Virtual Music: Computer Synthesis of Musical Style

Rating:
Summary: Not much difference from previous books by the author
Review: Mr Cope's work is fantastic. But as a reader of his previous books - from another publisher (A-R Editions) - I was rather disappointed with this one: it does not add much new content to the subject. If you have his other books, there is no point in buying this one.

Rating:
Summary: A fascinating overview of machine musicanship
Review: This book gives a fascinating overview of machine music as seen through the eyes of the author, who has been actively involved in this field for many years. The reading of this book is recommended for anyone who is interested in the extent of machine musicianship and musical creativity. Not being an expert in music theory should not dissuade one from its perusal. In fact, not possessing extensive knowledge of musical theory may be an advantage, in that one can read the passages and listen to the musical compositions on the accompanying CD with minimal bias as to what constitutes enjoyable or "good" music. Indeed, if one were to approach musical listening, musical composition, and music theory from the standpoint of being exposed only to 'virtual music', what would one then think of music composed solely by humans? Would one then judge "machine music" to be better than "human music"? As virtual music becomes more integrated into entire knowledge base of music, as it will in this century, there will be many who will be exposed to it more often than human-composed music. It will come to be accepted as beautiful music to listen to, and debates as to its "authenticity" will disappear. Even more interesting is the question as to what musical preferences the machines themselves will have. Will they debate among themselves about music theory and what constitutes compositional excellence? It will be interesting to see what kinds of music theory are generated (or preferred) by these machines, and if they are as biased to certain forms of music as their human musician counterparts frequently are.

The author characterizes 'virtual music' as being a category of machine-created composition that attempts to replicate the style of existing music. He points out that virtual music has predated the advent of computing machines, but that these machines have allowed, at a much larger scale, the composition of music in a pre-selected style. Early instances of virtual music discussed in the book include the 'figured bass' of the Baroque period, wherein composers could produce music that adhered to a particular composer style but would still be original. Another (fascinating) example is the 'Musikalisches Wurfelspiel', or "musical dice game" of the eighteenth-century. To obtain a composition, one constructs a matrix, with rows representing the results of the throw of dice, and the columns representing successive measures of music. To get a measure of music, the dice are tossed, and from the result in the correct row the measure is obtained from the column. Also mentioned is the work of Iannis Xenadis, which uses mathematical models to compose music, and the work of Kemal Ebcioglu, which uses first-order predicate calculus to create chorales in the style of J.S. Bach. In addition, the author mentions the work of Dominik Hornel and Wolfram Menzel, who make use of neural networks to create music with stylistic similarities to composers of the Renaissance and Baroque periods.

Early in the book, the author introduces what he calls "The Game" in order to warm the reader up to his study of virtual music. This game requires the reader to identify styles and composers of various examples of music, the notations of which are included in the book, and the actual music on the accompanying book. One of the objects of "The Game" is to be able to distinguish human-composed music from machine-composed music. Another goal of "The Game" is to determine not only which works are human-composed but also which ones (they are chorales) best follow the style of J.S. Bach. The third goal is to see whether four mazurkas are really in the style of Frederic Chopin. Players are scored according to their answers, and the author quotes a statistic with large groups of listeners between 40 and 60 percent on the average. The author notes that experts in musicology have failed to recognize many of the machine music examples.

For the philosophy-oriented reader, a debate between Douglas Hofstadter and the author is included in the book. It is the opinion of this reviewer that such debates do not add too much to the field of artificial intelligence, and that those who are involved in this field should declare a moratorium on philosophical debate, and instead spend their time on creating better thinking machines. Philosophical debate is best left to those who like to indulge in it: the philosophers. There are many like Hofstadter who do not want to accept the musical creativity and abilities of the machines. Convincing these individuals of these abilities is typically difficult, and requires large expenditures of time. This time is better spent on the development of more sophisticated musical machines.

There are many interesting discussions in this book, too many to be reviewed here in the space allotted. One of the most fascinating of all the topics is the author's discussion on musical patterns. He believes these patterns are critical to the recognition of musical style, and therefore designates these patterns as "signatures". These signatures are contiguous note patterns that recur in at least two works of the works of a composer, and are found by using pattern-matching algorithms from artificial intelligence. Interesting in this context is the use of what the author calls "controllers", which allow variations of patterns to count as matches. As the name implies, these controllers may be set by users or by the machine itself.

Besides signatures, "earmarks" are another device used by the author to do pattern-matching in musical compositions. Earmarks are a kind of measure of the "inevitability" in some musical works. The machines analyze the music in their databases for earmarks and then make sure that these are not placed in inappropriate locations in the scores or omitted entirely. The author views them as "principles rather than data", and so a pattern-matching algorithm that finds them must return an "abstraction" representing the type of material used rather than actual musical events. Detecting earmarks can be very difficult says the author. More research is needed.