By François Pachet.
Communications of the ACM 46(4), 71–75, April 2003. DOI 10.1145/641205.641207.
Read the paper (PDF)
ACM article (DOI)
About the paper
How can listeners find music in a large digital catalogue? This article examines the metadata needed to identify recordings, compare them, and build sequences for listening. It describes two Sony CSL prototypes that put these ideas into practice.
Abstract summary
The article compares what can be learned from the audio signal with what can be learned from external sources such as listener ratings and usage data. It uses recording identification, genre, and similarity to show where the methods can complement each other.
PathBuilder takes two tracks chosen by the listener and builds a playlist between them using similarities in genre, voice type, tempo, and other attributes. PersonalRadio gives listeners a slider that moves from familiar choices toward more exploratory ones.
Related work
The Active Listening project includes the PathBuilder and PersonalRadio examples. The Playlist Generation project traces the underlying work on constructing sequences of tracks.