RITMO: Understanding Drum Beats – Give the drummer some.... AI

Musicologists in the RITMO centre have told professional drummers to play "behind the beat", "on the beat", or "ahead of the beat". These are common terms in musical phrasing which refer to playing especially relaxed or "pushy". We would like to understand: what *exactly* have these drummers done, as they changed their way of playing?

See some Youtube examples here (not from RITMO):

https://www.youtube.com/watch?v=Y9SZ4i1eo00 (see from 1:20) https://www.youtube.com/watch?v=GlcSuJlOwJ0

The goal of this thesis is to automatically obtain this information from the timing data that we have available, and contribute to a better understanding of how professional musicians adjust their playing. This information can, for example, be useful in music education (how to learn to play behind the beat or ahead of the beat), or for music synthesis (to make drum machines sound more "human").

Bildet kan inneholde: tromme, musikk instrument, musiker, trommeslager, trommer.

Methodologically, the plan is to apply supervised Machine Learning (ML) and phrase the problem as a classification, where an ML model is trained to recognize the three different phrasing styles. Then, the model can be analysed to better understand the importance of input parameters (e.g., was it mostly the snare drum that allowed the ML system to recognize a drum pattern as being played behind the beat?). This can be done, e.g., by determining feature importance in case of a Random Forest model, or with SHAP value analysis in case of deep learning models.


Some related work:

Danielsen, Anne; Waadeland, Carl Haakon; Sundt, Henrik G. & Witek, Maria (2015). Effects of instructed timing and tempo on snare drum sound in drum kit performance.  Journal of the Acoustical Society of America.  ISSN 0001-4966.  138(4), s 2301- 2316 . doi:10.1121/1.4930950 

The TIME – Timing and Sound in Musical Micro-rhythm project at RITMO.

(Mari Romarheim Haugen, “Investigating Musical Meter as Shape: Two Case Studies of Brazilian Samba and Norwegian Telespringar”, Proceedings of the 25th Anniversary Conference of the European Society for the Cognitive Sciences of Music, 31 July-4 August 2017 )


The tasks of the project:

  • Get an overview of earlier work on motion capture data and machine learning analysis
  • Study available data sets and/or collect own data
  • Compare various machine learning algorithms to provide timing information about different drumming styles
  • Write master thesis.



Publisert 7. okt. 2019 13:36 - Sist endret 11. okt. 2019 09:35

Omfang (studiepoeng)