Marine ecosystems are an example of complex systems for large parts of which we still have only rudimentary understanding of underlying processes. Thus, understanding of marine ecosystems is critical to help us mitigate and limit the impact of human activities on it, especially given the rapid climate change. 

Underwater acoustics are complex and, when combined with the oceanographic and biological data the complexity increases even further. The nature of marine animal sounds is also very diverse (both in length and frequency), indicating differences between species, individuals, and context. The data characteristics are unique in the large bandwidth but also signal sparsity as at the same time.

Currently there are manual or human in the loop solutions, that are in dealing most of the time with a single dataset or species. To solve this problem with those specific properties requires specially trained models to detect and classify underwater animal sound characteristics. To achieve this we leverage the transformer model architecture, and signal processing techniques to generate adequate spectrograms for the deep learning model.The project also has co-supervisors in the Department of Biology where we get input from the experts in marine biology.