The sound(ness) of data: From audio separation to ecosystem monitoring with data-driven machine learning

This talk explores the intersection of audio signal processing and machine learning, showcasing how data-driven approaches are transforming our ability to interpret and act upon acoustic information for understanding complex environments. Bruno Masiero (Unicamp, Brazil) and Nilesh Madhu (ASPIRE, IDLab, Ghent University) will present a journey from the fundamentals of audio source separation and analysis - where machine learning models disentangle overlapping sound sources to detect and characterise sounds of interest - to real-world applications in environmental and ecosystem monitoring. Drawing on their complementary expertise, the speakers will highlight recent advances in deep learning architectures for audio analysis, discuss challenges in training models with limited or noisy data, and demonstrate how these techniques are being applied in a wide range of fields, from healthcare to communications, from monitoring biodiversity to assessing ecosystem health. The talk will also touch on the practical considerations of deploying such technologies in the wild, and the importance of interdisciplinary collaboration in building robust, scalable solutions. Whether you're interested in the technical underpinnings of audio ML or its broader implications for sustainability and conservation, this session offers a compelling look at how the "soundness" of data can lead to sound decisions in acoustic contexts.