Machine/Deep Learning categorisation of sub-kilohertz Arctic soundscapes

Arctic soundscapes are being modified by climate change, which is greatly amplified in the region. Cryophony (sounds from sea ice processes) will become more variable as ice floes become increasingly fragmented. These changes to the sea ice will also result in shifts in temporospatial patterns of marine mammal vocalisations and anthropogenic sounds. These markers of the state of the Arctic Ocean are monitored using passive acoustic technologies, however there are still no standard practices for exploring soundscapes in this region. Here we investigate Machine/Deep Learning (ML/DL) approaches for categorising deep-water Arctic soundscapes. Recordings from hydrophone moorings deployed along the Nansen Basin during the “Coordinated Arctic Acoustic Thermometry Experiment” (CAATEX, 2019-2020) have been considered for this study. We utilise AVES (Animal Vocalisation Encoder based on Self-Supervision) to identify sounds within recordings with broad descriptors. Training datasets for ML/DL algorithms usually consider a broader frequency range (beyond 20 kHz), but it is not always feasible to use these higher sampling rates, thus the robustness of algorithms requires testing with lower sample rate data (976 Hz here), where the frequency content of sounds is not always fully recorded. To study multiple sound sources at once (e.g. whale songs, anthropogenic sounds), we consider longer context windows (‘snippets’) of 120 seconds, currently seldom considered in acoustic ML problems. These techniques will be crucial for avoiding current bottlenecks in data processing, in particular in the Arctic, enabling more in-depth studies for marine mammal conservation and industrial regulation.