Evaluation of a scale performance on the piano using spline and regression models
Shinya Morita, Norio Emura, Masanobu Miura, Seiko Akinaga, Masuzo Yanagida · 2009
Recently, many systems have been developed for supporting novice and/or beginner pianists. Even though these systems use a set of performance data, they cannot properly evaluate performance proficiency because a procedure for calculating proficiency has not been proposed. To solve this problem, we have developed an evaluation model by introducing a spline curve or regression curve, assuming it to be a standard for evaluating proficiency for scale performances from an aesthetic viewpoint. This paper introduces other models for evaluation, comprised of a standard curve for evaluating a piano performance. Curves used are (1) a spline curve and (2) regression curves using n-dimensional models (18n810). Investigated here is the effectiveness of the curves, so as to determine a better curve for automatically evaluating performance. As a result, correlation coefficients between scores predicted by the spline curve model and evaluation scores by experts were 0.65, whereas the average of correlation coefficients between scores predicted by regression curves models and evaluation scores by experts was 0.58. In other words, a correlation coefficient of the spline curve model is confirmed to be higher than those of regression curve models. Therefore, it was confirmed that the spline curve model is more effective than n-dimensional curves at automatically evaluating performance.