Date of Submission
Fall 2020
Academic Program
Computer Science
Project Advisor 1
Sven Anderson
Abstract/Artist's Statement
Segmentation is a well-studied area of research for speech, but the segmentation of music has typically been treated as a separate domain, even though the same acoustic cues that constitute information in speech (e.g., intensity, timbre, and rhythm) are present in music. This study aims to sew the gap in research of speech and music segmentation. Musicians can discern where musical phrases are segmented. In this study, these boundaries are predicted using an algorithmic, machine learning approach to audio processing of acoustic features. The acoustic features of musical sounds have localized patterns within sections of the music that create aurally perceptible “events” that musicians identify as distinctive characteristics of a phrase. An experiment was conducted to gather data from musicians for the machine learning algorithm, and to set an upper bound on the performance of such an algorithm. The algorithm succeeded in detecting phrase boundaries, as determined by the participants, with accuracy scores of 0.91, 0.67, and 0.60 for the data from three participants, but there are still improvements to be made--specifically, the low specificity of the machine learner’s prediction is a challenge for a future endeavor.
Open Access Agreement
Open Access
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 4.0 License.
Recommended Citation
Petratos, Evan Matthew, "A Machine Learning Approach to the Perception of Phrase Boundaries in Music" (2020). Senior Projects Fall 2020. 23.
https://digitalcommons.bard.edu/senproj_f2020/23
This work is protected by a Creative Commons license. Any use not permitted under that license is prohibited.