We can contribute to a Cantonese G2P, please let us know how to proceed

[laubonghaudoi]

We maintain a state-of-the-art Characters-to-Jyutping library https://github.com/CanCLID/ToJyutping and we are interested in adding Cantonese to Kokoro. We can also offer 100+ hours of Cantonese training data. Please let us know how to contribute. Thank you!

[hexgrad]

I just wrote a Wiki page for this: https://github.com/hexgrad/misaki/wiki/Contributing

For Cantonese it would be yue.py

[laubonghaudoi]

Thank you! It looks quite straightforward and simple. We will send PRs later. We want to know about data requirements too. Do you need us to provide Cantonese TTS training data as well?

[JakubCzarlinski]

@laubonghaudoi This might be a good starting point: ASR datasets for Cantonese https://arxiv.org/abs/2201.02419

FYI: I have no idea regarding the state of the datasets - how clean the data is, suitability for TTS, etc...