Some models I finetuned a long time ago using Google Colab and the notebooks on PiperTTS's github repo. They were made by using a voice sample from Kemono Friends nexon, putting that into GPT-SoVITS to make an english dataset, then using that to finetune a default model from PiperTTS. They worked well for AlwaysReddy and actually sounded like the character they were replicating.