SCHUBOT: Machine Learning Tools for the Automated Analysis of Schubert’s Lieder
CitationNagler, Dylan Jeremy. 2014. SCHUBOT: Machine Learning Tools for the Automated Analysis of Schubert’s Lieder. Bachelor's thesis, Harvard College.
AbstractThis paper compares various methods for automated musical analysis, applying machine learning techniques to gain insight about the Lieder (art songs) of com- poser Franz Schubert (1797-1828). Known as a rule-breaking, individualistic, and adventurous composer, Schubert produced hundreds of emotionally-charged songs that have challenged music theorists to this day. The algorithms presented in this paper analyze the harmonies, melodies, and texts of these songs. This paper begins with an exploration of the relevant music theory and ma- chine learning algorithms (Chapter 1), alongside a general discussion of the place Schubert holds within the world of music theory. The focus is then turned to automated harmonic analysis and hierarchical decomposition of MusicXML data, presenting new algorithms for phrase-based analysis in the context of past research (Chapter 2). Melodic analysis is then discussed (Chapter 3), using unsupervised clustering methods as a complement to harmonic analyses. This paper then seeks to analyze the texts Schubert chose for his songs in the context of the songs’ relevant musical features (Chapter 4), combining natural language processing with feature extraction to pinpoint trends in Schubert’s career.
Citable link to this pagehttp://nrs.harvard.edu/urn-3:HUL.InstRepos:12705172
- FAS Theses and Dissertations