Bird 10

(Whole archive = BIRD10.tar.gz = 4.2Go)

The international tiny benchmark for bioacoustics machine listening 
Built by EADM GDR CNRS MADICS - 2016-2017
SABIOD DYNI LSIS UTLN team
contact : julien.ricard@gmail.com, glotin@univ-tln.fr
version 20161105


This dataset includes 10 amazon bird species (454 files in total, 15 secondes per files in average).
It also includes Chirplets representation as presented in [1]

- audio files, mono, are given in 2 different sample rates: 22050 and 44100 Hz
- chirplets: chirplets data in csv files or joblib (jl) dumps
- labels/labels.csv: class labels
- splits: list of files per sets (train / valid / test)

The data are extracted from the training set of LifeClef [2] 2015 / 2016 / 2017, 2000 bird species challenges, themself built from XenoCanto archives.

Please cite [1] and [2] if you use this set for your own research.

[1] H. Glotin, J. Ricard, R. Balestriero, 'Fast Chirplet Transform Enhances CNN-based Audio Classifier on Small Data', submited to ICLR217 
http://104.155.136.4:3000/pdf?id=H1Fk2Iqex

[2] H. Goëau, H. Glotin, WP Vellinga, R Planquè, A. Joly, 'LifeCLEF Bird Identification Task 2016: The arrival of Deep learning' CLEF (Working Notes) 2016: 440-449