[bagels]

There are real world effects in the real videos that are completely absent in the fake ones that give them away.

The fakes have completely clean audio with similar artificial noise (real ones have crowd, or other backround noise), the fakes have no microphone or environmental artifacts (echo/reverb, plosives, others).

What made them even easier to recognize is that accents were all wrong. They were exactly as another person said, caricatures of the original voices. I actually found that they were so obvious that most could be identified before the second or two that they make you wait to click the submit button.

If it were a foreign language, I think the success rate would be lower, but the unnatural lack of noise gives them away.