They predict from sound and content. They don't always get it right.