NIPS4Bplus: Transcriptions of NIPS4B 2013 Bird Challenge Training Dataset
datasetposted on 19.07.2019 by Veronica Morfi, Dan Stowell, Hanna Pamula
Datasets usually provide raw data for analysis. This raw data often comes in spreadsheet form, but can be any collection of data, on which analysis can be performed.
Veronica Morfi (1), Dan Stowell (1) and Hanna Pamula (2).
(1): Machine Listening Lab, Centre for Digital Music (C4DM), Queen Mary University of London (QMUL), UK
(2): AGH University of Science and Technology, Department of Mechanics and Vibroacoustics, Kraków, Poland
Version 7 Updates
Fixed: Annotation duration surpassing length of recording.
Fixed: Empty spaces at the end of some labels
Fixed: Label typos
The zip file contains 674 individual recording temporal annotations for the training set of the NIPS4B 2013 dataset in the birdsong classifications task (original size of dataset is 687 recordings).
Task and dataset description can be found in: http://sabiod.univ-tln.fr/nips4b/challenge1.html
Donwload zip file of dataset and weak annotations at: http://sabiod.univ-tln.fr/nips4b/media/birds/NIPS4B_BIRD_CHALLENGE_TRAIN_TEST_WAV.tar.gz
Transcriptions were produced using Sonic Visualiser: https://www.sonicvisualiser.org/ by an experienced birdwatcher, Hanna Pamula.
Number of missing annotations: 13 (6 of these files contained sounds which could not be unambiguously labelled and the rest 7 of them only included insects)
The original (weak) labels provided during the NIPS4B 2013 challenge were used for guidance. However, some files were judged to have a slightly different set of species present than was given in the original metadata.
Extra Unknown label was added to the dataset for the vocalisations that couldn't be classified to a specific species. Also, extra Human label was added for a few recordings that had human sounds present in them.
[Starting time (sec)],[Duration of event (sec)],[Label]