I doubt it will be that simple. Audio fingerprinting a la Shazam works by identifying key elements in some transformation of the sound, and setting some threshold for the number of matches. I suppose the output of this compression method offers a good way for identifying a song by a bit of sound, but the identification algorithm will be much more complex.

I'd like to see what happens when you modify the stream. If the representation is really that compact, making small changes should change the output considerably. Treat it like a synthesizer, really. That should tell us something about its usefulness in identifying audio.