This repository contains a collection of 15 epoch checkpoints of custom T2S models for use with GPT-SoVITS.

Compared to the stock GPT semantic model, it improves accuracy for English voices, especially with v4.

The model was trained on 1226 examples of Celune's speeches, totaling approximately 2 hours, and allowing the model to understand soft voices, whereas it could not by default.

The model is intended to be used with GPT-SoVITS based projects, and does not run as a standalone model.