Vocal Signatures in Rams: Exploring Individual Distinctiveness Across Different Contexts

Simple Summary

Vocal recognition helps animals stay connected and survive within their social groups. In sheep, it is especially important for mothers to recognize their lambs after birth. While vocal individuality has been extensively studied in lambs and ewes, little is known about rams, who are often reared separately in farming systems. This study explored whether adult rams exhibit vocal individuality across different emotional contexts and whether their calls convey affective information. We recorded vocalizations from rams in five distinct emotional situations: morning and evening isolation, auditory exposure to ewes’ bells, feed anticipation, and feed denial. Using proper statistical approaches, we assessed both individual vocal distinctiveness and contextual acoustic variation. Results showed that rams demonstrated vocal individuality with classification accuracy ranging from 59% to 80%. Although vocal distinctiveness was retained across contexts, it was less pronounced, indicating partial acoustic stability. Furthermore, calls were classified above chance level according to context, suggesting that rams use vocalizations to express emotional states. Among the acoustic parameters, formant dispersion and amplitude variation were most effective in distinguishing negative from positive contexts. These findings highlight context-sensitive vocal individuality in rams and its potential role in affective communication.

Abstract

Individual vocal recognition is essential for social cohesion and survival among conspecifics. In sheep, it facilitates postnatal identification and strengthens the mother–offspring bond. Although vocal individuality has been well-documented in lambs and ewes, little is known about whether rams—typically reared separately in farming systems—exhibit acoustic distinctiveness. This study investigated whether rams express vocal individuality across different emotional contexts and whether their calls convey contextual information. Adult rams’ vocalizations were recorded across five emotionally distinct contexts: physical and visual isolation (morning and evening), auditory exposure to ewes’ bells without visual contact, feed anticipation, and feed denial. Implementing discriminant function analyses and linear mixed models, we assessed individual distinctiveness and contextual variation in acoustic parameters. Rams exhibited vocal individuality ranging from 59% to 80%, with higher distinctiveness in negative contexts compared to positive ones. Vocal distinctiveness persisted across contexts, albeit to a lesser degree, suggesting some degree of acoustic stability. Calls were classified above chance by context, suggesting rams use vocalizations to convey affective states. Formant dispersion and amplitude variation were the most informative raw acoustic parameters under negative from positive contexts. The findings indicate that rams exhibit context-dependent vocal individuality, potentially conveying affective states through vocalizations.

1. Introduction

Animal vocalizations play a crucial role not only in survival but also in enabling communication and individual recognition within the same species. These acoustic signals can transmit multifaceted information about the sender, either static or more dynamic, such as age, body size, and sex, as well as emotional state and social status, respectively [1,2]. Unlike visual or olfactory-based signals, they are less susceptible to attenuation by distance or physical obstructions [3]. Depending on their frequency, sound waves are capable of traversing barriers more effectively than visual or chemical (olfactory) signals, thereby allowing auditory communication to occur over extended ranges. Consequently, individuals do not need to be in proximity to detect and interpret vocal information from conspecifics [4]. Furthermore, vocalizations can encode individual identity. Species that inhabit larger social groups tend to exhibit a greater degree of individual-specific acoustic variation than those residing in smaller social groups [2,4].

The source-filter theory has significantly advanced the analysis of vocal characteristics by offering a comprehensive framework for investigating vocal signal production in both human and non-human mammals [5,6,7,8]. The theory distinguishes two processes: sound generation via vocal fold vibrations (which constitutes the “source”) and sound shaping by the supra-laryngeal vocal tract (acting as the “filter”). The vocal folds determine the fundamental frequency (F0), while the vocal tract modifies the signal to produce formants (F1, F2, etc.) and shape energy distribution (amplification or attenuation of the signal). The theory has been widely applied to a range of mammalian species, including domesticated animals, uncovering links between vocal output and the anatomical or physiological traits of the emitter [8,9,10].

Discriminative acoustic signals play a pivotal role in species that exhibit communal living or complex social structures, since they can mediate social interactions among animals, such as mate choice, parental investment, and coordinated group behaviors. They can also be noticed in challenging farming stimuli like feed frustration or anticipation, mating signs (estrus), physical and/or visual isolation [11]. Individual-specific vocal responses are most evident when intra-individual variability is minimal and inter-individual variability is pronounced [12,13]. Numerous studies have documented the existence of distinctive vocal signatures in mammals [11,14,15,16,17,18,19], characterized by unique spectral and temporal features. These acoustic parameters could serve as reliable indicators of individual identity, facilitating recognition and social differentiation within populations [17,18,19,20].

