The algorithm more or less “listens” (there are a lot of different ways to do this, but it works similarly to how Cortana or Siri work) to the file and then matches what it “hears” to the sound of words it knows.
It seems to swing towards political and news stuff, so those are probably things it hears a lot so it favors those.
Off topic, but how on Earth this kind of algorithm helps anyone? If you were deaf you’d be better off just guessing by yourself. It’s just a ridicilous thought that someone has actully published the entire system and been serious.
It might not help anyone directly, though it can be used as training for future development.