PreAcoustix: A modular framework for bioacoustic signal preprocessing

Acoustic communication is widespread in animal taxa, from insects to whales, and many species use sequences of distinct sound elements for interaction. Birds and frogs exhibit unique call patterns that are valuable for species identification, behavioral studies, and ecological monitoring. Bioacoustics, the study of animal vocalizations and natural soundscapes, has emerged as a vital tool for observing species in their natural habitats. Passive acoustic monitoring (PAM) has become an essential and noninvasive method for wildlife conservation, enabling continuous data collection through autonomous recording units (ARU), especially for species that are elusive, nocturnal, or inhabit remote and inaccessible regions. These devices facilitate the acquisition of large volumes of acoustic data over extensive spatial and temporal scales. However, manual analysis of such data is labor intensive and impractical. Thus, acoustic preprocessing plays a critical role in enabling efficient acoustic analysis. This preliminary step is crucial for filtering out nonrelevant acoustics and ensuring the quality of the data used in subsequent feature extraction and machine learning applications. This study proposes PreAcoustix, a modular framework for acoustic preprocessing that can be applied to audio data collected from ARUs, encompassing downsampling, denoising, energy normalization and segmentation. This approach supports scalable and automated acoustic analysis, contributing to effective biodiversity assessment and ecological research. Effective preprocessing significantly reduces the computational and processing power needed for ARU data by eliminating approximately 89% unwanted signals, as demonstrated in the study.