Study of Music Information Retrieval Methods Using Music Structure

本研究对音乐结构的各个方面进行了研究，包括研究用计算机表示和分析音乐结构的理论基础，研究人类音乐结构认知本质的心理学实验，以及基于其结果的应用系统的实现和验证，重点是在音乐信息检索(MIR)中的应用。MIR是近年来一个非常活跃的研究领域，除了普通的基于表面线索的旋律检索外，还需要面向基于高层结构分析的检索和/或专门针对特定音乐特征的研究。可以看出，这项研究参与了这些正在进行的努力。本研究的个别主题和结果可以总结如下。研究中使用的音乐数据，包括钢琴作品(约300首曲目)和歌唱数据在…推出的服务器上以数据库的形式积累和组织本研究的主要结论如下：1.更多S授课，配合歌唱等原始数据进行心理实验。作为实现基于音乐结构的MIR的一部分，实现了一个从多声部音乐中检索结构音符的原型系统，并进行了测试。已经实现了一种使用打击音的口头(拟声词)模拟查询的打击乐模式检索系统，并获得了高识别率。将研究结果应用于支持打击乐声部输入的音乐创作和编曲系统--《声乐鼓手》。为了验证在检索查询中使用歌唱声音的可能性和合理性，分析了与歌唱技能评价相关的音乐/声学特征，并将结果应用于一个不使用乐谱信息的自动歌唱技能评价原型系统中。为了研究人类音乐结构认知的特点，验证其对MIR的适用性，对人类背诵未知旋律的方式进行了实验研究，并分析了歌唱错误的结构特征。上述部分结果仍在准备发表中，并计划在不久的将来出版。综上所述，虽然本研究在扩展原有研究计划的基础上取得了显著(和高度认可的)成果，但在音乐结构分析及其在MIR中的应用方面仍有许多尚未解决的问题，因此未来的工作将继续下去，目标是实现高质量和灵活的MIR方法。较少

This research looks into various aspects of music structures, with work including study on the theoretical basis of representing and analyzing such structures by computers, psychological experiments studying the nature of human music structure cognition, and implementation and verification of application systems based on their results with focus on use in Music Information Retrieval (MIR). MIR has been a highly active research field in the recent years, and besides ordinary melody retrieval based on superficial cues, research oriented towards retrieval based on high level structural analysis and/or specializing on specific musical features is desired. This research can be seen to take part in such ongoing efforts.The individual topics and results of this research can be summarized as follows.1. Music data used in the research, including piano pieces (approx. 300 pieces) and singing data have been accumulated and organized in the form of databases on the server machine introduced by thi … More s grant, together with original data such as singing data from the psychological experiments.2. As part of realizing MIR based on musical structures, a prototype system retrieving structural notes from multi-part music has been implemented and tested.3. A system for percussion pattern retrieval using queries verbalized (onomatopoeic) imitations of percussion sounds has been implemented and achieved a high recognition rate. The results were applied to "Voice Drummer", which is a system that supports input of percussion parts in music composition and arrangement.4. In order to verify the possibility and plausibility of using singing voice in retrieval queries, the music/acoustic features related to singing skill evaluation has been analyzed, and the results were applied to a prototype system of automatic singing skill evaluation that does not use music score information.5. In order to study the characteristic features of human music structure cognition and to verify their applicability to MIR, an experimental study on how humans recite unknown melodies was performed, and the structural features of singing errors were analyzed.Part of the above results is still in preparation for publication, and is planned to be published in the near future. In summary, while this research achieved notable (and highly recognized) results extending the original research plan on the one hand, there are still many unsolved issues related to music structure analysis and its application to MIR, so future work will continue with the aim of realizing high quality and flexible MIR methods. Less