Effects of noise on acoustic and visual signalling in the Croaking Gourami: differences in adaptation strategies in fish

ABSTRACT Numerous fishes produce sounds and their transmission and detection may be hindered by increasing levels of anthropogenic noise. We investigated acoustic communication during dyadic contests between male croaking gouramis (Trichopsis vittata, Osphronemidae) in the presence and absence of white noise. We hypothesised that fish modify acoustic signalling in the presence of noise in order to maintain intraspecific communication. Under controlled laboratory settings we compared agonistic behaviour, visual and acoustic signalling between noise and no-noise conditions. Trichopsis vittata produced sounds that were significantly lower in level and higher in dominant frequency under noise treatments. No difference was found in visual signalling or temporal sound characteristics. This study indicates that noise does not affect the amount of signalling during agonistic behaviour in a highly vocal fish. No increase in sound level was observed in croaking gouramis, indicating that a Lombard effect is not present in all vocalising fish. The lack of a Lombard effect shows that sound communication is potentially hindered by (human-made) noise in fish, which may affect territory maintenance and reproduction.


Introduction
Different biotic and abiotic factors can influence the behaviour and exchange of information in animals. In almost all terrestrial and aquatic environments, acoustic communication is severely impaired by the constant masking of high-intensity background noise from naturally occurring and man-made sources Ladich 2019) All animals rely on signalling, and anthropogenic noise clearly affects the behaviour and therefore the signalling in many taxa. This can have direct or indirect consequences for their ecology and wellbeing, as has been shown in terrestrial species such as birds (Brumm 2004, Brumm andZollinger, 2013;Dorado-Correa et al. 2018Zhou et al. 2019, anurans (Sun and Narins 2005;Zhao et al. 2018), as well as mammals (Eliades and Wang 2012;Stowe and Golob 2013;Schopf et al. 2016;Jiang et al. 2019), who clearly changed their signalling behaviour when confronted with anthropogenic noise.
Acoustic communication is an essential aspect of social interaction in animals, and sound is a very useful tool, especially in aquatic habitats where the visual range is reduced due to turbidity or poor light conditions (Holt and Johnston 2014). In the past decades These increasing noise levels in aquatic environments potentially affect acoustic communication in fish, prompting an increasing number of studies on the effect of noise on fish communication. Nevertheless, such effects were assessed under different paradigms and treatments, complicating comparisons between fish species and noise regimes (Alves et al. 2021). The present study therefore seeks a better understanding of how noise influences acoustic communication and social interactions in a highly vocal fish species. The croaking gourami, Trichopsis vittata, was chosen to study changes in visual and acoustic signalling during agonistic interactions in the presence of noise. Croaking gouramis defend their territories to reproduce successfully. Sound production during dyadic contests is well known in T. vittata due to several prior studies (Ladich 1998, 2007, Ladich and Maiditsch 2018, Maiditsch and Ladich 2022. Both sexes vocalise during agonistic interactions to maintain territories but only females produce sounds during courtship. The dominant frequencies of these croaking sounds are concentrated between 1.0 and 1.5 kHz, and thus within the best hearing range of this species. Croaking gouramis possesses an air-filled suprabranchial chamber for air-breathing laterally to the inner ears, and can detect sounds up to several kHz (Schneider 1941;Ladich and Yan 1998;Wysocki and Ladich 2001).
The study was designed to determine how noise influences signalling during agonistic behaviour. We hypothesise that (1) noise does not result in a reduction of signalling because croaking gouramis have to defend their territories and (2) that gouramis adapt characteristics of their croaking sounds to noise. In particular, sound levels should increase in the presence of noise and thus provide evidence for a Lombard effect in this species. Finally (3) if a Lombard effect is not present in T. vittata, we hypothesise that auditory masking is compensated by visual signalling. We chose white noise at 110 dB because this level is within the low natural ambient noise range. Males were selected because of availability, and they do not differ from females in signalling during agonistic behaviour (Ladich 2007;Ladich and Maiditsch 2018;Maiditsch & Ladich, 2021).

