Toward realizing automatic evaluation of playing scales on the piano

The aim of this study is to realize automatic evaluation of playing scales on the piano and to point out defects of the user in his/her playing scales. Records of playing scales together with corresponding subjective scores are collected in order to determine the evaluation criteria. Performances were recorded on a MIDI sequencer. Onset intervals, velocities, and durations for each note are recorded and a set of three regression curves for each performance is calculated by spline interpolation for the average value of each parameter calculated for each wrist position on the keyboard. The scores obtained are employed as both test data or training data in turn to achieve open tests. KL expansion and k-NN algorithm are tried to predict subjective scores of new scale-playing data. Outputs by the proposed system are relatively similar to subjective evaluation scores and the appropriateness of values for description parameters is discussed.