Apollonl3: A Training System for Emergency Situations in a Piano Performance

摘要

During a piano performance, there is always the possibility that the musician will cease playing on account of an unexpected mistake. In a concert, such a situation amounts to an emergency state in the piano performance. Therefore, we propose a system named "Apollonl3" that simulates emergency states by replacing particular notes with different ones, in the manner of mis-touches, by referring to the performer's degree of proficiency as determined by a performance estimation algorithm. From the results of user studies, we confirmed that Apollon13 is basically effective as a training system for handling emergency states. However, the estimation algorithm could not precisely identify the note-replacement points where the subjects become upset. Accordingly, we evaluated the estimation algorithm by comparing it with the player's subjective assessment based on the data of an experiment. As a result, we found a clear relationship between the subjective assessment and the points, obtained by experiment, at which players become upset. This result suggests that an algorithm could gain the ability to detect the "upset points" by approximating a human's subjective assessment.