README **BirdCLEF 50** 2015 12 08

*Short description*
Identify 50 bird species from all over South America  by analyzing their songs, as recorded in the wild.

more details : 
http://sis.univ-tln.fr/~glotin/SABIOD_LIFECLEF/BIRD50/ENS_CHALLENGE_BIRD.pdf

Online scoring : 
https://challengedata.ens.fr/en/challenge/12/classify_bird_songs.html
(it ends the 01/07/2016)


* Challenge context *
Large-scale analysis of environmental audio recordings is a powerful tool for the preservation of ecological diversity. In this setting, the BirdCLEF international challenge (1000 species) aims at fostering the development of song-based automatic identification of bird species. This identification task is made difficult by the transientness of typical bird chirps, the large behavioral intra-class variability, the small amount of examples per class, the presence of wildlife noise, and so forth.
For the students of the Challenge Data, BirdCLEF has selected a realistic subset of its collection, amounting to 13 hours of audio across 50 different species.

*Challenge goals*
In this challenge, 50 species of bird were recorded in their natural environment by amateur ornithologists. The goal of the challenge is to identify species of each recording. When several species are present, the goal is to find the foreground species, that is, the closest one to the microphone. Therefore, this challenge is a single-label, supervised classification problem with 50 classes.

*Data description*
All recordings are from LifeClef 2015 challenge, which is a normalized subset of the Xeno Canto data set. Date are in WAV format at a CD sample rate of 44,1 kHz. Every recording is assigned a unique ID number and a class. IDs in the training set range from 0001 to 0932 ; IDs in the test test range from 1001 to 1375. Class are numbered from 01 to 50. The recordings last about 40 seconds on average, although their durations vary a lot between 1 second and 10 minutes. Class imbalance is also very strong.


* Data set *
The whole data set is available at : 
http://sis.univ-tln.fr/~glotin/SABIOD_LIFECLEF/BIRD50/

Several publications on this challenge are available on : 
http://sabiod.org/publications.html

* Scoring * 
The Mean average Precision is the official scoring of this challenge.
You can submit online during the competition 
several of your runs to the Kaggle like ENS system 
and see the other challenger current scores

https://challengedata.ens.fr/en/challenge/12/classify_bird_songs.html

(it ends the 01/07/2016)


Or you can directly send your runs to glotin@univ-tln.fr


* External ressources * 
External data set are prohibited

* List of the species * 
The name of each species, from number 1 to 50, into the international code, is : 
['aaogct', 'abyhkz', 'acvtjw', 'addsdr', 'aecmdb', 'aeqwqe', 'aethwv', 'aezhvb', 'aflwfz', 'afzbzd', 'ahuzrk', 'aiigbg', 'aiiwiu', 'aiwvzm', 'ajqxje', 'ajwuzt', 'aksucy', 'alouyq','alwkew', 'amsvta', 'amvexh', 'anfvsy', 'apnhqx', 'arhvxn','arlrfh', 'armfvy', 'asvgnp', 'athqge', 'atrqbr', 'aufwve','ayfewp', 'ayyyjw', 'ayzwsf', 'aziqpv', 'azkncx', 'azrobg', 'azysgp', 'bafzon', 'baleaf', 'bbhech', 'bbtkon', 'bcbkld','bcieap', 'bcmlba', 'bfwanm', 'bfwjif', 'bgaifs', 'bgcpcn', 'bgjflj', 'bgothl']


* Contact * 
glotin@univ-tln.fr