Diff-VITS(WIP)

VITS with diffusion for zero shot tts.

Dataset

You should put your dataset in the dataset folder, and the dataset should be organized as follows:

dataset
├── p225
│   ├── p225_001.wav
|   ├── P225_001.text
│   ├── p225_002.wav
|   ├── P225_002.text
│   ├── ...
├── p226
│   ├── p226_001.wav
|   ├── P226_001.text
│   ├── p226_002.wav
|   ├── P226_002.text
│   ├── ...

and processed dataset will be saved in the processed_dataset folder under the same folder as the dataset folder.

Data preprocessing

Put the textgird files in the dataset folder, and then run the following command to preprocess the data.

python preprocess.py

Requirements

Use the following command to initialize the env and install the requirements.

bash init.sh

And env named vocos will be created.

Training

Run accelerate config to generate the config file, and then train the model.

Multilingual is in test, now only support mandarin.

accelerate launch train.py

Inference

Run python tts_infer.py for inference. You should change the text and model_path in the tts_infer.py file before running the command.

Q&A

qq group:801645314 You can add the qq group to discuss the project.

Thanks to sovits4, naturalspeech2 and imagen for their great works.