Guest editorial: Machine learning in and for music

Music has always held a special fascination for researchers in mathematics and computer science. It is both an abstract, formal system and a physical event; it relates to areas of human endeavour as diverse as the arts, aesthetics, philosophy, mathematics, acoustics, physiology, or history; and it poses deep questions regarding human nature, intellect, and emotions. Artificial Intelligence, in particular, has dreamed of devising intelligent musical machines almost from its inception. Music has also played its role in the history of machine learning research. Initially, musical tasks were mostly used as toy domains to demonstrate the properties of new learning algorithms, and the musical problems addressed were usually greatly simplified (the guest editor’s own early research is no exception). But recent years have seen a dramatic change in this regard. Machine learning researchers have begun to address musical tasks of real-world complexity—from the analysis of the artistic style of famous musicians to systems that learn to accompany, or improvise with, human musicians, and from programs that classify audio recordings into musical genres to computers that can recognise artists from the way they play. This trend towards ‘real-world’ applications has been accelerated by the dramatic changes that are currently going on in the digital music market. With the Internet becoming the central medium for the distribution of music and Web-based music stores offering literally millions of digital music tracks, there is a pressing need for things like intelligent and adaptive music search and recommendation systems, automatic style recognition and music classification, etc. The new field of Music Information Retrieval (MIR) is a direct response to this need, and the research published at the International Conference on Music Information Retrieval (ISMIR) and similar events is closely watched by major players in the music industry. The potential of this kind of research is also being recognised by the large international research funding organisations. At the European level, for instance, several large-scale research efforts in the area of intelligent computing and learning in music and audio have recently been funded—e.g., the projects SIMAC (www.semanticaudio.org), Semantic HiFi (http://shf.ircam.fr), and S2S2 (http://www.s2s2.org).