Editors' Notes

Given the tremendous growth of music-related data available digitally, both locally and remotely, there is a pressing need to find effective ways to organize and structure such data, and to provide tools that efficiently search these repositories, in particular by direct musical content. To accomplish this, music representation needs to be multidimensional; audio data, which can be voluminous, require particular care in storage and transmission while preserving quality; descriptive information that draws upon acoustic, musicological, and cultural features is needed to specify what is musically significant; and complex intellectual property rights must be addressed.

This is the impetus for the international conference series on music information retrieval, now in its fifth year, which brings together practitioners (both academic and industry-based) from many disciplines to exchange ideas. The program consists of invited talks and peer-reviewed papers and posters. Authors come from a wide range of fields, including archival and library science, computer science, engineering, jurisprudence, musicology, and psychology; their work draws heavily from concepts and methods from various areas within these fields, such as information retrieval metrics from library science, knowledge representation and machine learning techniques from artificial intelligence, signal-processing techniques from audio engineering, and models of music representation and perception from musicology and music cognition. To help foster interdisciplinary discussions, a range of tutorials are provided to introduce participants to terminology and concepts from other fields.

In October 2003, the Fourth International Conference on Music Information Retrieval, ISMIR 2003, was co-hosted by the Sheridan Libraries of Johns Hopkins University and the Library of Congress and held in Washington, DC, USA. In consultation with members from the ISMIR 2003 program committee, the authors of eight papers presented at the conference were invited to submit revised and extended versions of their work for potential publication in this special edition of Computer Music Journal. All authors accepted our invitation, and after another round of peer reviewing, five of the eight articles were selected for publication in this special issue, and—owing to a lack of space—the remainder will appear in the subsequent issue of Computer Music Journal.

The ISMIR conference series grew out of a symposium, envisaged and organized by Donald Byrd and J. Stephen Downie, that not only gave rise to the now slightly anachronistic acronym ISMIR, but also provided the starting point for this exciting and highly successful series of events. It is fitting, therefore, that the first article of this special edition be authored by J. Stephen Downie, one of the founding figures of the conference series, and that it is a forward-looking article that explores ways to develop a more rigorous framework for evaluation of music information retrieval systems. The article, "The Scientific Evaluation of Music Information Retrieval Systems: Foundations and Future," provides an introduction to the topic of music information retrieval (MIR) before discussing the many layers of complexity that surround the evaluation of MIR systems and techniques.

Napster put peer-to-peer networks on the map, in part due to its legal ramifications. It demonstrated the power of this means of access to digital audio even though retrieval was based around crude keyword searching. In "A Scalable Peer-to-Peer System for Music Information Retrieval" by George Tzanetakis, Jun Gao, and Peter Steenkiste, the authors describe an alternative way to organize such a topology that supports audio-content-based retrieval in addition to textual attributes and that displays better characteristics when scaled to large collections with many peers.

An archetypal example of music information retrieval is someone trying to locate a song based on a possibly faulty memory of a melodic fragment. It is an experience many people can identify with, and in recent years numerous systems have been developed to support this kind of information-seeking task. In "The MUSART Testbed for Query-by Humming Evaluation" by Roger Dannenberg et al., three such systems—each founded on a different principle—are studied and compared.

"Exploring Music Collections by Browsing Different Views" by Elias Pampalk, Simon Dixon, and Gerhard Widmer offers an alternative way to locate music. Recognizing that not all music retrieval tasks start with a formulated query of notes, the...