Cope, David. (1991). Computers and Musical Style. Madison, Wisconsin: A-R Editions.

I. History of Automatic Music

Computers may be fairly recent, but people have been exploring automatic music generation for centuries. All sorts of cultures have variations of wind chimes, bells, and aeolian harps. The development of the first "computer" in 1840 by Lovelace and Babbage allowed speculation that one day automatic music composition would be possible. Throughout the twentieth century there are many examples of invented instrument systems that automatically produced sounds according to formal systems, chance procedures, or natural physical phenomena.

Computers and other mechanical hardware have been extremely successful in generating highly serialized, randomized, and transduced (e.g. turning the earth's magnetic field pulsations into sound, see pg. 4) pieces of music. But where is a style that can be perceived, held in memory, and learned?

II. Beginnings

In 1981 David Cope began brainstorming to develop a computer-aided system to write new pieces in the styles of past composers. He envisioned a process where anyone, regardless of programming or musical abilities, could interact with a computer program to compose and tweak a novel piece that falls within the domain of a chosen style. In addition, Cope was struggling with composer's block and he felt that a computer would be a good way to cultivate a deeper understanding of musical style. If the definition of musical style could be operationalized and its components encoded into a computer, Cope could increase his creativity and productivity.

III. Defining Style

The essential elements that define musical style are patterns. Compositions may be superimposed upon each other in order to discover patterns that are unique to a particular era, a specific composer's techniques, or an individual piece. (This is similar to L. Meyer's Hierarchy of Constraints.) There are many broad definitions of style which are great for theoretical discourse, but a computer would need an explicit definition of style.

Language parallels pervade Cope's thinking and techniques. He seems reluctant to define style, but eventually he defines it as, "the identifiable characteristics of a composer's music which are recognizably similar from one work to another," (p.30). Cope isolates patterns by encoding multiple samples of a composer's works into a database and using computation functions to seek out a composer's style.

IV. The SPEAC system

This system was invented to give musical ideas (chords, motives) a greater degree of flexibility and to connect them together in smooth, logical orderings. The composition program works from a database of musical samples as it dictionary. Musical fragments from the database are combined according to a logic that parallels language syntax. The SPEAC logic system must be fine tuned so that it varies the inherited material and composes original music. If the program becomes too creative, connections may become abrupt.

In traditional tonal function analysis a 'five' chord is a 'five' chord. But, the SPEAC system acknowledges that harmonies function in multiple ways within a single key. This parallels language in which a word's meaning may change depending on context. The components of SPEAC are as follows:

1. (S)tatement: An idea may "simply exist 'as is' with nothing expected beyond iteration," (p.34).
2. (P)reparation: "Preparations modify the meanings of statements or other identifiers by standing ahead of them without being independent," (p.35).
3. (E)xtension: This is a way to prolong a statement. Basically, it is a codetta.
4. (A)ntecedent: These "cause significant implication and require resolution," (p.35).
5. (C)onsequent: These resolve the antecedent. "Consequents are often the same chords or melodic fragments as found in S. However, they have different implications," (p.35).

The SPEAC system is a way to save computational resources and to produce coherent and interesting music. Instead of building linear chains of harmonic progressions, a "succession rule" list allows particles of musical ideas to be flexible and appropriate to the context. Once a musical "sentence" or "paragraph" is formed it may be used over and over again as surface features are reapplied until the composer/programmer is satisfied. This process resembles a song contrafact in which new melodies are applied over existing chord changes to create an entirely new piece.

V. Bottom-up vs. Top-down processes

Cope acknowledges the influence of Schenker's theories upon his programming strategies. He states that Schenkerian analysis is a bottom-up process. It elicits the foundational structures of the music that lie under the musical surface. Cope's computer composition utilizes a top-down approach that first defines large-scale units and then fills in the surface. Since composition and analysis are reverse processes, Schenker and Cope are approaching music from a similar angle.

VI. How to distinguish composers?

1) Signatures: This is a short series (2 to 9 notes) of notes that distinguishes a particular composer or several composers, but does not make a certain piece unique (e.g. the Landini cadence). This concept is akin to L. Meyer's notion of "strategic play" or the ways in which a certain composer navigates through the general rules of the era. 2) Texture: What is the rate of activity within a voice? What is the number of voices? How often does variation occur? 3) Counterpoint: How much is used and where does it occur?

These three elements can be encoded and manipulated by a computer and form the basis for Cope's computer composition procedures.

VII. How do we prevent the computer from generating wandering melodies and harmonies?

We need ATNs (Augmented Transition Networks) - SPEAC is a ATN. Cope creates coherent music using a musical syntax that is inspired by language hierarchies. Original musical material must be composed over progression abstractions that are analogous to sentence components such as nouns, verbs, and adjectives. Each point in this progression acts a node where a choice may be made (drawn from a database) before moving to another node.

VIII. LISP (list processing) programming language

This is a programming language that utilizes very small functions to perform big tasks when used in combination. Recursive operations are the key. A recursive function is a loop that continually calls upon itself and runs until a desired state is reached. One will often see "if/then" statements in recursive programs. These types of operations are very important for automatically sorting through a large list of symbols according to the programmer's instructions.

IX. Other interfaces for style construction

Cope developed a style dictionary. This is a series of a hundred or so questions (yes or no) aimed at defining musical style parameters. Alternatively, the method of style selection that Cope favors is to encode a composer's works into the computer and to have it analyze the style by finding patterns.

X. Conclusion

The computer programming perspective has a unique advantage of "separating rules and data," (p.216). This allows Cope to retrofit standard practice musical logics with new surfaces. For example, one may define "dominant function in more general terms. Rather than a 'five' chord, dominant function may be generalized as a "repetition in location and time," (p.214). Many harmonies may function as dominant as long as their context and frequency ties them to the tonic. The computer medium encourages Cope to think of music as a combination of underlying logics with superimposed surface data.

Issues/Comments:

• Cope seems highly indebted to Meyer, yet he criticizes Meyers definition of style for being too vague. Cope's own definition of style seems to vacillate between including groups of composers, an era, or an individual composer.
• Cope criticizes Schenker, but then derives SPEAC from Schenker. On page 28, he disagrees with LaRue's visual analogy that expresses musical style analysis as moving from large scale structures to small details. Cope refutes this and says that large and small parts must be assembled in tandem. Yet, his whole approach goes on to first define large scale forms and later fill in the details.
• The SPEAC approach is admirable, but it ignores many other possibilities for musical progressions and movement. The whole approach seems to root upon standard practice foundations. This is a limitation because musical logics are style specific not universal. For example, a piece with the surface of Stravinsky and the logic of the Classical period may be interesting, but it is still fundamentally a classical piece.
• Chord inversion is allowed in the SPEAC paradigm, but the allowance seems more like a process of surface variety for the sake of variety. Chord inversions are not just surface features. They will determine the proximal syntax . What is the rule system for creating specific harmonic progressions that are determined by inversion context?
• As I began reading the Cope, I imagined a computer program that acted as an automatic probe to construct a musical style from scratch. This would be difficult. Cope's analysis and synthesis procedures are resting on an assumption of style as a product of an individual composer, but isn't this definition too narrow?
• Cope gives examples of his LISP code, but readers are never privileged to a complete program.

• Music 838C Main Page