AI-Driven Communication Analysis for Biodiversity Conservation: A Fruit Bat Prototype

Abstract: This research aims to explore how Artificial Intelligence (AI) could help us understand a lot better and interpret all the unique vocalizations/communications of various different kinds of species, with the goal of supporting Biodiversity Conservation (BD). The project will develop a prototype system using vocalization data from Egyptian fruit bats. By showing that AI can successfully decode communication in one species, this work lays the groundwork for expanding such models to more complex, multi-species ecosystems in the future.

The system uses a publicly available dataset of Egyptian fruit bat calls, which includes detailed annotations such as who made the call, the context, the intended recipient, and the outcome of the interaction. To prepare the data, the audio was segmented, cleaned of background noise, and converted into sound features like MFCCs and mel-spectrograms. We then tested several deep learning models like CNNs, LSTMs, and Transformers on four tasks: 1) Identifying the caller 2) Classifying the context 3) Recognizing the recipient. and 4) Predicting the interaction’s outcome. Model performance was measured using balanced accuracy, precision, recall, and F1-score, with the results being tested for statistical and numerical significance for p < 0.05.

Inside the fruit bat communication context prediction, our model achieved 63% accuracy (In “Isolation” context our model achieved 100%accuracy).

Keywords: Animal Communication (AC), Biodiversity Conservation (BD), Deep Learning (DL), Egyptian Fruit Bat (EFB), AI for Ecology (AFE), Species Communication Translation (SCE), Wildlife Monitoring (WM)