Conference proceeding
A Machine-Learned Model of Perceived Musical Affect
Proceedings of the ICMPC-SMPC 2016 Joint Conference, pp.169-172
2016
Abstract
Recent research has shown that affect perception in music can be reliably correlated with low-level acoustic parameters. In this study, we examine thirty excerpts from the Beethoven piano sonatas to generate a cognitive model of affect perception in this repertoire. Specifically, six low-level acoustic features have been derived from each excerpt using automatic feature extraction, including timbre, maximum amplitude, average amplitude, frequency of the dynamic signal, rate of change of the spectral centroid, and roughness. These acoustic features are used to predict perceived musical affect from 68 participants in this repertoire by using machine-learned neural networks. Nine affective dimensions are tested, resulting in nine models that are able to model the perception of affect based on low-level acoustic features. Specifically, between 7.1% and 41.5% of the variance in listener ratings are able to be accounted for by the resulting cognitive models.
Details
- Title: Subtitle
- A Machine-Learned Model of Perceived Musical Affect
- Creators
- Tess Dula - University of Mary Hardin–BaylorPaul Griesemer - University of Mary Hardin–BaylorJoshua Albrecht - University of Mary Hardin–Baylor
- Resource Type
- Conference proceeding
- Publication Details
- Proceedings of the ICMPC-SMPC 2016 Joint Conference, pp.169-172
- Publisher
- San Francisco, CA
- Copyright
- Copyright © 2016 the authors
- Language
- English
- Date published
- 2016
- Academic Unit
- School of Music
- Record Identifier
- 9985219827702771
Metrics
1 File views/ downloads
1 Record Views