Analyzing the perception of Chinese melodic imagery and its application to automated composition

摘要

This study investigated the influence of two music elements, mode and rhythm, on melodic imagery. By modulating different parameters of mode and rhythm, this study produced various melody samples. Kansei engineering was adopted as a research method for designing the experimental process. A set of music assessment rules was established by regularizing the existing modes and rhythms and analyzing and comparing their differences. The results were applied to automated composition in order to develop a comprehensive automated composition system. Before the hearing experiment, 39 adjective pairs were selected for the survey of melodic imagery by reviewing literature and engaging in discussions with experts. A Markov chain was adopted for analyzing the current modal music to establish a Markov switching table for the Chinese pentatonic scale, and metrical complexity was used for generating rhythms. Finally, 30 music samples were produced using the automated composition system by applying the various combinations of modes and rhythms. The hearing experiment was conducted on 35 participants for examining their perception of melodic imagery. Item analysis and a reliability test were performed on the experimental data for eliminating invalid samples and items. Thus, 14 adjective pairs with a Cronbach's alpha of 0.887 were obtained, indicating favorable reliability. In addition, this study involved conducting factor analysis and extracting three imagery factors, namely passionate-apathetic, beautiful-ugly, and romantic-rigid. Subsequently, multivariate analysis of variance was conducted in this study for determining the relationship between modes and metrical complexity and the 14 adjective pairs, and the estimated means were calculated. Multidimensional scaling was adopted for producing a two-dimensional perceptual map for the construction of a database. The experimental results of this study showed that modes and rhythms in music can effectively help musicians or computer composition systems compose melodies that create various perceptual imageries.