This page contains additional material for the paper:
Downbeat Tracking Using Beat-Synchronous Features and Recurrent Neural Networks
Florian Krebs, Sebastian Böck, Matthias Dorfer, and Gerhard Widmer. In Proceedings of
17th International Society for Music Information Retrieval Conference
(ISMIR), New York, USA, 2016.
The Python code of the proposed downbeat tracker can be found as part of the madmom package. To use it carry out the following steps:
- Install madmom and the models submodule (see the madmom documentation for detailed instructions)
- Checkout the branch BarTracker
- Compile the modules with Cython by running
python setup.py build_ext --inplace
in the madmom's root directory.
- Locate the script BarTracker in the bin folder. This is is what you need to run to use the downbeat tracker.
Here are some usage examples:
- For a list of available parameters use:
BarTracker -h
- To detect the beats and downbeats from an audio file, use:
BarTracker single WAVFOLDER/audio.wav
- To detect the beats and downbeats from a set of audio files, use:
BarTracker batch -o OUTFOLDER WAVFOLDER/*.wav
- You can specify the number of beats per bar, which the algorithm looks for, by:
BarTracker --beats_per_bar 2,3,4 single WAVFOLDER/audio.wav
- You can also feed beat annoations (text file, one beat time in seconds per row) and let the BarTracker identify the downbeats for you. The beat files have to have the same basename as the corresponding audio files.
BarTracker --load_beats --ext .beats batch -o OUTFOLDER WAVFOLDER/*.wav BEATFOLDER/*.beats
last edited by fk at Sep 15, 2016