By analogy, the fact that FLAC, mp3, and other audio codecs really do achieve a reduction in file size, actually constitutes a rigorous scientific proof of the algorithms' underlying assumption - that audio data (or at least, the kinds of audio data that humans are ever interested in listening to) is at least partially predictable.

The Shannon entropy in a minute of truly random CD-quality white noise approximately equals the size of that .WAV file, but the Shannon entropy of a minute of music is far lower. That's why it can be compressed at all!