Sheep are highly gregarious [21], with their vocal repertoire comprising two distinct types of bleats: high-pitched and low-pitched [21]. High-pitched bleats are typically emitted in contexts associated with stress or negative affective states, such as social isolation or maternal separation, and are characterized by elevated frequency and amplitude, often produced with an open-mouth posture. In contrast, low-pitched bleats are generated with the mouth closed and have a lower frequency and amplitude. They predominantly occur during early postnatal interactions between the ewe and her lamb(s), serving to strengthen the maternal–offspring bond [19,21]. Contrary to ewes, which are more vocally expressive in affiliative and caregiving contexts (i.e., ewes–lambs), rams’ vocalizations are more situational and less frequent—primarily linked to mating and dominance [21,22,23].

Generally, sheep demonstrate vocal distinctiveness, with acoustic signals varying across individuals in ways that support recognition and context differentiation [3,24,25,26]. Previous research [19,25] has predominantly examined vocal individuality within mother–offspring pairs of meat- or wool-type sheep breeds during the early postnatal period (2–15 days of age), a stage characterized by close physical proximity between them. A recent study [27] indicated that comparable levels of vocal individuality between dairy ewes and lambs persist at a later postnatal stage (40 days postpartum); however, these levels were lower than those observed during an early postnatal period [19,25]. Vocal individuality was also evident among lamb siblings; however, fewer acoustic parameters contributed to their vocal distinctiveness compared to those differentiating vocal cues between ewes and their lambs. Notably, dairy ewes exhibited vocal individuality not only during the suckling period but also at a distinct time point, that of the dry season, suggesting that individualized vocal signals may serve functions beyond immediate maternal–offspring recognition [26,27]. However, to the best of our knowledge, data on vocal individuality in rams remains limited, especially regarding its expression across various contexts.

Consequently, the aim of the present study was to investigate individuality coding in rams’ vocalizations within and across different contexts. Specifically, it aimed at investigating whether ram vocalizations are individually distinctive, their consistency across different context situations, and the variation in informational content. Additionally, we assessed which acoustic parameters most effectively encode individual identity both within and across contexts. We hypothesized that rams exhibit vocal individuality, which may vary in terms of distinctiveness and informational content depending on the context. Finally, we predicted that individual identity under different contexts would be conveyed through diverse acoustic parameters.

2. Materials and Methods

2.1. Site, Animals, and Set-Up

All data were collected in a commercial sheep dairy farm in Voiotia, Greece, during spring 2024 and before the mating period. The farm rears 120 ewes and 15 adult rams for breeding purposes, belonging, according to the pedigrees of the farmer, to the Assaf breed. All recordings were made in all 15 rams of the farm, which were kept separate (harem group; N 38°22′2.046″, E 23°7′10.83″) and approximately 600 m from ewes’ enclosures (N 38°21′46.534″, E 23°6′58.388″). According to farm records, the rams were 1.5–5 years old (mean: 3.27 ± 0.33 years) and with no relation to each other. One day before the recordings started, the wither height and the body length were measured (Appendix A:

Table A1. Age recordings (in years) and body and head measurements (cm) of the rams (Assaf breed).

Ram ID	Age (Years)	Withers Height (cm)	Body Length (cm)
1	3.5	96	98
2	2	88	98
3	2.5	96	101
4	3	94	100
5	5	97	101
6	2.5	96	96
7	5	96	98
8	4	101	99
9	4	95	92
10	4.5	96	93
11	3	87	91
12	1.5	92	98
13	2	92	94
14	1.5	91	94
15	5	98	101
Mean	3.27 ± 0.33	94.33 ± 0.96	96.93 ± 0.88

).

The vocalizations were recorded using a ZOOM SSH-6 microphone (20–20,000 Hz; Zoom North America, Hauppauge, NY, USA) attached to a recording system (H5 Handy recorder, Hauppauge, NY, USA). The microphone was suspended at the center of the experimental pen, approximately 130 cm from the ground. For shock and wind-noise reduction, the microphone was protected with a Zoom Windshield (HWS-6 Zoom Windscreen Hauppauge, NY, USA). The recording system was also equipped with a wired remote control to facilitate the vocal recordings from a distance. Recordings were carried out only when the weather was suitable. Each vocal sample was stored as a separate uncompressed file in “.WAV” format, using a 44.1 kHz sampling rate and 16-bit amplitude resolution.

