Skip to main content

Functions for working with this data repository:

Project description

Build Status DOI PyPI version License


Functions for working with data from the following repository:

The repository contains .wav files of Bengalese finch song from ten birds and annotation for the songs in .xml files.

This repository provides a great resource, and was used to benchmark a sliding window-based neural network for segmenting and labeling the elements of birdsong, as described in the following paper:
Koumura, Takuya, and Kazuo Okanoya.
"Automatic recognition of element classes and boundaries in the birdsong with variable sequences."
PloS one 11.7 (2016): e0159188.

The code for the network can be found here:

The original code was released under the GNU license:

The koumura module is used with the crowsetta package to make the repository a dataset available in the hybrid-vocal-classifier and vak libraries.

It's called koumura because that's the last name of the first author on the paper, and because I am too lazy to type PyBirdsongRecognition.


$ pip install koumura


The main thing that koumura gives you is easy access to the annotation, without having to deal with the .xml file format.

To access the annotation in the Annotation.xml files for each bird, use the parse_xml function.

>>> from koumura import parse_xml
>>> seq_list = parse_xml(xml_file='./Bird0/Annotation.xml', concat_seqs_into_songs=False)
>>> seq_list[0]
Sequence from 0.wav with position 32000 and length 43168
>>> seq_list[0].syls[:3]
[Syllable labeled 0 at position 2240 with length 2688, Syllable labeled 0 at position 8256 with length 2784, Syllable labeled 0 at position 14944 with length 2816]  

Notice that this package preserves the abstraction of the original code, where syllables and sequences of syllables are represented as objects. This can be helpful if you are trying to replicate functionality from that code.
Importantly, each song is broken up into a number of "sequences". You can set the flag concat_seqs_into_songs to True if you want parse_xml to concatenate sequences by song (.wav file), so that each Sequence is actually all the sequences from one song.
If you are using the annotation to work with the dataset for some other purpose, you may find it more convenient to work with some other format. For that, please check out the crowsetta tool, that helps with building datasets of annotated vocalizations in a way that's annotation-format agnostic.

The koumura package also provides a convenience function to load the annotation for an individual song, load_song_annot. This is basically a wrapper around parse_xml that filters out the songs you don't want.

>>> from koumura import load_song_annot
>>> wav1 = load_song_annot(wav_file='1.wav')
>>> print(wav1)                                                                                                  
Sequence from 1.wav with position 32000 and length 214176  

Getting Help

Please feel free to raise an issue here:


BSD License.


If you use this package, please cite the DOI: DOI

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

koumura-0.2.1.post1.tar.gz (11.0 kB view hashes)

Uploaded Source

Built Distribution

koumura-0.2.1.post1-py3-none-any.whl (10.2 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page