Automated data analysis using machine learning offers significant advantages for nature conservation, biology, and bird watching when it comes to identifying bird species through sound recordings (acoustic biomonitoring). In the past, bird song experts had to check each individual recording. Now, automated bird song recognition can at least provide an initial impression.

To make this work, Dawn Chorus has integrated the BirdNET analysis software from Chemnitz University of Technology and the Cornell Lab for Ornithology into the Dawn Chorus app. BirdNET uses machine learning to automatically recognize and classify bird songs. Currently, BirdNET can recognize around 6,000 bird species based on their songs and calls, and it is constantly learning more.

Challenges in automatic bird song recognition

However, birds – like humans – have varied voices and different dialects. And even software sometimes finds it difficult to recognize bird calls without error when several birds are singing “all at once.” Or when ambient noise overlaps the recording.