Contribution
Elevoc: Towards a novel methodology for encoding vocal communication in elephants
* Presenting author
Abstract:
Understanding the vocal communication of elephants helps ensuring their well-being in an environment that is increasingly dominated by humans. Elephants are thought to use their vocal system in sophisticated ways to convey information to their conspecifics. Previous research enabled the identification of individuals by their calls, yielded insights into the animals’ vocal repertoire, and found links between their vocalizations and arousal levels. However, decoding and interpreting the precise meaning of elephant vocal communication patterns remains challenging.While the current understanding of elephant communication is largely based on the manual analysis of pre-segmented sounds, we aim at extracting information by employing AI. Specifically, we adopt models of human sound production and perception to reflect current biophysical knowledge about elephants. We then use those models to mine acoustic patterns that potentially carry relevant information in our dataset, which comprises about 10,000 fully annotated and categorized elephant calls. Based on these patterns and on the adopted models, we synthesize naturalistic elephant calls, thereby deriving a framework to iteratively test the perceptibility and the producibility of the patterns, in the lab. Finally, we use synthesized calls in playback experiments with wild elephants to better understand the importance of high-level parameters, thus validating our approach.