A Bayesian Approach to Drum Tracking

Publication Type:

Conference Paper


SMC Conference 2011 (2011)


This paper describes a real-time Bayesian formulation of the problem of drum tracking. We describe how drum events can be interpreted to update distributions for both tempo and phase, and how these distributions can combine together to give a real-time drum tracking system. The proposed system is intended for the purposes of synchronisation of pre-recorded audio or video with live drums.
We evaluate the algorithm of a new set of drum files from real recordings and compare it to other state-of-the-art algorithms. Our proposed method performs very well, often improving on the results of other real-time beat trackers. We discuss the merits of such a formulation and how it makes explicit the assumptions that underlie approaches to beat tracking. We conclude by considering how such an approach might be used for other tasks, such as score following or audio alignment. The proposed algorithm is implemented in C++ and runs in real-time.

smc2011_submission_185.pdf288.09 KB
SMC paper: