Identifying Bird Calls in Soundscapes Using Convolutional Neural Networks

Birds are crucial for maintaining ecosystem balance and detecting environmental threats. Therefore, monitoring bird populations is essential for understanding ecosystem health. However, many birds are isolated in high-elevation habitats, making it challenging for researchers to study and monitor their populations. As a result, detecting birds based on their sounds can provide a passive, low-labor approach to monitoring bird populations. This research aims to develop a deep-learning model that accepts an audio waveform of arbitrary length and then acoustically recognizes the species. Toward this end, a subset of the BirdCLEF 2022 dataset was used. This dataset covers 152 bird species. However, four species were selected based on their ecological significance and data availability, allowing comparison and evaluation of different strategies and techniques. The developed model achieved an 89.66% macro F1 score on the test set. This promising result suggests that the techniques utilized in the model development process were effective and could be applied in similar contexts.