Separating Overlapping Birdsongs Enhances the Reliability of Avian Vocal Activity Analysis.

Bird vocalizations are important indicators in ecological research, used to assess population size and habitat use. Long-term monitoring of changes in bird vocalizations can also reveal patterns of environmental change affecting birds. However, recordings obtained through passive acoustic monitoring often contain overlapping calls from multiple bird species and background noise interference, reducing the reliability of automated identification and ecological analysis. This study uses an improved deep learning algorithm that dynamically processes mixed bird calls of uncertain numbers in the field through a counter and decoder, and employs three different strategies to verify the impact of the separation front-end on classification performance. We conducted a field study in the Huangmaohai Cross-Sea Channel area in southern China to verify the effectiveness of the separation algorithm. The study shows that separation effectively filters out misclassifications caused by noise or calls from other species, correcting biases in the original monitoring. By studying the diurnal variation patterns of calls of the common diurnal bird, the White Wagtail, and the nocturnal bird, the Black-crowned Night Heron, we demonstrate that separation can effectively eliminate false activity peaks of the White Wagtail in the evening and restore the masked nocturnal activity rhythms of the Black-crowned Night Heron. Separation also helps to reveal the suppression of bird vocalizations by high temperatures and precipitation, improving the statistical sensitivity of environmental factors. Furthermore, the separation results revealed the seasonal vocalization patterns of birdsong in the region. This study provides important support for sound analysis in multi-species resonance scenarios during ecological monitoring.