Bird Vocalization Monitoring Work at Anhui University（July 3rd)

Recently, I went to the School of Resources and Environmental Engineering, Anhui University, to assist postgraduate students with their ecological research. The core task was to build acoustic data models for bird vocalization monitoring, focusing on Shibalianwei Wetland and other typical wetlands across Anhui Province.

We adopted Raven, a professional sound analysis program developed by the Cornell Lab of Ornithology, to process field audio recordings. This research aims to explore the correlation between bird vocalizations and their behavioral patterns. We followed a two-step identification workflow: At first, the program applied built-in AI algorithms to automatically detect and classify bird sounds; then, I conducted manual rechecks to revise inaccurate AI recognition results and guarantee data accuracy.

During manual verification, I summarized four common flaws of the AI recognition system. First, ambient noises from transportation tools and electronic facilities such as alarms were frequently misidentified as bird calls. Second, overlapping chirps of multiple bird species easily triggered recognition errors—the AI either confused mixed sounds with a single wrong bird species or only detected one type of vocalization among all singing birds. Third, low-frequency bird calls were hard for the algorithm to capture and often escaped automatic identification. Fourth, several bird sound samples obtained accurate recognition results, yet the AI output unreasonably low confidence scores.

This hands-on work helped me understand both the value and limitations of bioacoustic monitoring software. Proper manual review is indispensable to optimize ecological data models and improve the reliability of wetland bird research.

The lovebird of one of the students is fierce.