Labeled Dataset of Dolphin Vocalizations Recorded During Structured Activities

This dataset contains a curated collection of vocalizations from seven bottlenose dolphins (Tursiops truncatus) recorded in a controlled environment at Oltremare marine park (Riccione, Italy) in November 2021. The data were collected continuously over a 24-hour period using a passive acoustic monitoring system installed at the bottom of the main pool. A total of 3,111 dolphin whistles were manually identified and extracted by a trained PAM analyst. Each whistle is provided as both a normalized WAV file and a high-resolution spectrogram image. In addition, an automated detection algorithm identified 1,277 pulsed vocalizations, including echolocation click trains, burst-pulse sounds, and feeding buzzes. All vocalizations are organized in relation to the dolphin activity type (e.g., training, play, feeding, rest), allowing for context-aware research. The dataset includes the full set of raw recordings (WAV format, 5-minute segments), labeled audio clips, spectrograms, and a structured spreadsheet summarizing vocalization counts and metadata per time block. This resource is intended to support research in marine bioacoustics, animal behavior, and machine learning, providing an accessible foundation for acoustic classification, behavior prediction, and the development of AI-based analysis tools.