Bioacoustic monitoring of nocturnal bird migration
BirdVoxDetect: detection and classification of flight calls
BirdVoxDetect is a pre-trained deep learning system which detects flight calls from songbirds in audio recordings, and retrieves the corresponding species. It relies on per-channel energy normalization (PCEN) and context-adaptive convolutional neural networks (CA-CNN) for improved robustness to background noise. It is made available both as a Python library and as a command-line tool for Windows, OS X, and GNU/Linux.
The simplest way to install BirdVoxDetect is by using the
pip package management system, which will also install the additional required dependencies
pip install birdvoxdetect
Note that birdvoxdetect requires:
- Python (3.6, 3.7, or 3.8)
- librosa (==0.7.0)
- tensorflow (>=2)
- scikit-learn (==0.21.2)
- birdvoxclassify (0.2)
From the command line
To analyze one file:
python -m birdvoxdetect /path/to/file.wav
To analyze multiple files:
python -m birdvoxdetect /path/to/file1.wav /path/to/file2.wav
To analyze one folder:
python -m birdvoxdetect /path/to/folder
--output-dir OUTPUT_DIR, -o OUTPUT_DIR Directory to save the output file(s); The default value is the same directory as the input file(s). --export-clips, -c Export detected events as audio clips in WAV format. --export-confidence, -C Export the time series of model confidence values of events in HDF5 format. --export-faults, -f Export list of sensor faults in CSV format. --export-logger, -l Export output of Python logger in TXT format. --threshold THRESHOLD, -t THRESHOLD Detection threshold, between 10 and 90. The default value is 50. Greater values lead to higher precision at the expense of a lower recall. --suffix SUFFIX, -s SUFFIX String to append to the output filenames.The default value is the empty string. --clip-duration CLIP_DURATION, -d CLIP_DURATION Duration of the exported clips, expressed in seconds (fps). The default value is 1.0, that is, one second. We recommend values of 0.5 or above. --quiet, -q Print less messages on screen. --verbose, -v Print timestamps of detected events. --version, -V Print version number.
import birdvoxdetect as bvd df = bvd.process_file('path/to/file.wav')
df is a Pandas DataFrame with three columns: time, detection confidence, and species.
Below is a typical output from the test suite (file
Time (hh:mm:ss) Species (4-letter code) Family Order Confidence (%) 00:00:08.78 SWTH Turdidae Passeriforme 70.14839
Please visit our website for more information on the BirdVox project: https://wp.nyu.edu/birdvox
The main developer of BirdVoxDetect is Vincent Lostanlen, scientist at CNRS, the French national center for scientific research.
For any questions or announcements related to BirdVoxDetect, please refer to our discussion group: https://groups.google.com/g/birdvox
Please cite the following paper when using BirdVoxDetect in your work:
Robust Sound Event Detection in Bioacoustic Sensor Networks Vincent Lostanlen, Justin Salamon, Andrew Farnsworth, Steve Kelling, and Juan Pablo Bello PLoS ONE 14(10): e0214168, 2019. DOI: https://doi.org/10.1371/journal.pone.0214168
As of v0.4, species classification in BirdVoxDetect relies on a taxonomical neural network (TaxoNet), which is distributed as part of the BirdVoxClassify package. For more details on TaxoNet, please refer to:
Chirping up the Right Tree: Incorporating Biological Taxonomies into Deep Bioacoustic Classifiers Jason Cramer, Vincent Lostanlen, Andrew Farnsworth, Justin Salamon, and Juan Pablo Bello In IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Barcelona, Spain, May 2020.
Release history Release notifications | RSS feed
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size||File type||Python version||Upload date||Hashes|
|Filename, size birdvoxdetect-0.4.0a1.tar.gz (29.6 MB)||File type Source||Python version None||Upload date||Hashes View|