Automatic Composition of Intention-Based Music for Multimedia Interfaces

Music is known to be able to support certain aims of presentations. In addition, there are a number of approaches to automatic music composition. Thus, it is a natural consequence, to establish automatic, intention-based music composition as an additional tool for multimedia presentation systems. However, many related applications show unsatisfactory results. An explanation for this phenomenon is that the consideration of musical structure at diierent abstraction levels is often neglected although many relevant theories do exist and have been reened to a high level. The aim of the project described in this paper is to build a system for the automatic composition of music that supports the intentions of a given presentation. The idea underlying the system design is that the composition process works on diierent abstraction levels, so that each subset of desired intentions can be realized at the appropriate level. The resulting coordination problems make great demands on the level of musical structure which functions as a link between the presentation intentions and the musical realization. It can be seen as a kind of \rhetorical" level of music. This paper discusses the relevant aspects concerning this structural level and introduces a composition system which exploits related theories. Feasibility People often have reservations about the automatic generation of music. That is because many approaches have gone too far by aiming at producing masterpieces. The goal of this project is not to satisfy a user's intellectual enjoyment, rather we have to accept the various restrictions concerning not only the realizability of automatic music composition but also the eeectiveness of music. The eeectiveness of musical strategies depends, for instance, on parameters such as the user's individual music experience Dowling & Harwood, 1986], attention, personality, intelligence Bruhn, 1988], and bias Pekrun & Bruhn, 1986], as well as parameters dependent on the temporal and spatial environment of the presentation Tauchnitz, 1990]. User modelling would become extremely complex as these, sometimes temporally changing, parameters would have to be considered simultaneously. The type and amount of non-musical information which can readily be presented through music is strongly debated, and it is known that many musical eeects reach their audience with less certainty than is often assumed Groomann, 1991]. If we consider people of western cultures with average music listening experience, then the attributes of most listeners include a preference for tonal, well-constructed music with normal harmonic functions and predictable musical expectations Bruhn, 1988]. On this …