Oh come on, this is an easy one. They need to include the complete audio channel in their training model, and identify important audio cues.<p>99% of Homer's characterization is really Dan Castellaneta's voice. The rest is simply the basic stretch-and-squash rubber band concept, as applied to the bug-eyed, yellow Simpsons visual character model.<p><a href="https://en.wikipedia.org/wiki/Rubber_hose_animation" rel="nofollow">https://en.wikipedia.org/wiki/Rubber_hose_animation</a><p>This is not an image processing task.