vits

VITS TTS model fitted for lithuanian language.

Installing

This project uses conda for virtual environment and package management.

Creating conda environment

1. conda env create -f environment.yml
2. conda activate vits

Training

Training required GPU with CUDA support!

Download the training dataset, then rename or create a link to the dataset folder: ln -s files/datasets/atrinkti_sak_3/wav DUMMY1

Run the following command to start training:

python train.py -c files/configs/lt.json -m lithuanian

Dataset

Dataset is not added in this repository since it is quite big and github charges for LFS usage if reached certain size treshold

Inference

To test trained model you can use inference.ipynb notebook in notebooks directory. To use trained model parameters replace load_checkpoint with trained model path from logs directory