A surprisingly large number of people in the computing field are also skilled musicians; this book will certainly appeal to them. However, this is not a volume that only interests the amateur. Important applications of the analysis of music are many: intelligent music retrieval, detection of plagiarism, understanding of composition, and other applications in musicology are just a few of the uses for computational tools for the analysis of music. It must be noted that music analysis is not a new field and excellent work, by hand, has been going for more than a century; many of the references in this book are from these earlier efforts. This volume is a collection of current papers that gives a good introduction to the current state of the field, popular techniques, and key challenges.
The first chapter, “Music Analysis by Computer: Ontology and Epistemology,” initially seemed a surprising choice for a technical volume; it sounds more suited for a philosophical treatise. In fact, it is a vital start to the book, as it sets out clearly the purpose, objectives, and assumptions of the field of computational music analysis. As one example, a question that is considered here is whether a “piece” includes characteristics of a given specific performance, such as inflections and precise timing--in other words, whether a musical piece is a repeatable entity or a short-lived phenomenon. The value added by this chapter is high, and it makes one wonder why such overarching essays are not part of other books on application areas of computing technology.
After this fundamental chapter, the rest of the book is organized into sections on chords and pitch operations, parsing structures, grammars and hierarchy, motivic and thematic analysis, and classification and distinctive patterns. Much analysis happens on a representation known as a musical surface: a sequence of pitch events that can be thought of as a rendition of the written musical score. This is part of a hierarchy of representation; other levels in the hierarchy can be considered to discover other important aspects of a musical piece. Moving from the musical surface to higher levels of representation requires parsing and segmentation, in which individual note events are grouped and segmented into musical phrases. Typically, most notes are not sounded and perceived individually but rather as part of simultaneous groups known as chords. One chapter describes different representation schemes for chords, as the recognition of chords depends on which intervals between pitches are regarded as consonant or dissonant. Another chapter presents several interesting representations of a chord complex, which is a multidimensional sequence of chords providing visualization of chord progressions. A chapter titled “Contextual Set-Class Analysis” directly addresses the hierarchical representation of a piece through defining systematic operations on the musical surface and interpreting the results at several levels.
In the section on parsing, the extraction of musical forms (simply, “verse-chorus-verse-chorus-chorus”) is addressed. Automated recognition of these forms is done by segmentation and analysis of results and is informed by the fact that long-recognized musical forms exist. Using widely available data sets of musical surfaces, segmentation and classification results may be compared and evaluated. Another interesting problem is the division of the musical surface into distinct notes and voices by a machine learning algorithm.
Several papers explore the analysis of music as an instance of language. “Analysing Symbolic Music with Probabilistic Grammars” applies probabilistic grammar models implemented in PRISM and compares them to Markov models, based on the efficiency of representation. Another chapter explores interactive strategies where an expert in music analysis works closely with a computational system to determine structure and properties; the important aspect of incorporating user feedback is essential here.
Other chapters explore: motivic analysis, identifying and understanding repeated passages called motives, through a multidimensional analysis using tries; pattern discovery through analysis using wavelets (the Haar wavelet is used here, but other wavelets may be promising); analysis of collections of points in a space defined by pitch and time, called “point-set” representation; and composer recognition through the application of support vector machines to features extracted from the musical surface. A final paper looks at entire corpora of music from specific composers to understand possible instances of reuse or borrowing of musical elements; this approach operates on point-sets in a geometric manner the authors call symbolic fingerprinting.
This book is well organized, deep, sound in methodology, challenging, and fascinating. It should be accessible for those with knowledge in both areas: computational methods and music. I have some concern that readers with limited knowledge in one area may find it difficult to read and understand, through no fault of the authors. Still, for its intended audience, which will have competence in both areas, it is a must-read.
