Repository | Book | Chapter
(2015) Mathematics and computation in music, Dordrecht, Springer.
A change-point approach towards representing musical dynamics
Katerina Kosta, Oscar F. Bandtlow, Elaine Chew
pp. 179-184
This study proposes a novel application of change-point techniques to the question of how dynamic markings in a score correspond to performed loudness. We apply and compare two change-point algorithms–Killick, Fearnhead, and Eckley's Pruned Exact Linear Time (PELT) method, and Scott and Knott's Binary Segmentation (BS) approach(-)to detecting changes in dynamics in recorded performances of Chopin's Mazurkas. Dynamic markings in the score, assumed to correspond to change points, serve as ground truth. The PELT algorithm has a higher average best F-measure (15.78 % for 0 tolerance threshold; 29 % for one-beat tolerance threshold) compared to the BS algorithm (10.94 % and 19.74 %, respectively), it also results in a smaller average Hausdorff distance–32.8 vs. 77 score beats for 0 tolerance and 32 vs. 52.2 score beats for one-beat tolerance. Applications of loudness change-point detection include audio-to-score transcription.
Publication details
DOI: 10.1007/978-3-319-20603-5_18
Full citation:
Kosta, K. , Bandtlow, O. F. , Chew, E. (2015)., A change-point approach towards representing musical dynamics, in T. Collins, D. Meredith & A. Volk (eds.), Mathematics and computation in music, Dordrecht, Springer, pp. 179-184.
This document is unfortunately not available for download at the moment.