This page lists some additional results not included in the paper:
Rhythmic Pattern Modeling for Beat and Downbeat Tracking in Musical Audio
Krebs, F., Böck, S., and Widmer, G.
Proceedings of the 14th International Society for Music Information Retrieval Conference (ISMIR), Curitiba, Brazil, November 2013.
Usually the first five seconds are excluded from the evaluation of beat tracking algorithms. However, on this page we decided to present results for the whole song, as we think this is equally meaningful considering only offline methods.
The proposed methods are named PS2 and PS8 and model two and eight rhythmic patterns respectively. The parameters of the observation model of PS8 were learned using the dancestyle labels as rhythmic pattern labels.
The dataset used in the evaluation is the Ballroom dataset, available here.
The annotations used in the evaluation are available here.
The evaluation measures are described in [1] and were implemented by Sebastian Böck in Python.
Please note that there are replicas among the songs of the dataset, as reported by Bob Sturm. Even though this means that some songs (13 of 698) appear in both train and test sets, we ignore this issue in the following and give results for all 698 files.
Beat tracking results on the Ballroom set (698 files):
Algorithm |
F-measure |
P-score |
Cemgil |
Goto |
CMLc |
CMLt |
AMLc |
AMLt |
D |
Dg |
PS2 |
0.819 |
0.779 |
0.738 |
0.740 |
0.620 |
0.658 |
0.822 |
0.875 |
2.342 |
1.501 |
PS8 |
0.855 |
0.839 |
0.772 |
0.868 |
0.745 |
0.786 |
0.818 |
0.865 |
2.499 |
1.681 |
Degara |
0.790 |
0.762 |
0.709 |
0.697 |
0.607 |
0.632 |
0.799 |
0.847 |
2.391 |
1.229 |
Ircambeat |
0.764 |
0.722 |
0.656 |
0.636 |
0.544 |
0.571 |
0.801 |
0.842 |
2.670 |
1.203 |
Klapuri |
0.728 |
0.694 |
0.651 |
0.614 |
0.517 |
0.539 |
0.787 |
0.817 |
2.619 |
1.290 |
PS8.genre |
0.903 |
0.898 |
0.816 |
0.976 |
0.843 |
0.888 |
0.851 |
0.896 |
2.586 |
1.960 |
[1] Davies, M., Degara N., and Plumbley M.
Evaluation methods for musical audio beat tracking algorithms.
Queen Mary University of London, Centre for Digital Music, Tech. Rep. C4DM-TR-09-06 (2009).