Animals
Test subjects were 42 male croaking gouramis (body weight: 0.82-1.32 g, standard length: 33.4-41.6 mm) obtained from a local aquarium store in Vienna. All fishes were kept in community tanks (100 x 50 × 40 cm) at 25 ± 1°C in mixed gender-groups, and a 12 h:12 h light : dark light cycle was maintained. Sexing of fish was based on the absence of the whitish ovary in males (see Supplement in Maiditsch and Ladich 2021). Holding tanks were planted; the bottoms were covered with sand and contain tubes and halved flowerpots as hiding places. Water was filtered by external filters. The fish were primarily fed frozen chironomid larvae or commercially prepared flake food (Tetramin) five times a week (Ladich and Maiditsch 2018;Maiditsch and Ladich 2021).

Experimental setup
The test tank (50 × 27 × 30 cm) was set up three weeks before start of the experiment. A UW30 (University Sound, Buchanan, Michigan U.S.A.) underwater speaker was placed on the left side of the aquarium, suspended behind a plastic mesh so that the fish could not reach the speaker (Figure 1). The test tank bottom was covered with sand, and a plant was placed in each half as a hiding place. The set-up within the test tank remained the same until all experiments were completed.
Prior to experiments, individual males were gently put on a paper tissue before being weighed on a scale (Sartorius GmbH Göttingen PT 120; accuracy: 0.01 g) and standard length measured with a sliding calliper (Workzone,Nr. 23,149,168). Only gouramis that differed by less than 10% in weight were paired to avoid asymmetries that may not result in dyadic contests and acoustic signalling (Table 1). Ten white noise (20 males) and eleven no noise contests (22 males) were staged; males were only used once and randomly assigned to either noise or no-noise treatments. Opponents were kept separately for five days in isolation tanks (50 × 27 × 30 cm), under conditions similar to the holding tanks, in order to reduce dominance pre-experience. On the fifth day (24 h before the start of the trial), fish were transferred randomly into the left or right half of the test tank in order to acclimate to the new tank. They were separated by a nontransparent plastic plate and could not see each other. On the experimental day a hydrophone (Brüel & Kjaer 8101; sensitivity: −186 dB re 1 V/μPa) was placed close to the plastic plate at the back wall of the tank (sectors ABC 5; Figure 1). Gouramis typically interacted agonistically at the borders of their territories.
Dyadic contests were performed under two different conditions -white noise and no noise (control or ambient noise) condition. The white noise test started after fish were acclimated for 30 min, during which fish were allowed to get used to the noise. After this period the plastic plate was lifted manually and the opponents could see each other. White noise was played back during the whole time of the contest and was stopped after the fight was decided. For the control condition no acclimation time was necessary before the plate was removed because no noise was played back.
The entire setup was enclosed in a walk-in semi-anechoic room, which was constructed as a Faraday cage. The test tank was placed on a table that rested on a vibrationisolated concrete plate. All experiments were conducted at the same time of the day (forenoon). After experiments, fish were returned to the community tanks (Ladich and Maiditsch 2018;Maiditsch and Ladich 2021).

Behaviour and sound recordings
Agonistic behaviour started when opponents detected each other visually. Agonistic interactions consisted of two phases, the lateral display phase (LD) followed by the frontal display phase (FD). The LD-phase was organised in bouts (or sequences), after which fish swam to the surface for air-breathing (see Figure2 in Maiditsch and Ladich 2021). During the LD-bouts, opponents erected their unpaired fins, circled head to tail at a distance of 1-3 cm and produced croaking sounds only during this behaviour (see Figure1 in Ladich 2007). Sounds were typically emitted alternately by opponents during LD-sequences. The LD-phase is followed by frontal displays during which fish protrude their mouths, bite each other's mouth, but do not vocalise (escalated phase) (Ladich 1998). Contests were typically decided during the LD-phase when one fish gave up and fled (one winner; non-escalated contests). If fights proceeded to the FD-phase (escalation contest), they were stopped by the experimenter (no winner-undecided) to prevent fish from biting each other (Ladich 1998;Maiditsch and Ladich 2021).
Croaking sounds consist of a series of bursts produced by one pectoral fin each, when enlarged fin tendons snap over bony elevations of fin rays (Kratochvil 1978). The soundproducing fish could be determined easily on the video by the rapid pectoral fin beating during which the whole animal was shaking. Acoustic signals and behaviour were recorded using the hydrophone connected to a microphone power supply (Brüel and Kjaer 2804), which was connected to the XLR mic input of a 4-K video camera (Panasonic HC-X1000). Recordings were controlled via the camera display and a video monitor (Sony PVM 4000). The entire setup was positioned behind a curtain so that animals could not see the experimenter (Ladich and Maiditsch 2018;Maiditsch and Ladich 2021).

