In this thesis, an autonomous accompaniment system which accompanies a keyboard soloist is proposed. The primary goal of this study is to develop an algorithm that performs the real-time interaction between soloist and computer accompanist.
There are two subproblems : (1) matching the input from a live performer against the performance score, and (2) adjusting the tempo of accompaniment to the tempo of the soloist. As it is expected that the live solo performers may make some mistakes, it is necessary for the autonomous accompaniment to accomodate the performance mistake as it matches the actual solo against the score. To solve this problem, an efficient dynamic programming method is presented to match between the performance and the score, and a tempo adjusting technique is adapted for this system. The entire system may be regarded as an expert system which consists of three modules of expertises : soloist, accompanist and music score.