Published April 16, 2025
| Version v1
Dataset
Open
BioDCASE 2025 Task 3: Bioacoustics for Tiny Hardware Development Set
Authors/Creators
-
1.
University of South Bohemia in České Budějovice
-
2.
Aarhus University
-
3.
Charles University
Description
General Description
The development set for task 3 of BioDCASE 2025 "Bioacoustics for Tiny Hardware" consists of 2 hours and 37 mins worth of audio recordings.
Yellowhammer songs were initially recorded with AudioMoth recorders placed near the birds favorite singing posts less then 5m from the singing birds. The best quality songs were selected and put into one track. The track contained in total 209 songs from 10 individuals, including in total 21 different song types (2-3 song types per male), with c.a. 10 repetitions of each song type.
The track was then played back and re-recorded in 7 different distances ranging from 6.5m to 200m. Note that the signal-to-noise ratio (SNR) is quite low in the last two distance categories and the songs that were not visible on the spectrogram were not included in the dataset which means that these distance categories may have less than 209 samples. Playback was done in two different environments (forest and grassland). Altogether, each song is repeated up to 15 times in the dataset (original distance + 7x forest + 7x grassland)
The dataset is split into training and validation sets. The recordings containing Yellowhammer vocalizations were split based on individuals: 8 individuals are in the training set, 2 are in the validation set and the remaining 2 are reserved for evaluation.
Folder structure
* Files under `Yellowhammer/` contain Yellowhammer vocalization (target species for detection)
* Files under `Negatives/` contain negatives samples which include vocalization from other species, as well as background noise recordings extracted for the full recordings.
|_Development_Set/
|__Training_Set/
|___Yellowhammer/
|____*.wav
|___Negatives/
|____*.wav
|__Validation_Set/
|___Yellowhammer/
|____*.wav
|___Negatives/
|____*.wav
File name format
The file names under 'Yellowhammer/' follow the following format 'YH_SongID_Location_Distance.wav':
- SongID: a 3 digit identifier for each song
- Location: 'forest', 'grassland' or 'speaker' for recordings that were part of the track played with the speaker
- Distance: A (original) - B - C - D - E - F - G - H
The file names under 'Negatives/' the following format 'Type_ID.wav':
- Type: 'Background' for background noise or 'bird' for recordings containing vocalizations from other species
- ID: a file identifier
Dataset statistics
-----------------------------------------------------
TRAINING SET
-----------------------------------------------------
Number of audio recordings | 3562
Total duration | 1 hours and 58 mins
-----------------------------------------------------
TRAINING SET/Yellowhammer
-----------------------------------------------------
Number of audio recordings | 1321
Total duration | 44 mins
Sampling rate | 16000 Hz
Recording duration | 2s
-----------------------------------------------------
TRAINING SET/Negatives (known species)
-----------------------------------------------------
Number of audio recordings | 956
Total duration | 31 mins
Sampling rate | 16000 Hz
Recording duration | 2s
-----------------------------------------------------
TRAINING SET/Negatives (background noise)
-----------------------------------------------------
Number of audio recordings | 1285
Total duration | 42 mins
Sampling rate | 16000 Hz
Recording duration | 2s
-----------------------------------------------------
VALIDATION SET
-----------------------------------------------------
Number of audio recordings | 1156
Total duration | 38 mins
-----------------------------------------------------
VALIDATION SET/Yellowhammer
-----------------------------------------------------
Number of audio recordings | 408
Total duration | 13 mins
Sampling rate | 48000 Hz
Recording duration | 0.5s - 3s
-----------------------------------------------------
VALIDATION SET/Negatives (known species)
-----------------------------------------------------
Number of audio recordings | 319
Total duration | 10min
Sampling rate | 16000 Hz
Recording duration | 2s
-----------------------------------------------------
VALIDATION SET/Negatives (background noise)
-----------------------------------------------------
Number of audio recordings | 429
Total duration | 14 mins
Sampling rate | 16000 Hz
Recording duration | 2s
License
This dataset is available under a Creative Commons Attribution 4.0 International (CC BY 4.0) license.
Contact
For further information or questions, please contact: [imorandi@prf.cju.cz].