Finetuning YourTTS

[e0xextazy]

Has anyone been able to fine tune the YourTTS model to the new speaker? I took experiment 4 as a starting point (since it saw the most speakers) and am trying to finish training only on my dataset, without the admixture of others. I also disabled validation because of the small training set. Made about 20 different experiments, changing parameters, but all of them are unsuccessful. I will attach my config here. Please help more experienced colleagues to understand my question. Thanks in advance!
config_default.txt

P.S. If more information on training is needed to resolve my question, I will provide it.

[e0xextazy]

I run the code with the command:
CUDA_VISIBLE_DEVICES="0" python TTS/TTS/bin/train_tts.py --config_path experiments/ft_4/config.json --restore_path experiments/ft_4/best_model_latest.pth.tar