Bibliography (3):

1. ByT5: Towards a token-free future with pre-trained byte-to-byte models

2. mT5: A massively multilingual pre-trained text-to-text transformer

3. https://github.com/lingjzhu/CharsiuG2P