.emusic



05.09.03 Algorithmic Clustering of Music, to compress=to know.


Tell me how you compress and I’ll tell you who you are. This is the apparent result of the research of Rudi Cilibrasi, Paul Vitanyi and Ronald De Wold, of the Dutch National Research Institure of Amsterdam. In ‘Algorithmic Clustering of Music‘, in fact, the three authors illustrate a pragmatic method for automatically telling the genre and author of a musical piece. The technique is similar to the one already used for texts. In this case it’s possible to tell the language a text is written in by comparing it to other texts written in different languages. The technique used is to compress every file separately, then compress them again, this time adding to each one the file whose language must be guessed; the archive whose size has changed less after the second compression is the right one. The algorithm used is zip, which is particularly efficient in reducing similar text sqeuences. The researchers have used the bzip2 standard to compress music pieces by Beethoven, Miles Davis and Jimi Hendrix, stripping everything but the rhythm and melody. The result is a tree structure which can precisely distinguish genres and authors. One of the possible uses of this technology is to find out the author of an anonymous piece by comparing it to other pieces whose author is known.