Labelled acoustic dataset of roding Eurasian Woodcock (Scolopax rusticola)
Contributors
Data collectors:
Supervisor:
Description
This dataset contains manually labelled audio data of roding Eurasian Woodcock (Scolopax rusticola).
Description
Bioacoustic surveys of roding Eurasian Woodcock were conducted in Baden-Württemberg, Germany in May and June in 2020 and 2021. The audio data of this collection was used for the evaluation of BirdNET as a means for the automated analysis of large quantities of audio data. The original dataset consisted of 12.236 minutes of recording, which were reviewed manually. Each call element of a male roding Woodcock (i.e. croak, whistle, chasing male) was annotated. Individual call elements were subsequently clustered into so called roding events, which are ecologically more meaningful. BirdNET was then tested against this manually labelled dataset.
The dataset uploaded to zenodo contains:
- audio data of 2545 woodcock call element selections with a duration of 145 minutes
- audio data of 782 aggregated woodcock roding events with a duration of 115 minutes
- selection tables for call elements and roding events
- associated metadata
Audio information in between roding events (i.e. non woodcock audio) ist not included due to data privacy reasons (see below).
Selections
Woodcock call elements were manually selected/annotated in Raven Pro with bounding boxes. For this dataset, all selections with a duration of less than 3 seconds were extended symmetrically until 3 seconds were reached. This may result in overlapping selections in the case of croaks that are directly followed by a whistle. Signals at the beginning or end of these selections may thus be included twice.
Roding events
A roding event was defined as a continuous series of Woodcock call elements with a maximum gap of six seconds between consecutive elements. Each event can be interpreted as a roding bird that passes by the recording location, similar to a typical woodcock roding survey conducted by a human observer. Roding events were not created with the extended 3 seconds clips described above, but with the original bounding box selections drawn in Raven Pro.
Audio files
- selections.zip: each wav-file contains a single selections. Filenames correspond to the column selec in the table selections.csv
- events.zip: each wav-file contains a single roding event, typically consisting of multiple call elements (croaks and/or whistles). In the case of faint signals of distant birds, roding events may consist of a single call element only. Filenames correspond to the column event.id in the table events.csv.
Data collection
All wav-files in this dataset originate from audio files that were recorded with autonomous recording units of the type AudioMoth. ARUs were housed in custom made waterproof casings (See details and files for 3D-printing: https://www.thingiverse.com/thing:6428228). ARUs were programmed to record continuously for 2 hours during dusk and were placed at edges of forest clearings. The devices were mounted to tree trunks at a height of approximately 1.5m above ground.
Metadata files
filename | content |
sites.csv |
contains locations of the recording sites. Since exact recording locations can not be made public, only recording sites (= cells of the 1km² UTM-grid) are provided. CRS: EPSG - 25832, ETRS89 / UTM 32N Data source of the underlying ETRS89 UTM 32N grid: https://gdz.bkg.bund.de/index.php/default/digitale-geodaten/nicht-administrative-gebietseinheiten/geographische-gitter-fur-deutschland-in-utm-projektion-geogitter-national.html columns site.id = unique id of recording sites, cellcode = official cellcode of the 1km²-UTM-grid elevation = mean elevation a.s.l. x.centroid = x-coordinate of centroid (EPSG: 25832) y.centroid = y-coordinate of centroid (EPSG: 25832) wkt.geometry = polygon geometry of the grid cell |
arus.csv |
metadata of the recording hardware
columns aru.id = unique id of recording device type = recorder type manufacturer = manufacturer of recording hardware hardware.version = hardware version of the recording device acquisition.date = date the device was purchased (for reasons of microphone degradation) |
deploys.csv |
information on recorder deployment, includes aru settings, location, recording times
