On Problem Solving and Decision Making in MUSAS, a Musical Arrangement System

Abstract

MUSAS is a system for automatic music arrangement. In our research, the definition of musical arrangement is “to get a four-part melody through selection of appropriate chords from the given monotonic melody,” and the problem of musical arrangement includes decision making using information combined with fuzziness and uncertainty. It is necessary to observe and analyze the intelligence of human thinking processes in order to build intelligent systems with higher performance. Therefore, we regard the realization of MUSAS not only as a development of expert systems, but also as a simulation of human thinking processes with intelligence through the problem of musical arrangement. MUSAS uses three stages to process the musical arrangement, the melody interpretation stage, the chord selection stage, and the harmony generation stage. In these stages, there are some sub-knowledge-based systems to solve each problem independently. But, in each stage, plural solutions obtained by some subsystems have to be reduced to one solution for one problem. Therefore, MUSAS consists of a cooperative-distributed problem solver with strata. The prototype system was implemented for Japanese nursery songs, and, we are now developing a second system for jazz standards. In this paper, the problem of the musical arrangement, mainly focusing on the problem of chord selection, is described, and the results of the simulation of chord selection through sub-systems are shown.