From Brain Connectomes to Soundscape Connectomes: A Graph-based Unsupervised Approach for Acoustic Heterogeneity Identification

Graph-based approaches have proven invaluable for uncovering complex relationships across scientific fields, notably in neuroscience for constructing brain connectomes, comprehensive maps of neural pathways that enhance our understanding of cognition and disease. Inspired by these methods, we propose an unsupervised framework to build “soundscape connectomes,” capturing relationships, similarities, and differences across diverse geographic sites by leveraging biophonic information. Our approach decomposes audio recordings from passive acoustic monitoring (PAM) into sonotypes, unique acoustic entities characterized by specific time-frequency features. These sonotypes form a site-specific “acoustic footprint,” which feeds into graph inference algorithms, where nodes represent locations and links encode acoustic similarities. Since our goal is to uncover non-obvious links among locations, the absence of labeled data presents challenges for reliably training state-of-the-art graph models. To address this, we employ flexible initialization strategies, propose an unsupervised evaluation metric for model selection, and implement stability-focused mechanisms to preserve consistent core links. This framework was tested on PAM datasets collected from Colombian regions, each featuring varying vegetation covers. The resulting soundscape connectomes reveal ecological patterns consistent with expected land cover types and expose relationships that may remain overlooked by standard remote-sensing techniques. This advantage arises from acoustic signals’ ability to capture subtle, understory-level changes, such as shifts in the presence or vocal activity of biophonic sonotypes, providing an added layer of ecological insight. Our proposal offers a promising avenue for identifying nuanced biodiversity patterns and guiding data-driven conservation strategies by extending connectome principles from neuroscience to soundscapes.

• Autores: Guerrero Muriel, Maria José; Isaza Narváez, Claudia Victoria; Einizade, Aref; Giraldo Zuluaga, Jhony Heriberto; Daza Rojas, Juan Manuel; Uribe Meneses, César Augusto
• Tipo de recurso: http://purl.org/coar/resource_type/c_6670
• Fecha de publicación: 2025
• Institución: Universidad de Antioquia
• Repositorio: Repositorio UdeA
• Idioma: eng
• OAI Identifier: oai:bibliotecadigital.udea.edu.co:10495/47732
• Acceso en línea: https://hdl.handle.net/10495/47732
• Palabra clave: Ecosystem sounds; Acoustic signatures; Biophony; Acoustic heterogeneity; ODS 15: Vida de ecosistemas terrestres. Proteger, restablecer y promover el uso sostenible de los ecosistemas terrestres, gestionar sosteniblemente los bosques, luchar contra la desertificación, detener e invertir la degradación de las tierras y detener la pérdida de biodiversidad
• Rights: openAccess
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/