DCASE 2022 Task 5: Few-shot Bioacoustic Event Detection Development Set

Inês Nolasco, Shubhr Singh, Ariana Strandburg‐Peshkin, Lisa Gill, Hanna Pamuła, Joe Morford, Michael Emmerson, Frants Jensens, Helen Whitehead, Ivan Kiskin, Ester Vidana-Villa, Vincent Lostanlen, Veronica Morfi, Dan StowellZenodo 2022 · DOI · OpenAlex

Abstract

<strong>General Description:</strong> The development set for task 5 of DCASE 2022 "Few-shot Bioacoustic Event Detection" consists of 192 audio files acquired from different bioacoustic sources. The dataset is split into training and validation sets. Multi-class annotations are provided for the training set with positive (POS), negative (NEG) and unkwown (UNK) values for each class. UNK indicates uncertainty about a class. Single-class (class of interest) annotations are provided for the validation set, with events marked as positive (POS) or unkwown (UNK) provided for the class of interest. <strong>this version (2):</strong><br> * fixes issues with unsorted events in the annotation files - all events from the validation set are sorted by Starttime;<br> * Removes the ML subset from the validation set.<br> <strong>Folder Structure:</strong> <em>Development_Set.zip</em> |_Development_Set/ |__Training_Set/ |___JD/ |____*.wav |____*.csv |___HT/ |____*.wav |____*.csv |___BV/ |____*.wav |____*.csv |___MT/ |____*.wav |____*.csv |___WMW/ |____*.wav |____*.csv |__Validation_Set/ |___HB/ |____*.wav |____*.csv |___PB/ |____*.wav |____*.csv |___ME/ |____*.wav |____*.csv <em>Development_Set_Annotations.zip</em> has the same structure but contains only the *.csv files <strong>## Dataset statistics</strong> Some statistics on this dataset are as follows, split between training and validation set and their sub-folders: -----------------------------------------------------<br> TRAINING SET<br> -----------------------------------------------------<br> Number of audio recordings | 174<br> Total duration | 21 hours<br> Total classes | 47<br> Total events | 14229<br> -----------------------------------------------------<br> TRAINING SET/BV<br> -----------------------------------------------------<br> Number of audio recordings | 5<br> Total duration | 10 hours<br> Total classes | 11<br> Total events | 9026<br> Ratio event/duration | 0.04<br> Sampling rate | 24000 Hz<br> -----------------------------------------------------<br> TRAINING SET/HT<br> -----------------------------------------------------<br> Number of audio recordings | 5<br> Total duration | 5 hours<br> Total classes | 5<br> Total events | 611<br> Ratio event/duration | 0.05<br> Sampling rate | 6000 Hz<br> -----------------------------------------------------<br> TRAINING SET/JD<br> -----------------------------------------------------<br> Number of audio recordings | 1<br> Total duration | 10 mins<br> Total classes | 1<br> Total events | 357<br> Ratio event/duration | 0.06<br> Sampling rate | 22050 Hz<br> -----------------------------------------------------<br> TRAINING SET/MT<br> -----------------------------------------------------<br> Number of audio recordings | 2<br> Total duration | 1 hour and 10 mins<br> Total classes | 4<br> Total events | 1294<br> Ratio event/duration | 0.04<br> Sampling rate | 8000 Hz<br> -----------------------------------------------------<br> TRAINING SET/WMW<br> -----------------------------------------------------<br> Number of audio recordings | 161<br> Total duration | 4 hours and 40 mins<br> Total classes | 26<br> Total events | 2941<br> Ratio event/duration | 0.24<br> Sampling rate | various sampling rates<br> ----------------------------------------------------- -----------------------------------------------------<br> VALIDATION SET<br> -----------------------------------------------------<br> Number of audio recordings | 18<br> Total duration | 5 hours and 57 minutes<br> Total classes | 5<br> Total events | 972<br> -----------------------------------------------------<br> VALIDATION SET/HB<br> -----------------------------------------------------<br> Number of audio recordings | 10<br> Total duration | 2 hours and 38 minutes<br> Total classes | 1<br> Total events | 607<br> Ratio event/duration | 0.7<br> Sampling rate | 44100 Hz<br> -----------------------------------------------------<br> VALIDATION SET/PB<br> -----------------------------------------------------<br> Number of audio recordings | 6<br> Total duration | 3 hours<br> Total classes | 2<br> Total events | 292<br> Ratio event/duration | 0.003<br> Sampling rate | 44100 Hz<br> -----------------------------------------------------<br> VALIDATION SET/ME<br> -----------------------------------------------------<br> Number of audio recordings | 2<br> Total duration | 20 minutes<br> Total classes | 2<br> Total events | 73<br> Ratio event/duration | 0.01<br> Sampling rate | 48000 Hz<br> ----------------------------------------------------- <strong>Annotation structure</strong> Each line of the annotation csv represents an event in the audio file. The column descriptions are as follows: TRAINING SET<br> ---------------------<br> Audiofilename, Starttime, Endtime, CLASS_1, CLASS_2, ...CLASS_N VALIDATION SET<br> ---------------------<br> Audiofilename, Starttime, Endtime, Q <strong>Classes</strong> DCASE2022_task5_training_set_classes.csv and DCASE2022_task5_validation_set_classes.csv provide a table with class code correspondence to class name for all classes in the Development set. DCASE2022_task5_training_set_classes.csv<br> ---------------------<br> dataset, class_code, class_name DCASE2022_task5_validation_set_classes.csv<br> ---------------------<br> dataset, recording, class_code, class_name <strong>Evaluation Set</strong> The Evaluation set for this task will be released on the 1st of June 2022 <strong>Open Access:</strong> This dataset is available under a Creative Commons Attribution 4.0 International (CC BY 4.0) license.<br> <strong>Contact info:</strong> Please send any feedback or questions to: Ines Nolasco - [email protected]
BibTeX
@inproceedings{nolasco2022dcase,
  author = {Inês Nolasco and Shubhr Singh and Ariana Strandburg‐Peshkin and Lisa Gill and Hanna Pamuła and Joe Morford and Michael Emmerson and Frants Jensens and Helen Whitehead and Ivan Kiskin and Ester Vidana-Villa and Vincent Lostanlen and Veronica Morfi and Dan Stowell},
  title = {{DCASE 2022 Task 5: Few-shot Bioacoustic Event Detection Development Set}},
  booktitle = {Zenodo},
  year = {2022},
  doi = {10.5281/zenodo.6477512},
}

Loading related content...

Figure