On Tuesday, Google and Cornell University kicked off a competition that should excite those in the overlapping bit of the birding-and-AI-enthusiast Venn diagram. The Cornell Lab of Ornithology and Google’s bioacoustics team are working together to develop machine learning tools to help identify birds.
Participants will have to create AI models to identify bird calls from complex audio samples. Three winners will share $25,000 and assist in the bioacoustic conservation community in its research and understanding of various ecosystems.
AI for the canaries in the coal mine — The competition serves to effectively separate individual aspects of nature recordings. Traditionally, people are a necessary component in sifting out the bird call of one species in an audio file overflowing with other animal sounds and the noise pollution of a modern world. Bird calls have proven to be highly valuable when assessing the quality of life for animals in a given ecosystem. An improved ability to classify different sound markers would allow for more widespread use of AI models in ornithological research and conservation efforts.
Shouting to be heard — Competitors will have to evaluate recordings from three different sites that are filled with many, often overlapping bird calls. Code must be submitted through Notebooks on competition platform Kaggle by September 15, but the last day to officially enter the competition is September 8. The prize money will be split across three individuals or teams: $12,000 for first place, $8,000 for second place, and $5,000 for third place.
In addition to Google’s bioacoustics team — which falls under its AI for Social Good initiative — the contest’s data sets were collected with the help of researchers from The Ayres Lab at Dartmouth College, The California Academy of Sciences, The Patricelli Lab at the University of California, Davis, and Xeno-canto’s crowdsourced bird call database. Just be careful if you're darker-skinned, live in New York, and want to test out your solution in Central Park, please.