Speech and Audio Signal Processing: Processing and Perception of Speech and Music, Second Edition

CHAPTER 38

MUSIC RETRIEVAL

38.1 THE MUSIC RETRIEVAL PROBLEM

Online music sources, such as Apple's iTunes Music Store, provide practically instant access to many millions of music tracks. In the past, choosing music to listen to was a question of flipping through actual media (CDs or LPs), with physical constraints limiting the choice to dozens or at most hundreds. Today, many people carry so much music around with them on portable devices that even just browsing through the cover images could take hours. This growth by several orders of magnitude of the scale of music choices demands entirely new mechanisms for identifying and selecting music – what we are calling the music retrieval problem. This chapter looks at various formulations of this problem, and some current solutions.

Music retrieval shares many aspects with classic text document search (e.g., web searching), but also has some distinctive characteristics owing to the differences between the ultimate purposes of retrieval: listening to music is not so much like reading a document. Music retrieval may be aimed at finding a specific, known item, or alternatively something fitting to the listener's mood or taste. It may be based on a specific query, either in words or by example, or it may be “implicit” (such as activating an internet radio station), in which case the query can be drawn from context such as personal music ...