To record the rams’ vocalizations, every ram was physically and visually isolated in a barn pen from the rest of the group to avoid overlapping vocalizations. The rams had undergone isolation in the past as part of standard farming procedures (i.e., animal inspection, vaccinations, etc.). Thus, they were accustomed to the process, and no novelty effect of the situation was anticipated. We recorded rams’ vocalizations when they were (a) isolated in the morning (09:00 h; hereafter “Context 1”); (b) isolated while hearing sounds from ewes’ bells caused by the owner of the farmer (without visual contact and in a distance of 20–30 m) at the external area of the enclosure (hereafter “Context 2”); (c) isolated from conspecifics in the late evening (20:00 h; hereafter “Context 3”); (d) isolated from conspecifics anticipating feed (alfa–alfa hay; hereafter “Context 4”); (e) isolated and denied feed (hereafter “Context 5”). To eliminate any influence introduced by management practices (e.g., cleaning, feed, etc.) during the process, the observations were carried out between 08:30 h and 14:00 h (Contexts 1–5) and between 20:00–21:00 h. Each animal was isolated for 3 min in total, during which the vocalizations were recorded. For every context record, all the rams were isolated on the same day following a predefined random order, which remained consistent across all contexts.

2.2. Recordings Classifications According to the Recording Contexts

The recording contexts were categorized as either positive or negative based on their presumed emotional valence [20]. The emotional valence of isolation, hearing sounds, anticipating, or denial feeding contexts was inferred based on the functions of emotions [20,28,29,30], and knowledge of livestock behavior [11,20,26,31]. Negative emotions are associated with the unpleasant-motivational system in animals, triggering avoidance of releasing stimuli, which are typically avoided by sheep in natural life [28,29,30]. On the other hand, positive emotions are part of the pleasant-appetitive motivational system, which triggers an approach towards releasing stimulus [28,29,30]. Subsequently, contexts related to physical/social isolation or denial of feed were assumed to be negative, while contexts related to feed anticipation and hearing familiar sounds related to females were assumed to be positive. Specifically, feeding anticipation was considered to be positively valenced, since feeding induces approach behavior and increases fitness in natural life. Exposure to familiar auditory sounds, particularly those related to ewes triggering the attraction of mating, is also deemed a positive context. In contrast, both physical and visual social isolation were considered to be negatively valenced, since sheep are highly gregarious, and being separated from the flock could threaten fitness. Similarly, denial of feed access was deemed a negative context, as it could induce frustration, lack of feed intake in the wild, and a threat to fitness [29,31].

2.3. Acoustic Analysis

Praat (v.6.1.09) [32] was used for the acoustic analysis. Bleats were visualized on spectrograms (FFT method, window length 0.03 s, time steps 1000, frequency steps 250, Gaussian window shape, dynamic range 50 dB). Bleats with good quality, thus low levels of background noise (i.e., sound produced by the isolated ram interacting with the enclosure) as visualized on a spectrogram, were only selected, resulting in a total of 428 high-pitched bleats (96% of the initial recorded vocalizations) to be used in further analyses (Appendix A:

Table A2. Number of vocalizations analyzed from each ram within each examined context, including the putative valence in which they were assumed.

Ram ID/Context Valence	Context 1	Context 2	Context 3	Context 4	Context 5	Total
	Negative	Positive	Negative	Positive	Negative
1	7	4	8	4	6	29
2	1	13	13	9	16	52
3	2	0	0	1	1	4
4	0	10	9	16	10	45
5	0	0	0	3	0	3
6	14	13	12	10	9	58
7	1	5	8	5	0	19
8	2	1	2	9	5	19
9	3	5	0	12	4	24
10	5	5	4	19	18	51
11	0	0	0	4	0	4
12	13	7	11	24	10	65
13	1	8	4	15	15	43
14	0	0	0	8	0	8
15	0	1	1	2	0	4
Total	49	72	72	141	94	428

). Twenty parameters related to source, filter, and intensity (

Table 1. Description of the acoustic parameters extracted from each vocalization.

Acoustic Parameter	Definition
Dur (s)	Mean duration of the call
Q50 (Hz)	Frequency value at the upper limit of the second quartile of energy
Fpeak (Hz)	Frequency of peak amplitude
F0mean (Hz)	Mean F0 frequency across the call
F0Range (Hz)	Difference between the maximum F0 and the minimum F0 frequency
FMRate (s−1)	Number of complete cycles of f0 modulation per second
FMExtent (Hz)	Mean peak-to-peak variation in each f0 modulation
F0var (Hz/s)	Average fundamental frequency variation per unit time
F0AbsSlope (Hz/s)	F0 mean absolute slope
Jitter%	Mean absolute difference between frequencies of consecutive F0 periods divided by the mean frequency of F0mean
Shimmer%	Mean absolute difference between the amplitudes of consecutive F0 periods divided by the mean amplitude of F0
F1mean (Hz)	Mean frequency value of the first formant
F2mean (Hz)	Mean frequency value of the second formant
F3mean (Hz)	Mean frequency value of the third formant
F4mean (Hz)	Mean frequency value of the fourth formant
AmpVar (dB/s)	Mean intensity variation per second
AMRate (s−1)	Number of complete cycles of amplitude modulation per second
AMExtent (dB)	Mean peak-to-peak variation in each amplitude modulation
RMS (dB)	Root mean square of the amplitude
VTLmax (cm)	Estimated maximum vocal tract length

) were extracted using Praat commands and custom-built scripts [26,27].

