>>36780
>>36781
I think the problem is the ai works better with some source audio than others, and when it struggles it defaults to its pre-trained voices instead of the one you're trying to train it on