xiaomaiiwn

/

vits-cantonese

like 4

I trained with this script
https://github.com/CjangCjengh/vits

I see. I’m able to load the D.pth but not the G.pth

And also when I run inference with input text 我愛你, 愛 is skipped in the output audio, is it becoz it didn’t appear in the training data?

And also when I run inference with input text 我愛你, 愛 is skipped in the output audio, is it becoz it didn’t appear in the training data?

There are no traditional characters in the dictionary. Can you try using simplified characters

I see. Could you share which dataset you used to train it? And how long it took

There are two methods:
Option 1: You can collect Cantonese dialects from videos or create your own voice dataset
Option 2: Search for your dataset on the website
The second method I use here is website address: https://magichub.com

ok thanks! I have added your model to
https://github.com/Keith-Hon/vits-cantonese
is it ok?

Ok, currently there is only one type of voice, and we will continue to train multiple types of voice in the future.

Thanks. For one voice, How much data did you use for training and how long does it take? I

Over 1000 audio tracks, training time 2-3 days

how long is each audio track? the training time seems a bit long. you trained with a single low spec gpu or becoz the audio tracks are too long?

how long is each audio track? the training time seems a bit long. you trained with a single low spec gpu or becoz the audio tracks are too long?

1 audio for 1-3 seconds, trained with t4

ok got it