Reviewer biography

Current Reviews

Review Articles

Book Reviews Archive

Music Query: Methods, Models, and User Studies

by Walter B. Hewlett and Eleanor Selfridge-Field, Eds.
Published by Center for Computer Assisted Research in the Humanities and The MIT Press, Cambridge, MA, 2005
Paper, $35
ISBN: 0-262-58256-2.

Reviewed by Joao Pedro Martins, Marcelo Gimenes and Qijun Zhang
Future Music Lab, University of Plymouth, UK

This is the third book by Walter B. Hewlett and Eleanor Selfridge-Field, both researchers at the University of Stanford, with the stamp of the same publisher. It introduces a number of articles by researchers around the world who are in the lead of music information retrieval, one of the currently most active fields in music technology.

An equivalent interest began to happen not long ago with the overwhelming growth in the availability of information through the Internet. Search engines have been created to enable the access to information with advanced word query methods.

Music entails the same sort of challenge. What if one is capable of singing or perfectly whistling a piece of music but can’t recall its title or composer? Furthermore, what if this individual remembers only parts of the music? How should it be possible to find out a specific tune among millions of others in a growing database?

This is where Music Query: Methods, Models, and User Studies comes to provide several clues on how the above tasks could be improved. The subject of music query methods is not only confined to web applications, though. They also play an important role in musicology by establishing links in the evolution of musical styles, and also by enhancing copyright protection mechanisms. The current trend is to establish distance measures that take into account cognitive processes, in order to better capture human perception and creativity.

The book starts with a chapter by Vlora Arifi and her colleagues from the University of Bonn, addressing the problem of synchronisation of streams of musical information. Synchronisation of different representation formats is pointed as being a promising possibility to enhance the performance of music retrieval systems. The idea is to link the representations of the same piece of music in a large database with the help of a well-defined cost function.

The analysis of rhythmic structures with machine learning techniques is the subject of the next chapter. Tillman Weyde presents his Integrated Segmentation and Similarity Model, which uses fuzzy logic rules to rate alternatives for rhythmic structures. This model shows a great concern for the inclusion of perceptual features when defining the fuzzy rules.

Another major issue addressed by music information retrieval systems is the special role played by melody on the recognition of songs and musical styles. Wei Chai, in chapter three, uses folk songs from different countries to illustrate how music styles can be distinguished by statistical features described by Hidden Markov Models. Several string-matching methods are proposed which combine both pitch and rhythm information. It also shows the importance of designing melodic representation according to different tasks.

In chapter four, Olivier Lartillot and Emmanuel Saint-James describe a new approach to musical-pattern discovery that consider the musical discourse as complex flows of smaller structures. Their proposal was implemented with OMKanthus, a library based on IRCAM1s Open Music, modelling cognitive mechanisms characteristic of music perception.

In the next chapter, Eleanor Selfridge-Field, proposes a preliminary distance melodic metric based on cognitive principles. Distance assessments take into account pitch and harmonic conformance in relation to metrical and accentual information.

Chapter six, by Frans Wiering, Rainer Typke and Remco C.Veltkamp, introduces the approach to use weighted dots for music-notation retrieval. Two transportation distance measures are defined and compared. According to the experiments with monophonic incipits, this approach has achieved higher performance, compared with similar researches.

In chapter eight, Daniel Muellensiefen and Klaus Frieler propose a research paradigm or optimal melodic similarity measure from the comparison of the results taken from a multitude of algorithms and from experiments with human music experts. Several techniques are considered, depending on the intended task and data to be analysed.

Finally, in chapter seven, Micheline Lesaffre, Dirk Moelants, and Marc Leman report psychological experiments on query type preferences of the population. The key issue of the experiment was not to initially constrain the choices of the subjects. All the steps of the research are very well documented and the results show that some widely used methods, like query-by-humming, received a lower share of preference.

This book is not only designed for those interested in music information retrieval systems. Anyone who intends to carry out research on music technology will certainly benefit from its reading.



Updated 1st June 2005

Contact LDR: ldr@leonardo.org

Contact Leonardo: isast@leonardo.info

copyright © 2004 ISAST