Published November 1, 2025 | Version v1

Michaud_CNN_SNR_Boreal_owl_vocalizations

  • 1. ROR icon Muséum national d'Histoire naturelle
  • 2. ROR icon Centre National de la Recherche Scientifique
  • 3. ROR icon Centre d'Écologie et des Sciences de la Conservation
  • 4. ROR icon Sorbonne Université
  • 5. EDMO icon Institute of Systematics, Evolution, Biodiversity (Paris)
  • 6. ENES Bioacoustics Research Laboratory
  • 7. ROR icon Office Français de la Biodiversité

Description

First set of audio files - Raw Dataset

This dataset contains a first set of 2619 1-min audio recordings subsampled by stratification based on the singing activity of three Owl species (Strix aluco, Glaucidium passerinum, Aegolius Funereus) from a larger set of 1-min audio recordings (i.e. soundscapes) collected in a temperate cold climat forest locally named ‘Risoux’ between July 2018 and July 2024 (1 min every 15 min, day and night). The forest is located in the East of France in the Jura Mountains at the Swiss border (46°31'60.0"N 6°05'03.8"E ). The forest soundscape was recorded using four automatic SongMeter 4 recorders (Wildlife Acoustics Inc, Concord, MA, USA), spaced one kilometer apart each other in the center of the forest. 

The 2619 audio files are suplemented by 2619 annotation .txt files containing bounding boxes annotated with Audacity. The labels are:

1 = Strix Aluco

2 = Aegolius Funereus

3 = Glaucidium Passerinum

4 = other birds

0 = everything else

The composition of the soundscapes is a mix of biophony (mostly birds, but also insects and bellowing deer), geophony (e.g. rain and wind) and technophony (e.g. planes, chainsaws and possibly some gun shots and sheep bells). 

Second set of audio files specific to the Aegolius Funereus - Segmented Dataset

The second set of audio files are the 2619 1-min audio recordings segmented in 31,428 five-second audio segments. They are suplemented by three .txt files. 

  1. ALL_ANNOTATED_DATA.txt contains all annotations of the 31,428 five-second audio segments
  2. ALL_SETS.txt contains the annotations for all five-second audio segments except 1,500 empty five-second audio segments kept apart for data augmentation. The 29,928 five second audio segments are distributed in ALL_SETS.txt between the training set, the validation set and the test set. The SNR values are included in that txt file.
  3. NOISE_DATA_AUGMENTATION.txt contains the 1,500 empty five-second audio segments kept apart for data augmentation. 

 

This work was supported by the Parc Naturel Régional du Haut-Jura who received funding from the Région Bourgogne-Franche-Comté 

Files

annotations_segmented_dataset.zip

Files (18.6 GB)

Name Size
md5:85a3db8b799aa53233dc23b569637969
845.0 kB Preview Download
md5:c78c14a63ad750646280d970f7a8c676
541.5 kB Preview Download
md5:ed64790b118f68646a99dedbb847334f
9.3 GB Preview Download
md5:c86519bf6c22a49d836fd99b38c192dc
9.3 GB Preview Download

Additional details

Related works

Has part
Dataset: 10.5281/zenodo.14856482 (DOI)
Is supplement to
Dataset: 10.5281/zenodo.10701274 (DOI)