columns deploy.id = unique id of recorder deployment aru.id = unique id of deployed aru start.date = date the aru was deployed in the field (YYYY-MM-DD) end.date = date the aru was collected (YYYY-MM-DD) firmware = firmware version used in this deployment rec.periods = number of daily recording periods (corresponds to start.rec1, start.rec2 ...) sample.rate = sample rate in kHz gain = gain setting sleep.duration = duration off stand-by phases in seconds, when set on a sleep/record-cycle rec.duration = duration of each recording in seconds, when set on a sleep/record-cycle start.rec1 = start of first recording period (UTC, hh:mm:ss) end.rec1 = end of first recording period (UTC, hh:mm:ss) start.rec2 = start of secondrecording period (UTC, hh:mm:ss) end.rec2 = end of second recording period (UTC, hh:mm:ss) site.id = unique id of recording site |
recordings.csv |
metadata of the audio files from which the roding events originate
columns recording.id = unique id of the recording deploy.id = unique id of aru deployment, during which the recording was made date = date on which the recording was made (YYYY-MM-DD) time = time of day at which the recording started (UTC, hh:mm:ss) duration = duration in seconds sampler.rate = sample rate in kHz channels = number of channels bits = bit depth samples = number of audio samples gain = gain setting of the aru voltage = battery voltage of the aru during recording temperature = ambient temperature during recording reviewer = anonymous id of staff who reviewed the file and annotated calls
|
selections.csv |
manually labelled woodcock call elements (i.e. croaks, whistles, chases). Short selections were extended to 3 seconds by symmetrically adding time before and after the original selection. In the format of raven pro selection tables.
columns selec = unique id of the selection. Corresponds to the filename of the wav-files in the archive selections.zip deploy.id = unique id of the aru deployment during which the roding event was recorded channel = audio channel start = start of the event in seconds from the start of the recording end = end of the event in seconds from the start of the recording bottom.freq = bottom frequency of the annotation bounding box top.frequency = top frequency of the annotation bounding box species.code = species code as used by BirdNET common.name = English common name as used by BirdNET annotation = contains annotations of call elements that are pooled in the roding event. Thus typcally equal to the number of annotated call element recording.id = id of the recording this roding eventoriginates from |
events.csv |
aggregated roding events consisting of contiuous sequences of manually labelled call elements. In the format of raven pro selection tables
columns event.id = unique id of roding event. Corresponds to the filename of the wav-files in the archive events.zip channel = audio channel start = start of the event in seconds from the start of the recording end = end of the event in seconds from the start of the recording bottom.freq = bottom frequency of the annotation bounding box top.frequency = top frequency of the annotation bounding box species.code = species code as used by BirdNET common.name = English common name as used by BirdNET annotation = contains annotations of call elements that are pooled in the roding event. Thus typcally equal to the number of annotated call element recording.id = id of the recording this roding eventoriginates from deploy.id = unique id of the aru deployment during which the roding event was recorded |
removed_audio_files.txt | selection ids and event ids of audio files that were deleted because they included voices. Their metadata is still included in the files described above |
Data privacy
Selections and roding events were checked for human voices and audio information was removed, in case it contained any. Audio segments that did not contain woodcock calls were not completely checked for human voices and can thus not be made available.
Files
selections.csv
Files
(1.0 GB)
Name | Size | Download all |
---|---|---|
md5:fced44eb9eb0da0410484e3d7702aca1
|
2.2 kB | Preview Download |
md5:96e42ef0d4707daeca0ca88a8c1cccb9
|
6.3 kB | Preview Download |
md5:4c25206f6551798de7a05474a099791b
|
102.7 kB | Preview Download |
md5:6d55ceffa54c12021b252d70a84e418d
|
33.6 kB | Preview Download |
md5:3e20de844ec237ef702900c5203112c2
|
207 Bytes | Preview Download |
md5:09dbf499229f30be689e3e947f54d2e6
|
442.1 MB | Preview Download |
md5:5202477c5432f06b4675f278e66e2b35
|
395.8 kB | Preview Download |
md5:3ae98e81f5002ebb5f40525b257a3e8c
|
559.2 MB | Preview Download |
md5:156ba49ab954d33ffe30c48b02b9de5c
|
12.6 kB | Preview Download |
Additional details
Funding
- Ministerium für Ländlichen Raum und Verbraucherschutz Baden-Württemberg
Dates
- Collected
-
2020/2021data collection
- Submitted
-
2024-04submission to zenodo