In many animal species, including humans, producer arousal state is considered a key modifier of vocal production and structure. Encoding of affective arousal state in vocalisations provides a rapid means of information transfer about an individual’s internal state, potentially reflecting its response to external stimuli. Meerkats, Suricata suricatta, are a highly vocal species. They use close calls to maintain group cohesion while foraging. Due to their patchily distributed prey, motivation for neighbour proximity varies; being too close results in competition (increased arousal - aggression), while too far results in risks of losing the group and predation threats (increased arousal - fear). We investigated how neighbour proximity, and also behavioural, social and environmental context, influence the acoustic structure of wild wild meerkats’ close calls. We found little effect of neighbour distance on the majority of the acoustic parameters measured. However, there was a consistent effect of the behavioural context in which the call was given across several acoustic parameters. Additionally, if the nearest neighbour was a pup calls became longer, lower and quieter. Overall, meerkat close calls potentially convey information on current behaviour and are modified in relation to social context. This highlights a potential mechanism in the diversification of acoustic signals. 

We recorded acoustic and behavioural data of subordinate adult (>1 year) meerkats during foraging, noting every time the focal produced a close call, the context in which the call was given, the distance to the nearest neighbour and the neighbour identity. Individuals were recorded with a Sennheiser directional microphone (ME66/K6) connected to a Marantz PMD-670 solid-state recorder (Marantz Japan Inc.; sampling frequency 48 kHz, 16 bits accuracy). A windshield (Rainhardt, W200) was attached to the microphone to ensure good-quality recordings under variable wind conditions. 

Vocalisations were imported at a sampling rate of 48 kHz and saved in WAV format at 16-bit amplitude resolution. Each recording was manually annotated in Adobe Audition for close calls and corresponding contextual information. Close calls were verbally indicated by the recording observer and confirmed visually by the spectrograms during annotation. We extracted each annotated close call from the recordings, and Hann band-pass filtered between 0.05 and 12 kHz for analysis. We only included calls confirmed to be from the focal individual, as indicated verbally during data collection by the recording observer, and only calls with no overlap with other meerkat or bird calls. We extracted and analysed 2,399 calls produced by 24 individuals (65-118 per individual) over 1,893 minutes of recordings. We then extracted acoustic parameters in PRAAT (Boersma & Weenink 2023) using a custom script for spectral and temporal analysis (adapted from Reby & McComb 2003; Charlton et al. 2009; Briefer et al. 2019; Wyman et al. in prep.

We conducted a statistical analysis in RStudio (version 2022.07.1). We began by running a Principal Component Analysis to eliminate redundancy within our set of variables and determine which parameters accounted the most for the differences observed between calls. We used the ‘broken stick’ method to select which Principal Components (PC) to keep as response variables in linear mixed models (LMMs; lmer function, lme4 library, (Kuznetsova et al. 2016)). The broken stick method selects PCs with observed variance (eigenvalue) greater than that of the total variance (sum of eigenvalues) when divided randomly amongst components, following a ‘broken stick distribution’ (King & Jackson 1999; Peres-Neto et al. 2005). For those PCs with eigenvalues greater than that of the broken stick distribution, we built models including the PC scores as a response variable and the following fixed effects; nearest neighbour distance (0-0.5, 1-2, 2-5, 6-9, 10+), behavioural context (scabble, forage, dig, eat, move, post-vigilance), nearest neighbour age category (pup, juvenile sub-adult, adult), nearest neighbour dominance status (subordinate or dominant) and sex (male or female), group size, and vegetation density (none, low, medium, dense). All fixed effects were categorical, with the exception of group size which was continuous. Interactions were not included as they were not supported by the model. Models contained individual ID nested within group ID as random effects, as previous work has confirmed individual- and group-signatures in meerkat close calls (Townsend et al. 2010, 2011a). We checked model assumptions for normality and heteroscedasticity, by testing model residuals distribution (KS test), dispersion and outliers (testResiduals function, DHARMa library (Hartig 2022)). All models’ residuals fit a normal distribution and thus no transformation was necessary. An information theoretic (IT) approach was applied for model selection, using Akaike’s information criterion (AICc) to rank the models (model.sel function, MUMin library (Bartoń 2022)) following the approach used by Richards et al. (2011). Models within AICc ≤ 4 of the model with the lowest AICc value formed the ‘top set’. Post-hoc analyses were performed using estimated marginal means on the top models, to investigate the significance of the fixed effects factor levels (emmeans function, emmeans library (Lenth 2023)) and pairwise comparison (contrasts function, emmeans library (Lenth 2023)).