Source-related acoustic parameters were measured by extracting the F0 contour of each bleat using the “To Pitch (cc) command” in Praat (settings: time step = 0.01 s; voice threshold = 0.2; silence threshold = 0.05; minimum and maximum F0 = 65 Hz and 550 Hz, respectively). From this contour, the mean fundamental frequency (F0mean), the F0 range (F0Range), the second quartile of energy (Q50), and the peak frequency (Fpeak) were also calculated. F0 variability (F0var) was estimated following the method described in [9]. Additionally, the number of complete cycles of F0 modulation per second (FMRate) and the mean peak-to-peak variation in each F0 modulation (FMExtend) were computed. Measures of vocal perturbation—Jitter % and Shimmer %—were obtained using the “Jitter (local) and Shimmer (local) commands”, respectively. In addition, filter-related acoustic features (formants) were extracted using the “To Formant (cc) command” (settings: time step = 0.01 s; maximum number of formants = 4; maximum formant = 5000 Hz; window length = 0.05 s). The mean values of the first four formants (F1–F4) were measured for each bleat. Spectrograms were visually inspected to verify the accuracy of Praat’s tracking and measurement of all acoustic features. Intensity-related parameters were estimated using the “To Intensity (cc) command”. These included the mean amplitude variation per second (AmpVar), calculated as the cumulative amplitude variation divided by the total bleat duration; the number of complete cycles of amplitude modulation per second (AMRate); and the mean peak-to-peak variation in each amplitude modulation (AMExtent). The root mean square of amplitude (RMS) and the duration of each bleat (Dur) were also included in the analysis. Finally, the maximum vocal tract length (VTLmax) was estimated using the equation VTLmax = c/2·ΔFmin (ΔFmin = minimum formant spacing; c = the speed of sound in air) and following the regression method previously described [9]. The VTLmax reflects the effective vocal tract configuration during vocalization, which may exceed static anatomical measures due to laryngeal descent and neck extension. The VTLmax was not included in our acoustic analyses but was analyzed separately to account for differences between rams. Each bleat was included in the statistical analysis as a separate unit.

2.4. Statistical Analysis

All statistical analysis was conducted using the R environment (v.4.4.3) [33] and appropriate packages described as follows. For each ram vocal call, individuals with at least three calls per vocal bout were used. This was performed to maximize the use of available data while minimizing the loss of selected individuals [34].

2.4.1. Individual Information Measures

To investigate which raw acoustic parameters can encode an individual signature, the Potential of Individuality Coding (PIC) value was first calculated for each ram within each context for each of the 19 acoustic parameters. The PIC value assesses the ratio of within-individual variation (CV_i) and between-individual variation (CV_b) of each of the acoustic parameters using the following formula:

$$PIC={\displaystyle \frac{CVb}{CVi}}$$

(1)

The CV values (CV_b or CV_i) were computed according to the following equation:

$$\mathrm{C}\mathrm{V}={\displaystyle \frac{100\times \left(1+\frac{1}{4\times \mathrm{N}}\right)\times \mathrm{S}\mathrm{D}}{\mathrm{X}\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}}$$

(2)

where

N = the number of calls,

SD = the standard deviation and

Xmean = the mean value of the parameters for one specific individual (CV_i) or for one sample of various individuals (CV_b).

If the PIC value is > 1, then a parameter is potentially able to encode the individual signature, since their within-individual variation is lower than their between-individual variation [35]. All analyses were performed in the R language v.4.4.3 [33] and using the IDmeasurer package [36].

