Editor choice


AI helps ecologists identify birds by their singing

Researchers at Canada's University of Moncton have developed an AI tool called ECOGEN that creates synthetic bird songs to improve acoustic monitoring of endangered avian species. By augmenting limited audio samples of rare birds, their deep learning approach boosts identification accuracy to support conservation efforts.

Field recordings and song libraries used to train bird call recognition models often underrepresent threatened populations with minimal data. As many species face declining numbers from climate change and habitat loss, tracking their presence via audio surveys has become vital.

ECOGEN enhances these crucial datasets by algorithmically generating additional "pseudo-calls" modeled after rare species' musical signatures. Adding these AI-fabricated samples makes classification software better at detecting overlooked types in the wild based on distinctive vocal elements.

In testing across 264 species, ECOGEN-enhanced training data increased ID success by 12% on average - a breakthrough for confirming elusive subjects. The tool can apply such realistic synthetic sound generation to mammals, insects and more.

Lead researcher Dr. Nicolas Lecomte states, "With significant disruptions to animal populations, automated monitoring tools need complete acoustic reference libraries. ECOGEN fills gaps without further disturbing species-at-risk."

Importantly, the open-source ECOGEN platform runs on basic hardware for accessibility. Field biologists globally stand to refine critical population tracking efforts without new infrastructure expenditures.

The team believes creative AI data augmentation techniques will grow more important for ecologists as technology aids conservation and climate sciences. Protecting biodiversity hinges on understanding ecosystems down to individual organisms - a new frontier their sound simulation tool could help explore.

Share with friends:

Write and read comments can only authorized users