_det.png)
To secure the survival and well-being of elephants in an increasingly human-dominated environment, it is necessary to have a deeper understanding of their behaviour, cognition, perception, and communication. African elephants (Loxodonta africana) communicate extensively using the vocal modality. They transmit biologically relevant information via their calls, yet, each call is unique, differing in multiple acoustic parameters in the time- and frequency-domain. To decipher the language of elephants, it will be necessary to work with extensive data sets that can hardly be manually or semi-automatically analysed by humans.
Currently, it is unknown which acoustic patterns encode relevant information. Our central question is whether Artificial Intelligence (AI) can help to decode elephant communication. We will approach this question using advanced acoustic models combined with machine learning (ML) techniques on the largest dataset of annotated/curated African savannah elephant vocalizations to date. Our primary objective is to develop a computational model that enables identifying acoustic cues relevant for elephant communication in a combined data- and knowledge-driven process. This interdisciplinary project between biologists and computer scientists tackles the challenge of decoding elephant communication for the first time. We will develope a computational model for elephant sound production and hearing and will evaluate its validity in the field on elephants in the wild.
Combining advanced models on ML and AI to decipher an animal communication system (in our case elephants) and verifying our findings not only in the lab, but on elephants in the wild, providing the highest level of authentication, distinguishes this project as exceptional.