>>16288I use either these two sites for AI covers
Applio:
https://huggingface.co/spaces/IAHispano/ApplioCoverGen_NO_UI colab:
https://colab.research.google.com/drive/1u1brjK8IZt647UsbZuGYfW29oFM2I4tkI mostly use CoverGen since Applio can take forever to output if a bunch of users are using it all at once, but I think Applio can produce better results compared to CoverGen when you get the chance.
I don't know where you're getting the models for the girls' voices, but this is where they are all hosted:
https://huggingface.co/filler5736As far as making it sound better, that mainly falls on the dataset used and messing around with the output settings like the Index Rate/Search Feature Ratio (how much little details are added to the output that the original voice has, but may produce more robotic consonants) or what pitch extration algorithm was used (ie. mangio-crepe or rmvpe).
Anything else is being a bit knowledgable on audio/music production so things like EQ/Compression/Reverb/etc. could help make the AI voices sound less shite.