White noise playback
Continuous white noise was played back at a sound levels of 110 dB (RMS) re 1 µPa. (Figure 2). This noise level was chosen because croaking sounds were still detectable and it is similar to low-level noise encountered in fish habitats. White noise was chosen because it has a flat power spectrum over the entire bandwidth (Scholik and Yan 2001). The noise was generated in Cool Edit 2000 (Syntrillium Software Corporation, Phoenix, AZ, USA) on a laptop (Getac B300) and sent via a soundcard (Edirol UA-25) to a 30band equaliser (Alesis MEQ 230) to obtain a flat noise spectrum underwater; it was then fed via a power amplifier (Brüel & Kjaer Type 2713) to the underwater speaker.
Ambient (No) noise levels were recorded for one minute after every experiment, after fish were taken out of the test tank. Sound pressure levels (RMS) were measured using a sound level metre (Brüel & Kjaer Mediator 2238) connected to the power supply. The ambient noise level ranged from 89 to 93 dB re 1 µPa in no-noises experiments.

Measurements of sound characteristics
Sound pressure levels (LAFmax, broadband A frequency weighting, RMS Fast time weighting) were recorded using a sound-level metre (Brüel and Kjaer 2250) connected to the microphone power supply. The equipment was calibrated with the hydrophone calibrator (Brüel and Kjaer 4229). All dB values were referenced to 1 μPa. Relative spectra of croaking sounds recorded under noise as well as no-noise conditions were created using S_Tools-STX 3.7.8 (Acoustics Research Institute, Austrian Academy of Sciences, Vienna, Austria). Relative FFT amplitude spectra were calculated (sampling frequency 48 kHz, filter bandwidth 45 Hz, 75% overlap, number of coefficients: 50, Hanning window). The spectra were then exported as ASCII Files and imported into Excel, The relative broadband RMS (Root Mean Square) determined in STX was then equated to the absolute LLAF measured with the sound level metre in parallel to the video and sound recordings, and the relative spectral levels were recalculated into absolute spectral levels for each croaking sound, following the algorithm described in Amoser et al. (2004) and Wysocki and Ladich (2005). The first 10 croaking sounds produced near the hydrophone (sectors A 4-6, B 4-6, C 4-6, D 4-6) were used to determine absolute spectral levels. Absolute spectral levels at the dominant frequency of sounds were measured instead of broadband RMS levels to exclude potential effects of white noise on SPLs of croaking sounds following the methods described by Holt and Johnston (2014). Control tests were carried out prior to fish experiments to clarify whether white noise affects spectral sound level measurements. A croaking sound was played back via a small loudspeaker (Fuji 7G06) at no noise and white noise conditions. The spectral levels at the dominant frequency of the sound did not differ between these two noise conditions.
Because the distances of the fish to the hydrophone changed, the test tank was divided into 40 sectors (5x5 cm) by a grid plotted to the front glass of the aquarium (Figure 1). The sector in which fish emitted sounds was determined. To compensate for different distances between the hydrophone and the croaking fish, a correction factor was calculated (Ladich et al. 1992;Ladich 2007). For this correction factor, a typical croak was played back three times at a constant level from a small loudspeaker (Fuji 7G06) in each of the 40 sectors and the mean SPL registered. The small loudspeaker was positioned in the centre of each sector because the exact position of each fish within a sector during circling could not be determined unequivocally. The SPL differences between the sector nearest to the hydrophone (5 cm away) and all other sectors were calculated and added to the SPL values measured while the fish produced sounds in a particular sector. This yielded a distance-independent absolute SPL for each sound emission.
Furthermore, the dominant frequency of calls was determined using the frequency at the highest spectral level (Figure 3) in the absolute sound power spectra (filter bandwidth 45 Hz, overlap 75%, Hanning window, max. frequency 3.5 kHz) (Noll 1967;Ladich 2007). Frequencies were not analysed above 3.5 kHz to avoid the resonance frequencies of the small tank, which are above 3.3 kHz according to the formula by Akamatsu et al. (2002).

