An efficient pipeline for the unsupervised segmentation of heterogeneous natural soundscapes

Thomas R. Napier

doi:10.1016/j.ecoinf.2026.103956

An efficient pipeline for the unsupervised segmentation of heterogeneous natural soundscapes

2026-08-05

0

OA

AI

T

Thomas R. Napier *

E

Euijoon Ahn

S

Slade Allen‐Ankins

L

Lin Schwarzkopf

I

Ickjai Lee

DOI：10.1016/j.ecoinf.2026.103956

Original

Original request for help

Save

Abstract

En 中文

• We propose a resource-efficient unsupervised framework for ecoacoustic segmentation. • Our pipeline integrates SED, MFCCs, UMAP, and HDBSCAN clustering methods. • Evaluation covers 19,230 segments across six biodiverse Australian soundscapes. • Clusters align with sound types, achieving F1-scores between 83.7% and 98.3%. • The framework reduces manual effort and supports biodiversity monitoring in large acoustic archives.

Keywords:

Clustering

Ecoacoustics

Large-scale data analysis

Passive acoustic monitoring

Unsupervised learning

An efficient pipeline for the unsupervised segmentation of heterogeneous natural soundscapes

Abstract

Journal

Organization

Cited Papers

Citing Papers