A Data Cleansing System for Musical Expression Analysis

Abstract

A new data cleansing system for musical expression analysis is introduced. In order to analyze musical expression and to create musical rendition models, it is necessary to develop this new system since the database used in this study is no longer available and it is impossible to collect any more data without the new system. Inputs of this system are actual musical performance data of classic piano music in WAV format, and additionally musical information, i.e. chords, musical symbols, etc., obtained from analysis of the corresponding music score. The harmonic tension values and melodic attraction values are also used as inputs, which are obtained from the prolongational reduction tree expressing the cognitional and hierarchical tension-relaxation structure obtained from the score. From the WAV data, the system extracts physical onset times and calculates local tempos as its musical expression. Using multiple regression analysis, music expression of each performance is analyzed and some interesting features are obtained, that are useful to create musical rendition models.