|
|
|
@ -157,8 +157,8 @@ python pre.py <datasets_root> \
|
|
|
|
|
#### 2.3.3 synthesizer模型训练:
|
|
|
|
|
超参数文件路径:`models/synthesizer/hparams.py`,需将`MockingBird/control/cli/synthesizer_train.py`移成`MockingBird/synthesizer_train.py`结构。
|
|
|
|
|
```shell
|
|
|
|
|
python synthesizer_train.py <name> <datasets_path>
|
|
|
|
|
-m <out_dir>
|
|
|
|
|
python synthesizer_train.py <name> <datasets_path> \
|
|
|
|
|
-m <out_dir>
|
|
|
|
|
```
|
|
|
|
|
其中 `<name>` 是训练产生文件的名称,可自行修改。
|
|
|
|
|
|
|
|
|
@ -170,7 +170,7 @@ python synthesizer_train.py <name> <datasets_path>
|
|
|
|
|
vocoder模型对生成效果影响不大,已预置3款。
|
|
|
|
|
#### 2.4.1 数据预处理
|
|
|
|
|
```shell
|
|
|
|
|
python vocoder_preprocess.py <datasets_root>
|
|
|
|
|
python vocoder_preprocess.py <datasets_root> \
|
|
|
|
|
-m <synthesizer_model_path>
|
|
|
|
|
```
|
|
|
|
|
|
|
|
|
@ -184,11 +184,12 @@ python vocoder_train.py <name> <datasets_root>
|
|
|
|
|
```
|
|
|
|
|
#### 2.4.3 hifigan声码器训练:
|
|
|
|
|
```
|
|
|
|
|
python vocoder_train.py <trainid> <datasets_root> hifigan
|
|
|
|
|
python vocoder_train.py <name> <datasets_root> hifigan
|
|
|
|
|
```
|
|
|
|
|
#### 2.4.4 fregan声码器训练:
|
|
|
|
|
```
|
|
|
|
|
python vocoder_train.py <trainid> <datasets_root> --config config.json fregan
|
|
|
|
|
python vocoder_train.py <name> <datasets_root> \
|
|
|
|
|
--config config.json fregan
|
|
|
|
|
```
|
|
|
|
|
将GAN声码器的训练切换为多GPU模式:修改`GAN`文件夹下`.json`文件中的`num_gpus`参数。
|
|
|
|
|
|
|
|
|
|