PteroSet
Authors/Creators
-
Ruiz, Daniela
(Contact person)1
-
Ulloa, Juan Sebastian2
-
Miao, Zhongqi1
- Betancourt, Nicolás2
-
Hernández, Andrés1
-
Demuro, Bruno1
-
Barona Cortés, Eliana2
-
Toro Gómez, Maria Paula2
-
Mendoza-Henao, Angela M.2
-
Sierra-Ricaurte, Andres Felipe2
-
Pérez-Peña, Sebastian Camilo3
-
Dodhia, Rahul1
-
Arbelaez, Pablo4
-
Lavista Ferres, Juan M.1
Description
This release provides a curated passive acoustic monitoring dataset of Neotropical bird vocalizations recorded in Colombia (Puerto Asís, Putumayo; Pijavay, Magdalena) between 2023 and 2025.
Audio was collected using autonomous AudioMoth recorders and stored as time-lapse WAV recordings, distributed across the five project folders (MAP1.zip, PPA1.zip, PPA2.zip, PPA3.zip, PPA4.zip). The collection comprises 563 recordings totaling 73.62 hours at a 192 kHz sample rate. A companion segment_manifest.csv is provided to identify and extract the original 10-second recording segments from each time-lapse file. The dataset contains 15,372 curated bird-event annotations at the taxonomic-group level, of which 6,702 include species-level determinations using standardized species codes in the species.csv lookup table.
Each recording is paired with expert-generated annotations created in Raven Pro (labels.zip) and subsequently harmonized into a COCO-inspired JSON schema for bioacoustics. These annotations are provided in two versions: annotations_identification.json, containing all annotations, and annotations_species.json, containing only species-level annotations. Annotations are provided as strong labels with explicit temporal boundaries (t_min, t_max) and spectral bounds (f_min, f_max), enabling time-frequency localization and event detection tasks.
Deployment details are distributed in metadata.csv. Additionally, model weights are included in checkpoints.zip, enabling benchmarking of bioacoustic models on this dataset.
All code used for data processing and technical validation is available at microsoft/PteroSet to ensure reproducibility. A detailed description of the methodology is provided in the accompanying preprint. Additional non-annotated recordings from these deployments are available upon request.
Files
annotations_identification.json
Files
(86.4 GB)
| Name | Size | |
|---|---|---|
|
md5:7e7ed085752da3ae6fd126c608b3b966
|
4.9 MB | Preview Download |
|
md5:fefae0bd5b9cad4ead154ebf4dfeaf37
|
2.3 MB | Preview Download |
|
md5:e66f6ad94a1bb2f992fd2753c4de6072
|
619.3 MB | Preview Download |
|
md5:91457497ae9692708c51171de5a8237d
|
622.6 kB | Preview Download |
|
md5:499ce381754892ebd8311433fd8bc99c
|
6.9 GB | Preview Download |
|
md5:23ce32af48bf7af1e90bcd6bf5a18e94
|
108.0 kB | Preview Download |
|
md5:d5fce3e9d743bad4988131b57aa1c9bc
|
15.3 GB | Preview Download |
|
md5:42b4d4c7c27c821beab86dd51dc2204a
|
21.6 GB | Preview Download |
|
md5:c8e4d32ec0ccac4fd0928ee4dbc91abe
|
23.5 GB | Preview Download |
|
md5:aba69c06c5b6d8f749e56d1c85150898
|
18.5 GB | Preview Download |
|
md5:78f95db6260750923c1dd7b8a3b56a8e
|
3.2 MB | Preview Download |
|
md5:ab0dd4b6ecfd526c749b86bd6a5e71cc
|
6.6 kB | Preview Download |