Acoustic analysis of bottlenose dolphin vocalizations for behavioral classification in controlled settings

Abstract

Understanding how bottlenose dolphins adjust their vocal behavior in response to daily routines can provide insights into social communication and welfare assessment in managed care environments. This study presents a detailed analysis of bottlenose dolphin (Tursiops truncatus) vocal behavior in relation to different daily activities within a controlled environment at Oltremare Marine Park (Riccione, Italy). 24 hours of continuous acoustic recordings were collected from seven dolphins during a typical day at the marine park, including training, feeding, playing, and unstructured activities. Signals were analyzed to quantify the variations in type and number of vocalizations in relation to dolphin activity. 3,111 whistles were manually extracted and stored as both normalized audio files and high-resolution spectrograms. Additionally, an automated algorithm identified 1,277 pulsed vocalizations, classified into echolocation click trains, burst-pulse sounds, and feeding buzzes, using signal-to-noise ratio (SNR) and inter-click interval criteria. Results revealed a significant increase in vocalization rates during structured activities compared to unstructured periods (p<0.001). This trend was consistently observed across all four vocalization types. Notably, play sessions elicited the highest rates of pulsed vocalizations (p<0.01), suggesting enhanced social and exploratory behaviors, i.e., interactions with conspecifics as well as curiosity-driven engagement with the environment. To test dataset reliability and usability, signal quality was analyzed by evaluating SNR. To support future research in marine mammal bioacoustics, behavioral ecology, and Artificial-Intelligence-based acoustic monitoring, the full annotated dataset was released as an open-access resource.