Behaviour and sound analysis
The behaviour was analysed manually in Sony Vegas Pro 13.0. The video camera recorded LPCM-coded sounds, which were afterwards rendered in Sony Vegas Pro 13.0 to wavformat (44.1 kHz, 16 bit). These sounds were analysed using CoolEdit and S_Tools-STX 3.7.8 (Acoustics Research Institute, Austrian Academy of Sciences, Vienna, Austria).
The following visual and acoustic variables were determined: for visual signalling (1) the delay until the beginning of a contest (time from removing the separating plate until begin of first LD), (2) number and duration of LD-sequences in a contest, (3) duration of all LD-sequences minus pauses, and (4) the duration of pauses between the LD sequences; for the calling activity (5) Number of sounds (total number of croaking sounds produced during a contest, during each LD-sequence and numbers produced per individual), (6) number of bursts in croaking sound and for the call characteristics (7) dominant frequency and (8) the spectral sound level at the dominant frequency of sounds (the first 10 croaking sounds, if present, were analysed per individual and trial).

Statistical analysis
All calculations were done using SPSS 26 (IBM SPSS Statistics) or SigmaPlot 12.0 (SPSS Inc., Chicago, IL, U.S.A.). Data were tested for normal distribution using the Shapiro-Wilk test. Data that failed the test for normality were performed with non-parametric Mann-Whitney-U tests. For those parameters that tested normal, a t-test was performed.
The entire LD-contest was analysed, regardless of contest length. For every male, all croaking sounds produced during the contest were counted and visual as well as acoustic variables were analysed. Means of behavioural variables and sound characteristics (Number of LD-sequences, duration of all LD-sequences, duration of the entire LDphase, pauses between LD-sequences, delay until begin of the contest, number of croaking sounds produced during LD-sequences and the entire contest, number of bursts, SPL and dominant frequency) were calculated for contests and each individual and used for further analysis. Agonistic sounds of 19 males in white noise trials and of 22 males in nonoise trials were analysed. One individual did not produce sounds during dyadic contest under the noise condition. Asymmetries in body weight and standard length were calculated between opponents for each trial and compared between no noise and white noise conditions.

Agonistic behaviour and amount of signalling
The amount of visual signalling was similar in both trials. Contest started in average after 117.5 seconds in the white noise contest after the plastic wall was removed and 86.7 seconds in the no noise trial, with no significant difference between treatments (Mann-Whitney U-Test: U = 76.0; N = 21; p = 0.152; Table 1). The mean duration of LD-bouts was similar in white noise and the control experiments (Mann-Whitney U-Test: U = 212.0; N = 21; p = 0.840), as was the number of LD-bouts per contest (Table 1). During the white noise experiments, 10.4 LD bouts were observed, and 13.6 during no-noise experiments (Mann-Whitney U-Test: U = 198.0; N = 21; p = 0.578). No differences were found in the total duration of all LD-sequences (minus pauses); competition between fish lasted on average 246 s during white noise and 260 s when no noise was played (Mann-Whitney U-Test: U = 220.0; N = 21; p = 1.000). The duration of pauses between LDs was also similar between treatments (Mann-Whitney U-Test: U = 220.0; N = 21; p = 1.000; Table 1).
The total number of sounds produced by individual fish during contests did not differ between no noise and white noise experiments (Mann-Whitney U-Test: U = 219.5; N = 42; p = 0.990). Individual croaking gouramis emitted up to 156 sounds with a mean of 44.7 in a noise environment and ten up to 173 sounds with an average of 46.8 during lab conditions ( Table 2). The total number of croaking sounds produced during a dyadic contest was similar in the noise (mean: 89.4) and no-noise trials (mean: 93.6; Table 1) (Mann-Whitney U-Test: U = 216.0; N = 42; p = 0.920), as was the number of croaking sounds produced per LD bout (Mann-Whitney U-Test: U = 220.0; N = 21; p = 1.000). Approximately 16 croaking sounds were emitted per LD bout during experiments (Table 1). Moreover, no difference was found in the number of bursts per croaking sound produced with a median of about five bursts in different trials (t-test: t = > −0.816; df = 40; p = 0.419).

Sound characteristics
Trichopsis vittata modified the call characteristics under noisy conditions. SPLs were significantly higher in no-noise contests with an average of 120 dB than in white noise with an average of 116 dB (Mann-Whitney U-Test: U = 221.0; N = 35; p = 0.009; Figure 4a). The dominant frequencies differed significantly between experiments (t-Test: t = 2.162; df = 38; p = 0.037; Figure 4b); T. vittata produced higher-frequency sounds with a mean of 1413.7 Hz (median: 1369.8) under white noise conditions (Table 2).

