AbstractsLanguage, Literature & Linguistics

Automatic melodic analysis

by Ryan Groves

Institution: McGill University
Year: 2016
Keywords: Music
Posted: 02/05/2017
Record ID: 2119400
Full text PDF: http://digitool.library.mcgill.ca/thesisfile141517.pdf


Melodic reduction is a process performed on symbolic music to discover the more important structural notes of a melody. In music theory literature the process has a foundation in the works of Heinrich Schenker, as well as Fred Lerdahl, and Ray Jackendoff. Among other applications, melodic reduction can be used for: harmony estimation, melodic similarity and comparison, compression of melodic representations, melodic search, and automatic or assisted composition. The process of melodic reduction can also involve auxiliary melodic analysis methods such as melodic segmentation (or grouping), metrical structure analysis, and melodic parallelism. This thesis investigates the use of a technique originally developed for Natural Language Processing (NLP) to identify hierarchies in sequential data—the Probabilistic Context-Free Grammar (PCFG).Using the PCFG, it is possible to encode the rules of embellishment, allowing the identification of melodic embellishments. After defining each rule for a PCFG, it can take a sequence of notes and automatically build a tree structure by iteratively applying the rules. The hierarchical tree structure represents a reduction, since every subsequent level contains less and less notes. This thesis involves the supervised training of a PCFG using melodic reductions that were annotated using The Generative Theory of Tonal Music (GTTM) (Lerdahl and Jackendoff 1983). The GTTM dataset provides the melodic reductions in tree format, which is compatible with the PCFG technique (Hamanaka, Hirata, and Tojo 2007b). By leveraging the existing reductions, one can model the distributions for each of the rules contained in the analyses, and use those to find the most probable melodic reduction for a new input melody. A standard evaluation methodology is used to test the efficacy of the melodic reduction grammar. La réduction mélodique est un traitement réalisé sur la musique symbolique dans le but de découvrir les notes structurelles les plus importantes d'une mélodie. Dans la littérature de la théorie musicale, le traitement trouve son fondement dans les oeuvres d'Heinrich Schenker, ainsi que de Fred Lerdahl et de Ray Jackendoff. Entre autres applications, la réduction mélodique peut être utilisée pour effectuer : Une estimation de l'harmonie, la similarité et la comparaison mélodiques, la compression de représentations mélodiques, la recherche mélodique, et la composition automatique ou assistée. Le processus de la réduction mélodique peut également impliquer des méthodes auxiliaires d'analyse mélodique, comme par exemple la segmentation (ou le regroupement) mélodiques, l'analyse de la structure métrique, et le parallélisme mélodique. Cette thèse étudie l'utilisation d'une technique originellement développées pour le traitement automatique du langage naturel (NLP) pour identifier les hiérarchies dans les données séquentielles La grammaire probabiliste indépendante du contexte (PCFG). En utilisant la PCFG, il est possible d'encoder les règles d'embellissement, ce qui… Advisors/Committee Members: Ichiro Fujinaga (Internal/Supervisor).