While this is fun, trying to decide if a clip is fake or real based on a 1 or 2 second clip is almost impossible unless you use the audible 'clicks' that appear (mostly) on the fake ones. Maybe that is sort of the point but to me basing a yes/no decision on anything less than about 4 seconds is pointless because even if the clip was fake - what possible advantage would a deep-fake message that was only a second long be able to convey? A much fairer comparison for humans vs. AI detection would be at least 10 seconds long.
See how well you can detect state-of-the-art audio deepfakes in this interactive game!