Discussion
Our results show that noise did not affect the amount of visual and acoustic signalling during agonistic behaviour in contests in T. vittata. The number and duration of LDbouts plus pauses in-between and the number of sounds (and bursts within sounds) did not differ between noise and no-noise trials. Thus our first hypothesis was supported, namely that signalling did not differ between noise conditions. Significant differences were found in two sound properties between males, namely in sound level and main frequency. The sound level, however, did not increase and thus no Lombard effect is present in croaking gouramis. Accordingly, the data of the present study do not support our second hypothesis.

Agonistic behaviour and amount of signalling
The fact that noise did not influence the amount of signalling during territory defence indicates the importance of maintaining territories in croaking gouramis, independently of noise. This is in agreement with our first hypothesis.
Does T. vittata switch to the visual communication channel when the acoustic channel is hindered by noise and fish lack the ability to overcome this constraint by increasing noise levels? Animals may switch to other channels of communication when the acoustic channel is masked by noise. In the painted goby, visual courtship becomes more important during mating when noise disturbs acoustic communication (de Jong et al. 2018). Contrary to expectations, this was not the case in T. vittata, but such changes due  to noise were observed in several other species of fishes. Continuous tonal noise changed visual displays during male-male territorial interactions in A. burtoni such that they spent more time with their eyebar displayed, suggesting an increase in visual signalling (Butler and Maruska 2020). In the cichlid Neolamprologus pulcher, dominant breeding pairs showed significantly more aggressive behaviour (visual signals: fin spreading, head-down display, head jerking, opercula spreading and S-shaped bending) towards subordinates during playback of boat noise than during ambient noise (Bruintjes and Radford 2013). Interestingly, de Jong et al. (2018a) reported that noise affected male courtship behaviour differently in two closely related gobiids. Painted gobies were much more active during noise playbacks than two-spotted gobies, whose visual displays did not differ between noise treatments. This indicates that noise may have various effects on different fish species. The croaking gourami, for example, modifies its visual and acoustic signalling in the presence of a biotic threat, namely a predator. In those predator trials, a decrease in the number and duration of LD bouts was accompanied by a significant decrease in the number of croaking sounds emitted (Maiditsch and Ladich 2022). This underlines that fishes possess diverse behavioural strategies to adapt to different ecological constraints. The fact that visual signalling in the current study was not influenced by noise shows that the lack of a Lombard is not compensated by more visual signalling. This indicates that our third hypothesis is not supported by the data.
Trichopsis vittata showed no change in calling activity (number of sounds) due to white noise under lab conditions. Further studies under standardised conditions showed different findings. In the Lusitanian toadfish, boat noise lowers communication distance, which leads to a significant decrease in the males' calling rate during boat noise playback (Alves et al. 2021). Calling also decreased in the two-spotted goby and the painted goby. Both species produced fewer drumming sounds when a lowfrequency harmonic tone was added under standardised lab conditions. This may reduce the attractiveness of male painted gobies because females were less likely to spawn, which may negatively influence their fitness (de Jong et al. 2018a). In a field study, the oyster toadfish reduced calling rates in a boat traffic area, suggesting that toadfish cannot call over loud vessels, which lowers their communication distance (Luczkovich et al. 2016).
In contrast one field study showed that noise had no influence on the calling activity. Higgs and Humphrey (2020) reported no clear effect of an increase in overall background noise in the round goby (Neogobius melanostomus). This supports the importance of acoustic signalling for mate attraction and may also be the case in the croaking gourami. Calling rates are more important for territory defence and close male-male encounters. Accordingly, noise does not seem to disturb agonistic interactions as much as predators do.

