This is just the tip of iceberg (spear?). A friend of mine is doing research in this area and what they are working on doing is distilling each song to its essence (as far as I understand) based on a few set parameters such as rhythm, etc. and developing a code to represent it. In this way, if presented with the original "night and day" and U2's remake of the same, the database would be able to understand that they are the same song and code them identically, even though they sound quite different.
This opens up all sorts of possibilities particularly for music customization. Imagine searching a database using one of your favorite songs as the search term. It would be coded in the way mentioned above and then all the songs with a similar coding would be shown to you, under the assumption that there is something in the basic rhythm of the song that corresponds to your musical taste. Of course, musical preference is much more complicated and this and it might turn out that you hate all of the "hits" turned up by such a search, but its an interesting line of research.