Open Naozumi520 opened 1 year ago
I am a Cantonese. I will focus on it. Training data is the biggest problem right now.
Ah yes.... Can we use Common Voice? I used sovits to convert the dataset into a single voice for vits.
It sounds okay, but without bert it's not very natural
https://github.com/wenet-e2e/wetts/assets/52615455/4a851a3d-aa5a-4ae9-9aca-67659267beb6
Is your feature request related to a problem? Please describe. I'm always frustrated to see that the Chinese dialect "Cantonese" doesn't get enough attention. As a HongKonger, Cantonese is the language that I speak everyday. However, there are not much resources of it. I know VITS_Chinese did support Cantonese. However, the result are not very good. I would like to see if
wetts
can support this dialect with 85.5 million of speakers.Describe the solution you'd like TTS that support Cantonese, also with bert.
Describe alternatives you've considered There are no alternatives.... VITS_Chinese and PaddleSpeech is the only one. However, as I said before, the result is not very good.
Additional context https://huggingface.co/indiejoseph/bert-base-cantonese https://github.com/yeyupiaoling/VITS-Pytorch/blob/master/mvits/text/cantonese.py https://github.com/PaddlePaddle/PaddleSpeech/issues/2669