Using Computational Musicological Approaches and Informatics to Characterize Soundscapes in Diverse Natural and Human-dominated Ecosystems
The overarching objective of my dissertation research is to use computational musicology and informatics methods to characterize soundscapes in myriad ecosystems across different disturbance gradients. Three questions guided my research: (1) Can my methods place sound signals into three major sound source bins (classes) better than current approaches? (2) How can traditional metrics quantify disturbance and temporal periods within a time-dependent structure? (3) Can a multi-labeling approach allow for content discovery in a soundscape ecology database? To address these questions, I organized the dissertation into three research chapters. The second chapter, “Spectral timbral analysis for discrimination of soundscape components,” explores sound beyond traditional acoustic metrics and utilizes spectral features as described in MIR systems as a novel approach to classify dominant soundscape compositions. Current soundscape analyses consider the acoustic properties of frequency and amplitude resulting in varied metrics, but rarely focus on the discrimination of soundscape components. Computational musicologists, however, ingest similar data but consider a third acoustic property, timbre. We used recording samples from three different ecosystems from the soundscape library at the Center for Global Soundscape to demonstrate the efficacy of timbre in distinguishing among dominant soundscape components. The third chapter, “A rapid assessment monitoring framework to characterize a loud sound event stressor on a vocalizing bird community in a US Midwestern prairie,” addresses the two research questions that broach traditional metrics to quantify disturbance in an ecosystem to a rapid assessment approach in a coupled-human natural system. This chapter is the result of two years of field experience at a restored prairie in a US Midwestern prairie. We investigate how a proposed stressor-response monitoring framework could be used to quantify changes in a vocalizing community’s response to a loud sound event (LSE) at an urban historical park. The framework utilizes time-dependent data to assess an LSE stressor using a passive acoustic recorder network. The fourth chapter, “Data mining for soundscape content using a multi-label kNN approach" presents a case study for rapid assessment of disturbance in a temperate forest. The multi-label approach combines practice of knowledge discovery with music information retrieval systems using recordings from a long-term study on climate change in Tippecanoe County, Indiana. The aim was to see if soundscape content analysis using a supervised clustering method could contribute to a system that assesses the impact of altered soundscapes on wildlife communities and human systems. We propose a soundscape content analysis framework for improved knowledge outcome with assistance of the multi-label (ML) concept. Finally, the fifth chapter provides summary remarks about each chapter and the future of the field.
Pijanowski, Purdue University.
Off-Campus Purdue Users:
To access this dissertation, please log in to our