Determining computable scenes in films and their structures using audio-visual memory models