So you mean, the AI should figure out when a problematic scene is coming, and only then add labels and whatnot? Not impossible, just somebody must teach them, same with subtitles positioning.