Leveraging tropical reef, bird and unrelated sounds for superior transfer learning in marine bioacoustics

• DOI: 10.48550/arxiv.2404.16436
• Publication Date: 2024-04-25
• Authors: Ben Williams; Bart van Merriënboer; Vincent Dumoulin; Jenny Hamer; Eleni Triantafillou; Abram B. Fleishman; Matthew McKown; Jill E. Munger; Aaron N. Rice; Ashlee Lillis; C White; Catherine Hobbs; Tries B. Razak; Kate E. Jones; Tom Denton
• Keywords: Bioacoustics; Transfer of learning

Machine learning has the potential to revolutionize passive acoustic monitoring (PAM) for ecological assessments. However, high annotation and compute costs limit field's efficacy. Generalizable pretrained networks can overcome these costs, but high-quality pretraining requires vast annotated libraries, limiting its current applicability primarily bird taxa. Here, we identify optimum strategy a data-deficient domain using coral reef bioacoustics. We assemble ReefSet, large library of sounds, though modest compared libraries at 2% sample count. Through testing few-shot transfer performance, observe that on audio provides notably superior generalizability ReefSet or unrelated alone. our key findings show cross-domain mixing which leverages bird, during maximizes generalizability. SurfPerch, network, strong foundation automated analysis marine PAM data with minimal costs.