If it is actual (local, in-engine) Text-to-Speech, I’d see that as more forgivable. Less space taken up by audio files, better for modding/user-generated content.

Though given the mention of AI and a AA/AAA game I highly suspect they aren’t going that route.