With help from the US Air Force Research Laboratory, the inventors found they could bury around 20kb of speech data in a song without affecting how it sounds. Their technique exploits the fact that the tones of a musical instrument are made up of a complex pattern of randomly phased harmonics. These phase shifts are so small they are imperceptible to the ear. But a software decoder, which could be built into MP3 players or file-sharing applications, detects the phase shifts and turns them into speech.
Weird huh? Remember that the example I gave above could very well be false, as I couldn't find more details. I thought I'd write it to spark some interesting comments from you guys however, so come on then, hit the Post Your Comment button ;-).
Source:
New Scientist