Sound characteristics
What could influence the reproductive success is the effect on sound characteristics (SPL). Based on laboratory experiments, we know that pairing a male and female croaking gourami can end in either agonistic behaviour or mating behaviour, whereby mating behaviour itself can be subdivided into three stages, the first of which involves females producing purring sounds to initiate spawning behaviour (Ladich 2007). This is the case in a closed environment (tank), but what exactly attracts females in the wild remains unclear; sound level and frequency could play a key role in this scenario.
Trichopsis vittata were unable to increase sound levels in a noisy environment, indicating that a Lombard effect is lacking in this highly vocal species. Under lab conditions, Holt and Johnston (2014) reported that sounds produced by C. venusta during noisy conditions were significantly louder (spectral levels increased by 5.7 to 9.5 dB depending on the sound type). A Lombard effect was also mentioned in the oyster toadfish. Sound levels of vocalisations increased during the playbacks of vessel noise (Luczkovich et al. 2017). In the plainfin midshipman fish, males exposed to noise even lowered the fundamental frequency of their vocalisations by 4.6 Hz and increased the amplitude of their calls by approximately 7.9 dB (Brown et al. 2021). A field study done by Siddagangaiah et al. (2021) indicated a Lombard effect in species belonging to the family Sciaenidae, where the intensity of their chorus increased by 5 to 10 dB, during sustained turbine operation. The opposite effect in the present study is difficult to explain. It may be due to methodological difficulties because we could not measure absolute SPLs in each sector of the test tank but only close to the hydrophone (12 out of 40 sectors) where SPLs were above noise level. In an earlier study when 50 sectors were used for SPL correction, slight but significant differences in SPLs were found between sexes and behavioural contexts in croaking gouramis (Ladich 2007). Female agonistic sounds are lower in level than male agonistic sounds, most likely because females possess smaller pectoral sound-generating mechanisms (Kratochvil 1978, Ladich 2015. In addition, females produce two types of sounds, a high-level agonistic sound and a shorter lower level prespawning sound (Ladich 2007). This indicates some variability in sound output in T. vittata. The lowlevel courtship sounds may not be detectable during noise conditions. Croaking gouramis possesses an air-filled suprabranchial chamber for air-breathing laterally to the inner ears, which extends their hearing range up to several kHz (Schneider 1941;Ladich and Yan 1998). The dominant frequency of croaking sounds between 1.0 and 1.5 kHz is positioned within the most sensitive frequency range of this species (Ladich and Yan 1998). Noise will affect auditory thresholds (Ladich 2013) by increasing hearing thresholds, which reduces the ability of T. vittata to detect conspecific sounds and assess opponents and mates.
Are there any other differences in sound characteristics in the presence of noise in fishes? Neither in the blacktail shiner nor in oyster toadfish were any differences in main frequencies mentioned (Holt and Johnston 2014;Luczkovich et al. 2017). The dominant frequency was higher in noise treatments in T. vittata, which is difficult to explain when the interacting experimental fish are similar in size. The dominant (main) frequency typically depends on body size in those fish producing pulsed sounds (Myrberg et al. 1993). This has been shown in several studies in the genus Trichopsis (Ladich et al. 1992;Ladich and Maiditsch 2018;Ladich and Schleinzer 2020). Within birds, higher song frequencies in cities could be an adaptation to low-frequency anthropogenic noises, as suggested for great tits (Slabbekoorn and Peet 2003;Nemeth and Brumm 2009).

Conclusion and future research
This study indicates that noise does not affect the amount of acoustic and visual signalling during agonistic behaviour in a highly vocal fish under lab conditions. No Lombard effect was observed in T. vittata under these standardised noise conditions. We therefore assume that croaking gouramis are unable to adapt their sound characteristics to noise. We conclude that intraspecific acoustic communication is hindered by noise in male -male interactions in croaking gouramis. Hearing may be masked, as has been shown in several noise studies in fish (Ladich 2013). This may affect the assessment of opponents during agonistic interactions and of mates during courtship and spawning. The assumption is that female's low-level prespawning sounds are solely detectable under low noise conditions (Ladich 2007). Thus, even low noise levels may have a major impact on communication during reproduction. The amount of masking in the presence of various noise types and levels as compared to no noise conditions remains to be determined. A future study using noise from different habitats would help to clarify if and how noise affects nest defence and reproductive success under natural conditions.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Data availability statement
The data that support the findings of this study are available from the corresponding author, Isabelle Pia Maiditsch, upon reasonable request.

Ethical statement
Agonistic interactions between croaking gouramis consist of two phases: a lateral display phase followed by a frontal display phase. Croaking gouramis produce visual and acoustic signals only during the lateral display phase, without any physical contact between opponents. As the intention was to analyse signalling during contests, the agonistic interactions were not allowed to proceed to the frontal display phase during which fish bite each other (Ladich 1998;Maiditsch and Ladich 2021). All applicable national and institutional guidelines for the care and use of animals were followed (permit numbers: Animal Ethics and Experimental Board, Faculty of Life Science 2017-010; BMWFW-66.006/0035-WF/V/3b/2017 by the Austrian Federal Ministry of Science, Research and Economy).

Funding
This study was funded by the Austrian Science Fund (FWF grant no. P31045 to FL). The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript; Austrian Science Fund (FWF) [P31045].