Talking Head Generation with Audio and Speech Related Facial Action Units