Can we use AI to connect people to wildlife while at the same time gathering important data on bird populations? That is the idea behind Chirps, a project developed by Satyan Sharma, Falk Van der Meirsch, and Tim Bauer, at Data Science Retreat Berlin.
Chirps is based on a neural network that has been trained on hundreds of hours of bird recordings from the Xeno-Canto database. Check out our github repository if you want to learn more about how it works.
Chirps currently works for 100 bird species common in central Europe. It analyzes your recordings and returns the 5 species it considers the best fit. Under good audio conditions, the correct bird is among the selection in 80% of cases. On average 59% of the top predictions are correct.
For now Chirps is able to detect the following 100 species: