Soundscape Connectomes: Unsupervised Graph-Based Approach for Soundscape Mapping

Maria J. Guerrero, Aref Einizade, Jhony H. Giraldo, Victor M. Martinez-Arias, Claudia Isaza, Cesar A Uribe

Keywords: soundscape connectomes, unsupervised graph inferece, ecoacoustics, passive acoustic monitoring

Abstract: We introduce soundscape connectomes, which are graph representations of acoustic relationships of a landscape where nodes are geographical sites and edges reflect relations derived from each site’s biophony. Soundscape connectomes are built from passive acoustic monitoring (PAM) recordings and are enabled by the unique acoustic signatures of habitats. However, in ecoacoustic analysis, ground-truth graphs or labels are often not available. We propose an unsupervised pipeline that decomposes recordings into sonotypes, builds per-site acoustic structures, infers graphs with several methods, and compares them using a smoothness-based, unsupervised criterion that scores reconstruction of held-out nodes. We apply the proposed method to a large-scale real-life data set acquired in the Colombian Andes with over 19,598 recordings on 17 sites and 292 sonotypes. Results show the stability of the generated connectomes for a variety of graph inference methods. These results provide a practical way to select a graph model without prior information and position soundscape connectomes as a complement to remote-sensing analyses for monitoring and conservation.

Submission Number: 11