ttts/train_step.py

'''
1. train bpe tokenizer
2. train vqvae on mel
3. train gpt on mel vq
(3.5). train clvp on text and mel vq
4. train diffusion on gpt latent
'''