Birdwerdz is an automated birdsong recongnitzion tool based on an algorithm published by Fantana and Kozhevnikov (2014).
Birdwerdz is a Python 2 package which requires the packages Cython, numpy, scipy, matplotlib, and arf. The package incorporates a Cython extension module which should work cross platforms but has only been tested on Linux.
To install, clone this repository, cd into the repository folder, and enter this command as root:
python setup.py install
The algorithm uses dynamic time-warping to find potential matches it a recording to a given template. These matches represent local minima in the space of mappings between the template recording and the recording to be analyzed. This set of matches contains all of the examples of the template vocalization, as well as many false positives. In the second step of the algorithm a simple cluster analysis is performed on the potential matches in order to obtain the final results.
The command-line interface works with hdf5 files saved according to the arf standard. Your recordings should be saved as arf datasets. Make sure that the datatype attribute of these datasets is set to 1 (the code for audio data), so that the program knows which recordings in the file to analyze. The arf file should also contain a template recording, and a label dataset which labels events or intervals in the template.
To use the birdwerdz command-line interface, enter commands of the form "birdwerdz [command]." The available commands are "classify", 'cluster", "label", "plot", and "select." Entering "birdwerdz -h" will list these commands and their descriptions. Entering "birdwerdz [command] -h" will print the documentation for each command.
Reference:
Fantana, A. L., & Kozhevnikov, A. (2014). Finding motifs in birdsong data in the presence of acoustic noise and temporal jitter. Behavioral Neuroscience, 128(2), 228–236. http://doi.org/10.1037/a0035985