Quick and dirty hack done during HAMR@ISMIR 2014 for conduction tracking of motion features loaded from a file using the bar pointer model [1] and a DBN with fitted GMMs as in [2].
References:
[1] N. Whiteley, A. T. Cemgil, and S. Godsill. Bayesian modelling of temporal structure in musical audio. In Proceedings of the 7th International Conference on Music Information Retrieval (ISMIR 2006), pages 29–34, Victoria, BC, Canada, October 2006.
[2] F. Krebs, S. Böck, and G. Widmer. Rhythmic pattern modeling for beat and downbeat tracking in musical audio. In Proceedings of the 14th International Society for Music Information Retrieval Conference (ISMIR 2013), pages 227–232, Curitiba, Brazil, November 2013.
Run python setup.py build_ext --inplace to build the dbn module.
Required packages:
- python 2.7
- cython
- numpy
- scipy
- gcc
Run export CC=gcc to switch to gcc if you get OpenMP related errors.