Computational models of musical microtiming in drum kit performance
By Fredrick William Caiskey Bruford (Queen Mary Univ. of London), visiting researcher at RITMO during October-November 2019.
Fredrick Bruford (Photo: Nina Krogh)
In October 2019, I was lucky enough to have the opportunity to carry out a short research visit at RITMO, working primarily with Olivier Lartillot and the members of the TIME project, supported by a Short-Term Scientific Mission grant from the Nordic Sound and Music Computing Network. The aim of my 5 week stay was to collaborate on a project with the aim of developing computational models of musical microtiming in drum kit performance.
The overall aim of my PhD is to design intelligent methods of exploring and navigating of a large library of expressive symbolic drum loops. To enable an intelligent exploration tool, we have been developing and evaluating feature sets to computationally model similarity between drum loops. Our current focus is in developing new features to model perceptually salient aspects of drum loops that are not adequately accounted for in existing models. These algorithms could then be applied both in intelligent library management systems, but also computationally creative systems such as automatic drum loop generators.
In a previous study, we found that swing and expressive timing seemed to be important factors in the perception of similarity between drum loops. Much existing work in music perception also points to microtiming variation being a significant factor in rhythm perception. Currently however, features and models for estimating drum loop similarity work on quantized drum patterns, neglecting this microtiming information.
The initial aim of the research visit was therefore to develop computational models that can describe the perceptually significant factors of microtiming deviations common in expressive drum performance. Being integrated within the TIME project, devoted to the study of musical microtiming, was therefore the perfect environment to begin studying this problem. While microtiming was the focus initially, as the visit progressed we begun to work more generally on other novel approaches to similarity modelling, as well as further evaluation techniques based on data collected for a previous study.
During the stay we developed several computational models drawing from microtiming research being carried out within the TIME project and beyond. This included models of pushed and laidback feel based on the detection of microtiming interactions between instruments, and between instruments and a metronome; onset pattern-based similarity models that incorporate microtiming deviations when measuring onset displacements in similar rhythms; and models of swing and metrical feel that can account for both the swing ratio and the perceptual salience of swing in a rhythm, and whether a rhythm is closer to compound or duple meters. Towards the end of the research visit we begun to investigate rhythm transformation models and their use in removing ornamentation from rhythms, with the intention of adapting these for use as a rhythm similarity measure. We also begun to investigate new techniques for feature selection and evaluation based on perceptual data. We have a conference paper soon to be submitted based on the outcome of this work, including description and evaluation of the algorithms we developed, and public release of implementations of said algorithms so they can be used by both music informatics and musicology researchers alike.
During the course of my visit I had the opportunity to present my work 3 times, first as a Food & Paper to the whole research group, secondly to members of the TIME project during the Son workshop, and thirdly during the SAMA reading group meeting, all providing opportunities for valuable discussions with both researchers in the TIME project and wider RITMO group. The Son workshop in particular was a highlight; it was fantastic to spend two days on focused collaboration and get to see a beautiful part of Norway, and I am extremely grateful to have been part of it. Overall my short time in RITMO was both hugely enjoyable and intellectually stimulating, and I would like to thank everyone at RITMO and in the TIME project particularly for giving me the opportunity for such a great collaboration and making me feel so welcome. I hope to be back to visit again soon!
This research stay was supported by a Short-Term Scientific Mission grant from the Nordic Sound and Music Computing Network.