint_path wav2lip_gan.pth --face test.jpg --audio test.wav --outfile output.mp4
```
### 报错解决
#### 【debug】TypeError: mel() takes 0 positional arguments but 2 positional arguments (and 3 keyword解决

经查是librosa版本问题。原版本0.9.0，改成0.7.1后，问题解决

```python
pip uninstall librosa
```
```python
pip install librosa==0.7.1
```

#### No module named 'numba.decorators'
    
    找到安装的librosa的库中的decorators.py文件，一般在这里.\\Lib\\site-packages\\librosa\\util\\decorators.py，然后注释对应部分
    
    ![](upload/attach/202310/1_N4RBCUPBW8X55KM.png)