LEARNING-PERFORMANCE RULES IN A MUSIC INTERPRETATION SYSTEM

It has been reported that if music is realized as nominally described by the score, a musically unacceptable performance emerges. Music inte rpretation is necessary to generate a musically acceptable performance . We have been constructing a rule-based music interpretation system t hat generates sophisticated performance from a printed music score. Th e whole performance data are given by the product of expression of eac h dynamic mark explicitly drawn in the score and expression of analyze d motives. This paper describes the function of learning how to play m usic, which is the most important process in music interpretation. The target to be learned are expression rules and grouping strategy; expr ession rules are used in order to convert dynamic marks and motives in to concrete performance data, and grouping strategy is used in order t o extract motives from note sequence. They are learned from the given virtuoso performance. The delicate control of attack-timing, duration and strength of notes are extracted by the music transcription functio n. The performance rules are learned by investigating how the same or similar musical primitives are played in the performance. As for the g rouping strategy, the system analyzes how the player grouped music, an d registers dominant note-sequences to extract motives.