DS-2020-12: Experienced listeners: Modeling the influence of long-term musical exposure on rhythm perception

DS-2020-12: der Weij, Bastiaan van (2020) Experienced listeners: Modeling the influence of long-term musical exposure on rhythm perception. Doctoral thesis, University of Amsterdam.

[thumbnail of Full Text] Text (Full Text)
DS-2020-12.text.pdf

Download (2MB)
[thumbnail of Samenvatting] Text (Samenvatting)
DS-2020-12.samenvatting.txt

Download (4kB)

Abstract

This thesis investigates rhythm perception using computational modeling techniques and develops a set of tools and techniques for the definition and evaluation of probabilistic generative models of music perception. We argue that previously proposed computational models of rhythm perception insufficiently account for how perception has been shaped by culturally embedded listeners' prior experience, practice, and training. Motivated by predictive processing theories of perception, we propose a probabilistic generative model of meter perception which, compared with previous models, to a greater extent learns from patterns and regularities in datasets of rhythms (representing a musical environment). The outcome of this learning process simulates the effects of the long-term exposure that listeners receive to rhythms in their musical environment.

The emphasis on patterns and regularities in musical rhythms in the environment leads us to evaluate the model in a cross-cultural context. We investigate how varying degrees of sensitivity to statistical patterns, the style or cultural origin of rhythms from which the model learns, and that of rhythms on which it is evaluated factor into the model's performance. That is, the model, together with an alternative model representing traditional Western theories of meter perception is evaluated on culturally familiar rhythms (of the same style as the rhythms it has learned from) and culturally unfamiliar rhythms (of a different style than the rhythms it has learned from). In this way, we investigate whether there is between-style variety in the patterns and regularities useful for inferring meter, and the amount and type of sensitivity to statistical patterns necessary for detecting this variation.

Concretely, we investigate empirical samples containing rhythms of Western folksongs and Turkish makam music. We find that the Western as well as the Turkish rhythms contain regularities that allow a listener familiar with these regularities to infer meter from individual rhythms. These regularities involve patterns of multiple rhythmic events and are more complex than schematic patterns of expectation associated with traditional theories of meter perception. Furthermore, we find that some of these patterns occur only in Western or only in makam rhythms. However, the results also suggest that the patterns and regularities by which meter can be inferred are to a significant extent shared between makam and Western rhythms.

Additionally, this thesis presents a framework for the design and implementation of discrete dynamic Bayesian network models with deterministic constraints. The framework enables formal, concise, and explicit definitions of such models that can straightforwardly be translated into functional and executable implementations. The framework aims to enhance the transparency and reproducibility of modeling research and to make it easier for other researchers to build further on modeling work that uses the framework. The framework is suitable for defining predictive-processing based theories of music perception. Such cognitive models, including the models discussed in this thesis, can be seen as sequential music prediction models and can be used to model musical expectancy and uncertainty evolving dynamically over time, in lockstep with the temporal progression of music.

Concerning the structure of this thesis, Chapter 2 surveys a variety of previously pursued approaches to modeling rhythm perception and reviews ways in which rhythm perception has been found to be influenced by prior experience, training, and practice. Chapter 3 develops the technical details of the modeling framework. Chapter 4 demonstrates its use by presenting an adaptation of a previously proposed generative model of rhythm and meter perception. Chapter 5 provides technical definitions of two models that are subjected to empirical evaluations later in the thesis. Chapter 6 defines and motivates the main model proposed in this thesis (Chapter 5 contains a more technical description of this model). Finally, Chapter 7 presents cross-cultural evaluations in which the models and the methodology developed in Chapter 5 are applied to empirical samples of Turkish makam music and Western folksongs.

Item Type: Thesis (Doctoral)
Report Nr: DS-2020-12
Series Name: ILLC Dissertation (DS) Series
Year: 2020
Subjects: Computation
Depositing User: Dr Marco Vervoort
Date Deposited: 14 Jun 2022 15:17
Last Modified: 14 Jun 2022 15:17
URI: https://eprints.illc.uva.nl/id/eprint/2181

Actions (login required)

View Item View Item