Deforestation causes up to 80% of vertebrate species loss. While there are conservation efforts to promote responsible lodging, existing techniques (e.g., tree coverage) do not track biodiversity effects. This is mostly because objectively tracking large amounts of taxa is difficult and expensive.

Consequently, there is a pressing need to develop new methodology to infer biodiversity metrics. Our insight is that most species are more easily heard than seen. Hence we are investigating novel machine learning techniques to infer biodiversity from sound recorded in tropical forests. Here is a comparison of two soundscapes, recorded in a forest with reduced biodiversity, and an untampered forest. Notice the higher degree of variability in the later, reflecting healthier diversity.