Browse Books

Reviewer: Soubhik Chakraborty

Music is both a science and an art. All of its different dimensions, whether melody, rhythm, timbre, mood, harmony, or counterpoint, are of interest to users of music information retrieval (MIR) systems. In addition, users have a large reservoir of music collections from which to make a selection. This book's aim is to popularize the research trends in the area of MIR. The book is divided into four sections, covering MIR methods and platforms, harmony, music similarity, and content-based identification and retrieval. The specific topics covered include: indexing polyphonic music; symbolic representation of rhythm; spectral and cepstral features; chord and tonal harmony analysis; statistical modeling of music; harmonic and percussive sound separation; properties of violin modes, and a mathematical model for violin sound; automatic indexing of music by emotions, including a chapter on Hindustani vocal music; song identification; song similarity on the basis of "text statistic features"; melodic segmentation in MIR; and the role of the Artificial Immune Recognition System (AIRS) to classify music genres from various cultures. I strongly recommend this gem of a book for scientific libraries. My personal view, however, is that musical emotion is very complex, and is different from common human emotions such as happiness or sadness. Generally classifying two musical pieces as "happy" or "sad" is like saying that a mango is sweet (or sour), and so is an orange. But how is a mango different from an orange__?__ Perhaps it would be helpful to consider diversity in emotion, and not just commonality. Online Computing Reviews Service

Reviewer: Pierre Jouvelot

Given the importance of music in society and the economy at large, finding new, efficient, and intuitive ways of managing the wealth of audio content available, either online via the Web or offline in public or private audio libraries, is a very active area of research. If searching text databases can be argued to be already a somewhat mature field-although some would beg to differ, given the paucity of semantic analysis usually encoded within even well-known search algorithms-the domain of music information retrieval is clearly still in its infancy. Even though it is relatively easy for humans to recognize and analyze music pieces, computers seem to have a hard time acquiring these skills; yet, given the practical importance of such a capability, more effort is clearly warranted. Ras and Wieczorkowska have compiled a 17-paper book that gives readers a glimpse into some of the current research approaches to such a challenge. The book is structured into four parts. The five papers in Part 1, "Music Information Retrieval," introduce new ways to advance the state of the art in music representation formalisms. They discuss structuring methodologies for building meaningful indexes and test platforms where the relevance of these techniques can be assessed. Typical representations of (digital) music range from time-stamped audio samples, to frequency spectral and cepstral features, to more symbolic, score-like representations such as the musical instrument digital interface (MIDI). Mid-level representation formats are introduced to provide a bridge between raw data and symbols such as notes and chords. Finally, there is a historical assessment of MIREX, an evaluation exchange platform where researchers may quantitatively assess how their MIR search algorithms, built upon such internal representations, compare to others. The following three papers make up Part 2, which is dedicated to automatic chord analysis. Harmony, the art of chord progression, is a key facet of music practice and understanding, which can be put to good use to enhance information retrieval. The harmonic structure of musical pieces can be uncovered, at least in part, using various techniques, such as voice tracking, perceptrons, and n -grams. Part 3 goes beyond music representation and harmonic structure to focus on the content itself. Its papers address wide-ranging issues: modeling music data via hidden Markov models; distinguishing in real time harmonic sounds from percussive ones; automatically assessing the quality of violin playing; and indexing music via the emotions it may spur, both for Western and Hindustani music. A natural technique to enable music retrieval is to look for similarities between existing music pieces, an approach addressed in Part 4. Such similarities are present in cover songs-new recordings of existing pieces-with varied performance characteristics, such as tempo and orchestration. Apart from providing a survey of existing techniques to determine the presence of similarities, this part also includes contributions that suggest the use of text lyrics statistics, melodic segmentation, and classifiers based on artificial immune recognition systems, to reach such a goal. A rather short glossary of basic music terms closes the volume. Unfortunately, there is no index. The book gathers the contributions of almost 50 authors. As is usually the case when there are so many authors, the quality of the writing and presentation varies widely, as do the subjects addressed and their scope, from surveys to narrowly focused problems. Also, given the editors' affiliation-they are members of the Warsaw-based Polish-Japanese Institute of Information Technology-some of the papers involve, not surprisingly, members of their institution. Despite these caveats, the book's information, be it of a survey or research nature, will provide readers who are interested in music information retrieval with a fair description of the state of the art of this economically important research domain. Online Computing Reviews Service