Dataset of Bottlenose and Common Dolphins Acoustic Recordings in the English Channel

We provide dolphin whistles from two species: bottlenose dolphins (Tursiops truncatus) and short-beaked common dolphins (Delphinus delphis). Data are available through CSV files containing the time–frequency characteristics (contouring) of detected whistles across different acoustic encounters, with 74 whistles of bottlenose dolphins and 199 of short-beaked common dolphins. Due to confidentiality constraints, only a subset of the acoustic recordings collected in the English Channel could be shared (wave files). The dataset is described and analysed in a study entitled “Discrimination of two sympatric dolphin species from their whistles using machine learning models” (manuscript submitted). Methods Data collection Short-beaked common and bottlenose dolphin whistles were recorded North of the Dieppe coast (France), in the English Channel between July 2024 and January 2025. Audio files were collected continuously at a sampling rate of 156 kHz using RUBHY acoustic buoys (produced by RTsys, France) with preamplified hydrophone GP1516 (produced by Co.l.Mar, Italy; sensitivity: -168 dBVre 1 V/μPa). RUBHY buoys were deployed at 750m from a jack-up vessel to monitor acoustic activity under the construction of Dieppe-Le Tréport offshore windfarm foundations. At the same time, sightings of marine mammals were collected from the jack-up vessel by eight qualified Marine Mammal Observers (MMO). For each marine mammal encounter observed within a radius ~2km, the species was identified based on morphological criteria. Since the acoustic recorder was deployed within the observation radius of the MMOs and considering the acoustic detection range of the delphinid species we considered that the whistles recorded at the time of visual detection are associated from the individuals observed. Whistle contour extraction Whistles contours were extracted with a custom-written interface called ‘Caracall’ on MATLAB (developed by F. Cassiano). Frequency data were then interpolated every 0.01 seconds to eliminate inconsistent values resulting from manual annotation. Each CSV file corresponds to a single whistle and contains its time–frequency characteristics.