Bottlenose Dolphin Vocalizations in Controlled Environments: A Dataset for Behavioral Classification

The present dataset includes a set of acoustic recordings of common bottlenose dolphins (Tursiops truncatus) acquired during a specific training activity. Data was recorded at the Oltremare Marine Park in Riccione (Italy), which hosted 7 bottlenose dolphins (2 males and 5 females). 
The dataset contains:
One .zip file named “Raw_recordings_Day1_1” which contains 3 folders named “Raw_recordings_Day1_pti” i :1-3 including the first part of the raw acoustic recordings of dolphin vocalizations collected on November 20, 2021, split into five-minute audio segments (WAV files). One .zip file named “Raw_recordings_Day1_2” which contains 3 folders named “Raw_recordings_Day1_pti” i :4-6 including the second part of the raw acoustic recordings of dolphin vocalizations collected on November 20, 2021, split into five-minute audio segments (WAV files). One .zip file named “Raw_recordings_Day1_3” which contains 3 folders named “Raw_recordings_Day1_pti” i :7-9 including the third part of the raw acoustic recordings of dolphin vocalizations collected on November 20, 2021, split into five-minute audio segments (WAV files). One .zip file named “Raw_recordings_Day1_4” which contains 2 folders named “Raw_recordings_Day1_pti” i :10-11 including the fourth part of the raw acoustic recordings of dolphin vocalizations collected on November 20, 2021, split into five-minute audio segments (WAV files). One .zip file named “Raw_recordings_Day2_1” which contains 3 folders named “Raw_recordings_Day2_pti” i :1-3 including the first part of the raw acoustic recordings of dolphin vocalizations collected on November 21, 2021, split into five-minute audio segments (WAV files). One .zip file named “Raw_recordings_Day2_2” which contains 3 folders named “Raw_recordings_Day2_pti” i: 4-6 including the second part of the raw acoustic recordings of dolphin vocalizations collected on November 21, 2021, split into five-minute audio segments (WAV files). One .zip file named “Raw_recordings_Day2_3” which contains 2 folders named “Raw_recordings_Day2_pti” i :7-8 including the third part of the raw acoustic recordings of dolphin vocalizations collected on November 21, 2021, split into five-minute audio segments (WAV files). A single .zip file named “Whistle_Spectrograms” including spectrogram images of the dolphin whistles identified in the recordings (PNG files). A single .zip file named “Whistle_Signals” including the acoustic signals of the dolphin whistles identified in the recordings (wave files). A single Excel file named “Vocalization_Characteristics” which contains the quantitative vocalization characteristics, including the total number of whistles and pulsed vocalizations associated with the specific dolphin activity. A single .zip file named “Labels” including timing and labeling data for each detected vocalization.

The recording session ran continuously, starting on 11/20/2021 at 10:15 a.m. and ending on 11/21/2021 at 10:30 a.m.. The following Table 1 reports the scheduling of the dolphin training during the recording.
Day Start End Activity 20/11 10:20 11:00 ORD   12:00 12:45 ORD   14:45 15:15 ORD   15:20 16:00 PLAY   16:05 16:45 FFR   16:50 17:25 ORD 21/11 9:30 10:00 ORD
Table 1: Training activities during the recording: ordinary activity (ORD), play activity (PLAY), fish from the roof (FFR).

Ordinary sessions (ORD) were characterized by standard training exercises where dolphins receive rewards upon accomplishing tasks. These sessions lasted between 30 and 40 minutes. During the single play session (PLAY), the trainers stimulated interactive behaviors among the dolphins by introducing floating toys. This session lasted approximately 40 minutes. A further session, lasting around 40 minutes and named "fish from the roof" (FFR), was introduced for experimental purposes. In this session, additional fish were thrown into the pool near the hydrophone location to encourage specific behaviors within the context of the study. During the remaining time, the dolphins were free to move around and engage with one another, without receiving specific instructions or being involved in trainer-led activities.
The details of the data included in each file are reported below:
Raw recordings: The raw recordings are provided as 19 files containing a total of 286 audio segments. The 11 files Raw_recordings_Day1_i, i:1-11, include the recordings made on November 20, 2021, and the 8 files Raw_recordings_Day2_i, i:1-8, include those made on November 21, 2021. Each segment has a duration of five minutes and is stored in standard uncompressed WAV format. The file name is in the format: YYYYMMDD_hhmmss_192.wav. Over 30 GB of continuous raw signals are released in 19 separate files to simplify downloading for users.
Whistle spectrograms: The spectrograms of each whistle were generated within the 0-25 kHz range and grouped into subfolders that share the same name as the 5-minute block from which they were extracted. The file name is the same as the original recording file in its first part (YYYYMMDD_hhmmss_192), followed by "-colspectro-W-OFFSET.png", where W stands for whistle and OFFSET represents the time, in seconds, elapsed from the start of the individual recording at which the segment was extracted.
Whistle Signals: The WAV files of each spectrogram are provided in subfolders named after the original file from which they were extracted. The extracted signal was normalized within the range 0-1. The file name is the same as the original recording file in its first part (YYYYMMDD_hhmmss_192), followed by "-colspectro-W-OFFSET.wav".
Vocalization characteristics: This file includes structured information extracted from the raw dataset. Each row of the spreadsheet corresponds to a 5-minute audio segment, identified by its start timestamp in the format YYYYMMDD_hhmmss. For each segment, the file reports the total number of whistles and pulsed vocalizations, classified as echolocation click trains (ECT), burst pulse sounds (BPS), and feeding buzzes (FB). Whistles are classified based on their duration (d) into the following classes: 
- Class 1: d ≤ 0.2 s
- Class 2: 0.2 < d < 0.4 s 
- Class 3: 0.4 < d < 0.8 s
- Class 4: d ≥ 0.8 s
The number of whistles grouped by duration class and the type of dolphin activity associated with each time block were also reported. 
Labels: The timing and labeling data for each detected vocalization (whistle, ECT, BPS, and FB) are provided as tab-separated .txt files, including:
Vocalization start time expressed in seconds as the time stamp in the original 5-minute recording, counting from the start of the recording; Vocalization end time expressed in seconds as the time stamp in the original 5-minute recording, counting from the start of the recording; classification tag: W for single whistle, MW for multiple whistles, ECT for echolocation click trains, BPS for burst pulse sounds, FB for feeding buzzes, and NOISE for noise segment where no vocalization was detected.
Each row of the spreadsheet corresponds to a 5-minute audio segment, identified by its start timestamp in the format YYYYMMDD_hhmmss. All the parameters reported in these files are also depicted as labels that can be imported directly into the Audacity software for visualization (file à import à labels).