even when it makes mistakes they're usually consistent with the speech so the ai won't mind