Acoustic recordings and sound analysis
Acoustic experiments lasted from mid-April to October, which overlaps the natural reproduction season of O. croaticus . During acoustic recording sessions, a hydrophone (H2A-XLR hydrophone, Aquarian Audio & Scientific, Anacortes, WA, USA; sensitivity: -180 dB re. 1 V µPa-1; frequency range ± 4 dB from 0.01 to 100 kHz), was placed in the centre of the experimental tank above the shelter, and connected to a IRIG PRE preamplifier (Aquarian Audio & Scientific). Sounds were recorded using a ZOOM H4n portable digital audio recorder (16 bit/44.1 kHz sample rate; ZOOM, Tokyo, Japan). The hydrophone was placed within the attenuation distance from the emitter (less than five cm), and we obeyed the laboratory protocol for minimum resonant frequency for small glass tanks (e.g. , 2.7 kHz for 170 L tanks, according to Akamatksu et al., 2002). Sounds, monitored with headphones and noted by the observer, were stored on the recorder memory card as .wav files. Recordings were later band-pass filtered (0.05 – 3 kHz) to improve S/N ratio and subsampled at 4 kHz, and further amplified (10 dB) for better auditory and visual inspection of the audio tracks. Digitalised sounds were analysed using Avisoft - SASLab Pro 5.2 Software (1024-point FFT, FlatTop window; 100% frame; Avisoft Bioacoustics, Berlin, Germany). Ten audio recordings (2.5 per male, each lasting approx. 30 min) were aurally and visually inspected. Each O. croaticus sound was labelled using the “insert label” function of Avisoft - SASLab Pro. In this study, we recorded 367 pulsatile sounds from four soniferous males, but not all sounds presented a good signal to noise ratio (S/N) for acoustic analysis. From ten recordings presenting the best S/N ratio, we analysed 20 randomly selected sounds. Temporal features were measured from oscillograms, while frequency-related variables were obtained from the logarithmic power spectra (FlatTop window, 512-points FFT, 96.87% overlap; resolution 8 Hz). For sounds, we measured the following acoustic properties (Figure A1 ): (1) sound rate (SR, number of sounds emitted in 1 min); (2) sound duration (DUR, total length of the call, measured in milliseconds); (3) number of pulses (NP); (4) pulse repetition rate (PRR; NP divided by DUR and multiplied by 1000; Hz); (5) pulse duration (PD; ms); (6) pulse period (PP; average peak-to-peak interval of consecutive pulses, ms); (7) fatigue (FAT; ratio between the average PP of the last three and first three pulses representing the decrease in pulse emission rate possibly due to muscle fatigue, following Amorim et al., 2013); (8) frequency modulation (FM, after the sound has been divided into three temporally identical sections, FMi - initial, FMm - middle and FMf - final - see Figure A1 , frequency modulation was calculated as the difference between the final and initial pulse repetition rate and expressed in Hz; FMi, pulse repetition rate of the initial section of a sound and FMf, pulse repetition rate of the final section of a sound); (9) peak frequency (PF, the peak with the highest energy from the logarithmic power spectrum function, Hz). In order to follow the previous recording protocols as closely as possible (Amorim & Neves, 2007; Amorim et al., 2013), we also calculated the vocal activity parameters per male: (I) sound rate (number of sounds produced per min), (II) maximum sound rate (maximum number of sounds emitted in 1 min) and (III) calling effort (percentage of time spent calling, i.e., seconds of sound production divided by the duration of the recording in seconds). Despite the fact that the variables PP and PRR indicate the pulse repetition pattern, they were deliberately indicated separately here in order to facilitate comparisons with the goby literature on sound production.