Speaker
Description
Bird migration is the seasonal movement of birds between regions, often influenced by weather, food availability, and breeding needs. Passive acoustic monitoring (PAM) provides a useful way to study bird migration by recording the calls of birds passing through an area. These calls are often short and can be difficult to detect, especially when recordings contain environmental noise, overlapping sounds, or technical artefacts. In this study, we focus on the Redwing (Turdus iliacus), a migratory bird commonly detected during nocturnal migration by its characteristic flight call. We explore different approaches for automatically detecting and classifying Redwing calls in acoustic recordings. These include existing open-source bird sound recognition tools such as BirdNET, supervised methods using different preprocessing and audio representation techniques, and agile models based on Perch embeddings that support active learning. The results are discussed in terms of detection performance, practical usability, and robustness under varying recording conditions. By comparing these approaches, this study provides insight into the possibilities and limitations of automated call detection for monitoring nocturnal bird migration.