CoarseSoundNet: Building a reliable model for ecological soundscape analysis
4 affiliations
- 1TUM University Hospital, CHI – Chair of Health Informatics, Ismaninger Str. 22, Munich, 81675, Bavaria, Germany; MCML – Munich Center for Machine Learning, Munich, Bavaria, Germany; Corresponding author at: TUM University Hospital, CHI – Chair of Health Informatics, Ismaninger Str. 22, Munich, 81675, Bavaria, Germany.
- 2TUM University Hospital, CHI – Chair of Health Informatics, Ismaninger Str. 22, Munich, 81675, Bavaria, Germany; MCML – Munich Center for Machine Learning, Munich, Bavaria, Germany
- 3University of Freiburg, Faculty of Biology, Geobotany, Schaenzlestr. 1, Freiburg, 79104, Baden-Württemberg, Germany
- 4TUM University Hospital, CHI – Chair of Health Informatics, Ismaninger Str. 22, Munich, 81675, Bavaria, Germany; MCML – Munich Center for Machine Learning, Munich, Bavaria, Germany; Imperial College London, GLAM – Group on Language, Audio, & Music, London, UK
Abstract
A soundscape is composed of three types of sound: biophony (sounds made by animals), geophony (natural abiotic sounds) and anthropophony (sounds made by humans). A key research question in the field of soundscape ecology is how these components interact with each other, specifically how biophony responds to geophony and anthropophony. Nevertheless, as of today, there are not many analytical instruments that enable the distinct quantification of these elements. Recent machine learning (ML) approaches aim to support automated analysis but often rely on task-specific or clean data, limiting generalisation to noisy passive acoustic monitoring (PAM) recordings. This study presents a reproducible framework to build ML models for coarse soundscape classification and introduces CoarseSoundNet , a deep learning model trained to distinguish biophony, geophony, and anthropophony under realistic terrestrial PAM conditions. We systematically investigate model architectures, the influence of an additional training class, data composition, and evaluation strategies. Our findings suggest that model performance improves with additional PAM data, especially when similar to the target domain, and by introducing an explicit silence class during training. Post-hoc target-domain calibration employing class-specific decision thresholds further enhance performance, while incorporating minimum annotation durations yields higher scores for anthropophony and geophony. Error analyses exhibit challenges for anthropophony due to masking effects and confusions for silence and insect sounds for geophony and biophony. Finally, we conduct an ecological case study showing that pre-filtering recordings with CoarseSoundNet yields acoustic index trends broadly similar to ground-truth filtering, although the resulting improvements in their relationships with avian α -diversity were limited.
Related
- Assisting patients with reliable information gathering: A guide to building efficient QA systems for medical use
- A field observation method to characterize site soundscape and improve detection probability estimation for eastern bat species in North America
- Urban planning leaves an audible legacy in a megacity's bird soundscape
- Reef soundscape response to a severe convective storm event in the Northern Red Sea
- Audio embeddings identify novel sounds and track soundscape change within a tropical restoration landscape
BibTeX
@article{gebhard2026coarsesoundnet,
title = {{CoarseSoundNet: Building a reliable model for ecological soundscape analysis}},
author = {Alexander Gebhard and Andreas Triantafyllopoulos and Dominik Arend and Sandra Müller and Svenja Schmidt and Michael Scherer-Lorenzen and Björn W. Schuller},
journal = {Ecological Informatics},
year = {2026},
month = oct,
doi = {10.1016/j.ecoinf.2026.104101},
url = {https://doi.org/10.1016/j.ecoinf.2026.104101},
note = {CC BY},
}
RIS
TY - JOUR
TI - CoarseSoundNet: Building a reliable model for ecological soundscape analysis
AU - Alexander Gebhard
AU - Andreas Triantafyllopoulos
AU - Dominik Arend
AU - Sandra Müller
AU - Svenja Schmidt
AU - Michael Scherer-Lorenzen
AU - Björn W. Schuller
PY - 2026
DA - 2026/10/30
JO - Ecological Informatics
DO - 10.1016/j.ecoinf.2026.104101
UR - https://doi.org/10.1016/j.ecoinf.2026.104101
AB - A soundscape is composed of three types of sound: biophony (sounds made by animals), geophony (natural abiotic sounds) and anthropophony (sounds made by humans). A key research question in the field of soundscape ecology is how these components interact with each other, specifically how biophony responds to geophony and anthropophony. Nevertheless, as of today, there are not many analytical instruments that enable the distinct quantification of these elements. Recent machine learning (ML) approaches aim to support automated analysis but often rely on task-specific or clean data, limiting generalisation to noisy passive acoustic monitoring (PAM) recordings. This study presents a reproducible framework to build ML models for coarse soundscape classification and introduces CoarseSoundNet , a deep learning model trained to distinguish biophony, geophony, and anthropophony under realistic terrestrial PAM conditions. We systematically investigate model architectures, the influence of an additional training class, data composition, and evaluation strategies. Our findings suggest that model performance improves with additional PAM data, especially when similar to the target domain, and by introducing an explicit silence class during training. Post-hoc target-domain calibration employing class-specific decision thresholds further enhance performance, while incorporating minimum annotation durations yields higher scores for anthropophony and geophony. Error analyses exhibit challenges for anthropophony due to masking effects and confusions for silence and insect sounds for geophony and biophony. Finally, we conduct an ecological case study showing that pre-filtering recordings with CoarseSoundNet yields acoustic index trends broadly similar to ground-truth filtering, although the resulting improvements in their relationships with avian α -diversity were limited.
SN - 1574-9541
SN - 1878-0512
ER -