El Silencio Multi-Taxonomic: A passive acoustic monitoring dataset for multi-taxonomic identification in the Middle Magdalena, Colombia

Description: 

We introduce El Silencio, a bioacoustics dataset designed to benchmark multi-taxonomic identification using passive acoustic monitoring. The dataset features 770 one-minute soundscape recordings captured by AudioMoth recorders in the Middle Magdalena Medio region of Colombia between March and August 2023. These recordings, totalling 12.8 hours, were collected from seven sites with varying forest compositions, with a focus on dawn and dusk periods when biological activity is highest. The dataset contains ~15k strongly labelled bounding boxes representing four different taxonomic groups. El Silencio was also used as a test set for the BirdCLEF+2025 Challenge. By making this dataset available, we aim to foster collaboration between machine learning researchers and ecologists to better understand how global change is impacting biodiversity.

Dataset Structure & Temporal Coverage:

• Hxx.zip: Raw audio data collected in the field, organised into sub-folders representing each monitoring site. The audio .wav files follow the naming convention {site}_{date}_{time}.wav. There are 7 folders, each containing 110 files.
• strong_labels.csv: High-level annotations featuring bounding boxes of species-specific calls and sonotypes. This file utilises robust measurements (e.g., 5% and 95% frequency bounds) to ensure reproducible acoustic analysis. The group column defines the broad taxonomic class, while the label column provides the specific 6-character species or standardised sonotype code. Caution when using the "UNKNOWN" label, as it is not a stable class. It represents mostly several birds that were not identified. Community contributions are highly encouraged. If you can identify any of these species or spot a correction, please reach out so we can update the dataset. 
• taxonomy_dict.csv: A reference dictionary mapping the 6-character codes found in the annotations to their highest known taxonomic resolution. Identified species are listed by their scientific names. Sounds that could not be identified to the species level are grouped by taxonomic class (e.g., AVE (BIRD), INS, ANU) and assigned a unique sonotype ID (e.g., INS_SON022).