An AI voice model can only be as good as the audio it learns from. Give it a clean, controlled recording and you get a model that works in a cinema mix. Give it a session with an air conditioner humming underneath, and you may end up with a voice model that sounds like it has strong opinions about indoor climate control.