Anatomical analysis
For anatomical dissections, additional O . croaticusindividuals were collected in October 2020 from the same watercourse near the village Brečići (43°7’11.30”N, 17°29’4.03”E) using electrofishing. Five individuals were collected, of which three males (40 – 50 mm LT) were immediately euthanised with an overdose of MS-222 (tricaine methane sulphonate; Pharmaq, Overhalla, Norway), and stored for one week in 7% formaldehyde fixative solution and then transferred to 70% ethanol. Specimens were dissected and examined with a Wild M10 binocular microscope (Leica Camera, Leica, Wetzlar, Germany) equipped with a camera lucida to study the anatomy of the putative sound producing mechanism. Since earlier research on gobies highlighted the role of the pectoral girdle and (pectoral) fins in sound production, dissections primarily addressed the muscles related to this body part. The nomenclature used to designate muscular parts was based on earlier research (Winterbottom, 1974; Adriaens et al., 1993; Parmentier et al., 2013, 2017). Additionally, one specimen was subjected to micro-computed tomography (μCT) scanning to visualise the fish skeleton at the level of the neurocranium and pectoral girdle. Scanning was completed using a RX EasyTom (RX Solutions, Chavanod, France; http://www.rxsolutions.fr), with an aluminium filter. Images were generated at 75 kV and 133 μA, with a frame rate of 12.5, 5 average frames per image. This procedure generated 2897 images at a voxel size of 10 μm. Reconstruction was performed using X-Act software from RX Solutions. Segmentation, visualisation, and analysis were performed using Dragonfly software (Object Research Systems (ORS) Inc, Montreal, Canada, 2019; software available at http://www.theobjects.com/dragonfly). Three-dimensional (3D) 16-bit images were produced and subsequently converted into 8-bit voxels using ImageJ (Abràmoff et al., 2014). Three-dimensional processing and rendering were obtained after semi-automatic segmentation of the body using a ‘generated surface’, according to the protocols described by Zanette et al. (2014). Direct volume renderings (iso-surface reconstructions) were used to visualise a subset of selected voxels of the anterior skeleton in AMIRA 2019.2.