Cemgil A.T. & Kappen, B. (2002) Tempo Tracking and Rhythm Quantization by Sequential Monte Carlo In: Advances in Neural Information Processing Systems 14, Tom Dietterich, Sue Becker and Zoubin Ghahramani (Eds.), MIT Press.
Abstract
Automatic music transcription refers to computer generation of some high level description of a musical performance, for example in form of western music notation. In this paper, we constrain the problem to transcription of MIDI recordings, i.e. to generation of a notation from a list of onset times. In principle, one can always exactly represent an onset list by using a very complex but useless musical notation. The other extreme is generating a very simple notation that does not capture the necessary rhythmic information. An acceptable transcription successfully balances these two extreme cases. In this paper we exploit the inherent relationship between quantization and tempo tracking in a computationally feasible way and without compromising perceptual issues. Our simulation results suggest that our model has the potential to outperform existing methods for tempo tracking.
Full paper (pdf)