This item is available under a Creative Commons License for non-commercial use only
Music Performance Science (MPS), sometimes termed systematic musicology in Northern Europe, is concerned with designing, testing and applying quantitative measurements to music performances. It has applications in art musics, jazz and other genres. It is least concerned with aesthetic judgements or with ontological considerations of artworks that stand alone from their instantiations in performances. Musicians deliver expressive performances by manipulating multiple, simultaneous variables including, but not limited to: tempo, acceleration and deceleration, dynamics, rates of change of dynamic levels, intonation and articulation. There are significant complexities when handling multivariate music datasets of significant scale. A critical issue in analyzing any types of large datasets is the likelihood of detecting meaningless relationships the more dimensions are included. One possible choice is to create algorithms that address both volume and complexity. Another, and the approach chosen here, is to apply techniques that reduce both the dimensionality and numerosity of the music datasets while assuring the statistical significance of results. This dissertation describes a flexible computational model, based on symbolic approximation of timeseries, that can extract time-related characteristics of music performances to generate performance fingerprints (dissimilarities from an ‘average performance’) to be used for comparative purposes. The model is applied to recordings of Arnold Schoenberg’s Phantasy for Violin with Piano Accompaniment, Opus 47 (1949), having initially been validated on Chopin Mazurkas.1 The results are subsequently used to test hypotheses about evolution in performance styles of the Phantasy since its composition. It is hoped that further research will examine other works and types of music in order to improve this model and make it useful to other music researchers. In addition to its benefits for performance analysis, it is suggested that the model has clear applications at least in music fraud detection, Music Information Retrieval (MIR) and in pedagogical applications for music education.
O'Doherty, E. (2019) Measuring Expressive Music Performances: a Performance Science Model using Symbolic Approximation, Doctoral Thesis, Technological University Dublin.