BangladeshiBirdsCall: A Comprehensive Dataset for Real-time Detection of Bangladeshi Birds by Their Call

Urbanization and rapid environmental change impact the natural habitats of various living species, including birds, putting avian biodiversity at risk throughout the world. This study addresses the critical challenge of reliably identifying and classifying bird species in urban and rural areas, with a particular emphasis on Bangladeshi bird species. To solve this challenge, we present the development of an audio solution, Birds Sound Classification, to accurately identify and classify bird species in Bangladesh, a biodiversity-rich region with limited resources collected in an organized manner for bird calls. To overcome this, we propose the BD Birds Song dataset, a carefully selected collection of more than 25 hours of annotated audio recordings covering 23 unique Bangladeshi bird species. This dataset captures a variety of acoustic habitats, including rural, urban, and forest environments, ensuring robustness against background noise and overlapping calls. We apply BirdsSoundClassification model, a pretrained WAV2VEC2 model to evaluate the quality of our dataset, achieving a classification accuracy of 97% across the 23 species, with strong performance in precision, recall, and F1-score. This high accuracy demonstrates the potential of the model in identifying Bangladeshi bird species and supports its use in avian biodiversity monitoring and conservation efforts.