On the analysis of musical performance by computer

McGilvray, Douglas (2008) On the analysis of musical performance by computer. PhD thesis, University of Glasgow.

Abstract

Existing automatic methods of analysing musical performance can generally be described as music-oriented DSP analysis. However, this merely identifies attributes, or artefacts which can be found within the performance. This information, though invaluable, is not an analysis of the performance process. The process of performance first involves an analysis of the score (whether from a printed sheet or from memory), and through this analysis, the performer decides how to perform the piece.

Thus, an analysis of the performance process requires an analysis of the performance attributes and artefacts in the context of the musical score. With this type analysis it is possible to ask profound questions such as “why or when does a performer use this technique”. The work presented in this thesis provides the tools which are required to investigate these performance issues.
A new computer representation, Performance Markup Language (PML) is presented which combines the domains of the musical score, performance information and analytical structures. This representation provides the framework with which information within these domains can be cross-referenced internally, and the markup of information in external files. Most importantly, the rep resentation defines the relationship between performance events and the corresponding objects within the score, thus facilitating analysis of performance information in the context of the score and analyses of the score. To evaluate the correspondences between performance notes and notes within the score, the performance must be analysed using a score-performance matching algorithm. A new score-performance matching algorithm is presented in this document which is based on Dynamic Programming. In score-performance matching there are situations where dynamic programming alone is not sufficient to accurately identify correspondences. The algorithm presented here makes use of analyses of both the score and the performance to overcome the inherent shortcomings of the DP method and to improve the accuracy and robustness of DP matching in the presence of performance errors and expressive timing.

Together with the musical score and performance markup, the correspondences identified by the matching algorithm provide the minimum information required to investigate musical performance, and forms the foundation of a PML representation. The Microtonalism project investigated the issues surrounding the performance of microtonal music on conventional (i.e. non microtonal specific) instruments, namely voice. This included the automatic analysis of vocal performances to extract information regarding pitch accuracy. This was possible using tools developed using the performance representation and the matching algorithm.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Keywords: Musical Performance Analysis, Score-Performance Matching, Score-Performance Alignment, Music Representation
• Subjects: Q Science > QA Mathematics > QA75 Electronic computers. Computer science; T Technology > T Technology (General); M Music and Books on Music > M Music
• Colleges/Schools: College of Science and Engineering > School of Engineering
• Supervisor's Name: Nicholas, Dr. Bailey J.
• Date of Award: 2008
• Depositing User: Mr Douglas McGilvray
• Unique ID: glathesis:2008-451
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 09 Jan 2009
• Last Modified: 21 Mar 2019 12:07
• URI: https://theses.gla.ac.uk/id/eprint/451