My intuition is that training on audio will be trivial if they can accomplish this for video. Maybe I'm wrong.