Machine listening of music

The analysis and recognition of sounds in complex auditory scenes is a fundamental step towards context-awareness in machines, and thus an enabling technology for applications across multiple domains including robotics, human-computer interaction, surveillance and bioacoustics. In the realm of music, endowing computers with listening and analytical skills can aid the organization and study of large music collections, the creation of music recommendation services and personalized radio streams, the automation of tasks in the recording studio or the development of interactive music systems for performance and composition. In this chapter, we survey common techniques for the automatic recognition of timbral, rhythmic and tonal information from recorded music, and for characterizing the similarities that exist between musical pieces. We explore the assumptions behind these methods and their inherent limitations, and conclude by discussing how current trends in machine learning and signal processing research can shape future developments in the field of machine listening.