Vocal communication is an essential aspect of primate social behaviour. The bearded capuchin Sapajus libidinosus is endemic to Brazil and some studies have described specific vocalisation types for this species; however, there is still no complete description of its vocal repertoire. Thus, this study aimed to describe the vocal repertoire of a group of S. libidinosus living in the Parque Nacional de Brasília (PNB), a protected area in the Cerrado area of Central Brazil. We carried out focal samplings and recording of vocalisations of members of a S. libidinosus troop in different behavioural contexts. The call analyses revealed 25 different types of vocalisations, and each call presented significant structural variation. We grouped these vocalisations according to the context of the emission or acoustic structure into the following categories: contact calls (contact note, infant babbling, trill, teeth- and lip-smacking, and sirena); foraging calls (chihui, grgr, and patinado); whistle series (WS; food-associated WS, long-distance WS, and inter-group encounter WS); aggressive calls (aggressive contact note, ascending rapid staccato, cough cough, and pip); calls in response to aggression (scream, squeal, and pulsed scream), sexual display calls (chuck and raspy oestrous call), and stress-related calls (alarm call/ bark, hiccup, hip, double hip, and wah wah). S. libidinosus presented a very rich vocal repertoire, revealing a pattern consistent with the repertoire of other capuchin monkey species. This is the first comprehensive description of the S. libidinosus vocal repertoire and highlights the complexity of neotropical primate communication.

Vocal communication is an essential aspect of primate social behaviour. The bearded capuchin Sapajus libidinosus is endemic to Brazil and some studies have described specific vocalisation types for this species; however, there is still no complete description of its vocal repertoire. Thus, this study aimed to describe the vocal repertoire of a group of S. libidinosus living in the Parque Nacional de Brasília (PNB), a protected area in the Cerrado area of Central Brazil. We carried out focal samplings and recording of vocalisations of members of a S. libidinosus troop in different behavioural contexts. The call analyses revealed 25 different types of vocalisations, and each call presented significant structural variation. We grouped these vocalisations according to the context of the emission or acoustic structure into the following categories: contact calls (contact note, infant babbling, trill, teeth- and lip-smacking, and sirena); foraging calls (chihui, grgr, and patinado); whistle series (WS; food-associated WS, long-distance WS, and inter-group encounter WS); aggressive calls (aggressive contact note, ascending rapid staccato, cough cough, and pip); calls in response to aggression (scream, squeal, and pulsed scream), sexual display calls (chuck and raspy oestrous call), and stress-related calls (alarm call/ bark, hiccup, hip, double hip, and wah wah). S. libidinosus presented a very rich vocal repertoire, revealing a pattern consistent with the repertoire of other capuchin monkey species. This is the first comprehensive description of the S. libidinosus vocal repertoire and highlights the complexity of neotropical primate communication.

Study site and subjects

The study was conducted between September 2018 and June 2019 in Parque Nacional de Brasília (PNB), a protected area located in Distrito Federal, Central Brazil (15°35'–15°45' S, 48°50'–48°53' W). The study area was a gallery forest fragment, a phytophysiognomy of the Cerrado. Within the fragment, an area containing two artificial swimming pools was open to the public, where the subject animals often interacted with humans. We observed a group of ten bearded capuchins, comprising five adults (three males and two females), one subadult male, one juvenile female, and three infants. 

Data collection / Recording vocalisations

We used focal animal sampling [Altmann, 1974] to collect data on the bearded capuchin group's vocal repertoire. We recorded vocalisations using a  shotgun microphone connected to a digital recorder. Each focal session lasted 10–15 min, in which we followed the focal subject, recorded all emitted vocalisations, and recorded the behavioural context. Each time the focal animal vocalised, we dictated its activities to a headset microphone connected to the same digital recorder. We didn’t record infant vocalisations during focal sessions. Occasionally, we used ad libitum sampling [Altmann, 1974], and we included the calls recorded by this method to describe the bearded capuchins' vocal repertoire. We registered the activities of the focal animal according to the following behavioural categories: foraging (searching, approximating, manipulating, or inspecting a food item), feeding (ingesting a food item), locomoting (walking or running), resting (sitting or lying), and interacting (playing, attacking, grooming, or copulating). We registered other behaviour categories, such as sharing food items, specific body postures or displays, and interactions with humans. 

Data analysis

We saved each call recorded during sampling as an individual WAV file, from which we later produced spectrograms. We classified each call based on its acoustic structure and sound. We followed the classification used by Di Bitetti [2001] and Di Bitetti and Wheeler [2017] for S. nigritus. We named the vocalisations following Di Bitetti and Wheeler [2017]. After this first classification, we further selected high-quality files, removing those with excessive background noise, and proceeding with acoustic measurements. Only calls with more than ten high-quality samples files were used, resulting in the measurement of ten call types (n = 389 files). In general, all individuals contributed to the samples for each call type in the analysis. The exceptions were the sirena and raspy oestrous calls, which only had samples from males, and the teeth- and lip-smacking calls that only had samples from the juvenile female and one adult male. We used Fourier-transformed spectrograms, using the Avisoft-SASLab Pro v5.2.14 software (Avisoft Bioacoustics, Berlin, Germany), with the following settings: FFT-length = 1024, Frame = 100%, Hamming window, and 50% overlap. Whenever necessary, we used the finite impulse response (FIR) filter, high- and low-pass filters to remove noise that did not overlap with the vocalisations.

As the call types were very different in structure, we used simple acoustic variables that could be measured in all call types, based on the work of Di Bitetti and Wheeler [2017]. We measured the following variables from the selected spectrograms: number of elements in a call, call duration, duration of the longest element, initial peak frequency, peak frequency at the end, maximum peak frequency, minimum peak frequency, range of spectral energy, peak frequency on the mean spectrum, and change in frequency from start to end. We took measurements on the whole call when it had more than one element. We measured all variables at the fundamental harmonic of the call using automatic measurements in Avisoft-SASLab software. We checked the resulting values manually using a cursor at the spectrograms and then transferred automatically to an Excel spreadsheet.

To test the classification of the calls based on the measured acoustic variables, we performed a discriminant function analysis (DFA) with a cross-validation method, using the LDA function of the MASS package [Venables and Ripley, 2002] in R software [R Core Team, 2020]. Since the acoustic variables had different scales, we standardised the acoustic variables using the z-score method before performing the DFA. All variables were centred at zero, with a variance of one. One of the premises of LDA is that the number of acoustic variables in the model cannot exceed the number of calls assigned to each call type [James et al., 2013]. Therefore, we included only call types with more than ten samples in the analyses since we measured ten acoustic variables. First, we randomly divided the samples into two subsamples of the same size. One subsample served as a training sample to construct the discriminant function, while the other was the test sample used to verify how well the discriminant function assigned vocalisations to their a priori categories. After the cross-validation process, we performed a permutation test to compare the number of correct classifications with that expected in 1000 random permutations [Slow, 1990].

The data set is composed by three files, "Acoustic_Measurements.xls" contains the measured acoustic variables that were used to perform our analysis (columns 6 – 15), columns 1 – 3 contains information about the sex, age and identity of the individual that emitted the call, column 4 contains the call type and column 5 contains the file name that were used to take the measurements. The file entitled "measured_calls.rar" contains the 389 call files that were used to take the acoustic measurements.  The third file, "25_calls.rar", contains one example of each of the 25 Sapajus libidinosus' call types that were described in our work.