Music Signal Analysis Methods and Their Applications to Bowed String Instruments Based on Expressiveness Modeling
悠馬 小泉 · Institutional Repositories DataBase (IRDB) · 2014
The essence of music is the "expression" of each per former, namely, the deviations in amplitude, pitch, tim- bre and tempo/rhythm that they add to their performance. Hence, in computational applications of music, deviation analysis is important. However, because the musical tone of excitation-continuous musical instruments changes com- plexly in accordance with the level of controllability, it is difficult to analyze the deviations. This paper proposes five deviation analysis methods, focused on statistical consis- tency and repetition, for bowed string instruments. Results of analysis using each of the methods are presented. By us- ing CMKLD, which is an acoustic feature based on aural characteristics, the error rate of musical score alignment was decreased to 63.2 percentage points. By modeling the sound control indeterminacy due to performance expression, the er- ror rate of intra-note segmentation was decreased to 89.4 (A- to-S) and 48.8 (S-to-R) percentage points, respectively. By using a generative model of amplitude contour, focused on statistical consistency, a performer's phrasing and variation of articulation could be analyzed. By using a physical model of a violin in the frequency domain, high-quality sound could be synthesized via quantitative expressiveness parameters. By removing deviations that have no statistical consistency, misplayed sounds could be adjusted. These results show that the proposed methods can be used to analyze the expressive deviation of bowed string instruments, and can be applied to musical information processing systems such as automatic performance and automatic skill evaluation.