I bet it's a Self-Organizing Map (Score 1)

I bet it uses one of the Self-Organizing Maps variants to identify the sound snip. You may be able to fool this algorithm by injecting sound frequences in your upload that are too high to hear for people, but is still represented digitally. Of course this depends on *how* you inject the sounds as not all microphones would pick it up, but if you just edited it in a studio or patch program, no problem.