El Silencio Soundscapes: A dataset for leveraging unlabelled data in machine learning models for passive acoustic monitoring

Description: El Silencio Soundscapes is an extension of the original El Silencio Multi-taxonomic dataset, which served as a target dataset for the BirdCLEF+ 2025 challenge. This repository provides a collection of soundscape recordings captured via passive acoustic monitoring (PAM) in the Middle Magdalena region of Colombia. Dataset Structure & Temporal Coverage:The dataset comprises a total of 10,980 unlabelled audio files across 63 distinct monitoring sites, some of which geographically overlap with the labelled parent dataset. Format: 1-minute raw audio files. Uniform Site Distribution: Each of the 61 sites contributes exactly 180 recordings. Continuous Duty Cycle: The data captures a 38-day continuous monitoring window spanning from April 19 to May 26, 2023. 24-Hour Coverage: Because the autonomous recording units utilised a continuous duty cycle, the hourly distribution of the files is remarkably uniform. The acoustic data evenly represent all 24 hours of the diurnal and nocturnal cycles, free from targeted time-of-day sampling bias. Applications:This dataset is publicly available to motivate machine learning researchers to explore innovative methods for computational bioacoustics using real-world deployment scenarios. It is particularly well-suited for benchmarking models, developing deep feature extraction pipelines, and advancing self-supervised learning techniques on unlabelled acoustic data. Citation: If you find the El Silencio Soundscapes useful for your research, please consider citing the paper as follows: Cañas, J. S., Kahl, S., Denton, T., Toro-Gómez, M. P., Rodriguez-Buritica, S., Benavides-Lopez, J. L., ... & Joly, A. (2025, September). Overview of BirdCLEF+ 2025: Multi-taxonomic sound identification in the Middle Magdalena, Colombia. In Conference and Labs of the Evaluation Forum (CLEF 2025) (No. 4038, pp. 2909-2919). CEUR-WS.