#DeepLearning has huge potential in #ecoacoustics, but as we don't yet fully understand the ecological role of sound, using uninterpretable, 'black box' models comes with a risk of generating confusion rather than insight, especially when analysing whole soundscapes
In order to advance, we need to understand the basis of prediction: which features are used and what do they represent? To address this issue and more we explored a novel application of variational autoencoders. 1/