We further investigated whether the calls of the rams expressed under the different contexts carry a different individual information load. We calculated the stereotype Beecher’s index (Hs) as described in previous studies [12,13,35], which serves as a means of estimating the information content (entropy calculation; 2^Hs) of an examined parameter, forms a measure of encoding individuality, and is a robust standard method allowing for cross-species comparisons [36]. The Hs values stand for bits of information and can be translated as the approximate number of individuals that can theoretically be distinguished using a given signal, calculated as 2^Hs. To compute Hs value, the Kaiser-Meyer-Olkin criterion (function KMO, EFAtools package [37]) was first calculated, which confirmed the data were appropriate for a principal components analysis (PCA; overall KMO score: 0.72). Then a PCA including all extracted acoustic parameters was run (function prcomp, stats package [33]), and we used all extracted principal components to calculate Beecher’s information statistic [12] (function calcHS, IDmeasurer package [36], which automatically provides Hs values for both all and significant variables only).

2.4.2. Vocal Individuality Within and Across Vocal Signals

Individuality among rams within each context was explored using a discriminant function analysis (DFA; one analysis per each context; Context 1: n = 5 rams; Context 2: n = 9 rams; Context 3: n = 7 rams; Context 4: n = 13 rams; Context 5: n = 9 rams) to assess the extent to which individuals (rams) could be classified based on their bleats. The selected calls of each individual were subjected to a discriminant function analysis (DFA) with a leave-one-out cross-validation, performed separately for each call within each context and the temporal data (MASS package, function lda; [38]), with the individuals (rams’ ID number) as a test factor and with the scores of PCs with eigenvalues >1 as response variables. To estimate the overall significance of the classification with DFA, we used a two-tailed binomial test corrected with the percentage of classification expected by chance (Fisher test; stats package [33], function fisher.test).

We further performed a permuted discriminant function analysis (pDFA; [39]) pooling all available vocalizations from all individuals and from five contexts together to investigate the individuality of rams’ vocalizations across the vocal cues (repertoire) of all contexts. A pDFA with a nested design was conducted (pDFA.nested function provided by [39], based on function lda of the MASS package [38]), using rams’ ID (individual) as a test factor and context as a restriction factor, using the scores of PCs with eigenvalue >1 as input variables. We ran a total of 1000 permutations for the analysis.

2.4.3. Differences in Vocal Individuality Across Contexts

To investigate the difference in the vocal individuality of calls across the examined contexts, we performed a permuted discriminant function analysis (pDFA; [39]) pooling all available vocalizations from all individuals and contexts together, as above. In this case, context was included as the test factor and the individual identity of the rams (animal’s ID number) as the control factor.

To investigate the direction of changes among contexts in the VTLmax parameter as well as in the acoustic parameters, we used linear mixed models (LMM; lmer function, lme4 package [40]). To avoid running multiple models separately on each acoustic parameter, enhancing the possibilities for biases (type I error), one representative source-related parameter, one filter-related parameter, and one intensity-related parameter were chosen as previously reported [41], based also on the highest values of PIC. These parameters are also considered valuable and reliable indices of vocal communication in mammals [20,42]. The chosen parameters were (a) F0mean, (b) the formant dispersion (the average spacing between the formants) calculated as Fd = [(F2mean − F1mean) + (F3mean − F2mean) + (F4mean − F3mean)]/3 [20,43], and (c) AmpVar. The models included the acoustic parameter as a response variable, the context as a fixed factor, the body length as a covariate to control for body size, and the ram identity (animal’s ID number) as a random factor to control for repeated measures of rams within contexts. Data normality was assessed using Q-Q plots (qqnorm function, stats package [33]), and where they were needed, they were log-transformed (F0mean). The p-values (LMMs) were obtained using the PBmodcop function (pbkrtest package [44]).

3. Results

The mean values of the determined acoustic parameters in each context (Contexts 1–5) are presented in

Table A3. Mean values (±std. error) of each analyzed acoustic parameter in each context (Contexts 1–5).

Acoustic Parameter	Context 1	Context 2	Context 3	Context 4	Context 5
Dur (sec)	0.78 ± 0.02	0.80 ± 0.02	0.70 ± 0.02	0.84 ± 0.02	0.88 ± 0.02
Q50 (Hz)	695.27 ± 32.55	594.66 ± 23.75	690.66 ± 22.84	557.41 ± 23.31	460.10 ± 21.63
Fpeak (Hz)	480.19 ± 36.01	527.12 ± 26.87	536.58 ± 26.25	476.58 ± 28.34	471.26 ± 31.24
RMS (dB)	0.025 ± 0.003	0.064 ± 0.006	0.054 ± 0.005	0.053 ± 0.003	0.049 ± 0.003
F0mean (Hz)	147.60 ± 7.85	171.65 ± 9.46	157.96 ± 7.24	154.62 ± 4.93	174.72 ± 8.33
F0Range (Hz)	203.21 ±20.47	192.61 ± 14.43	203.84 ± 16.99	229.68 ± 12.85	284.41 ± 18.41
FMRate (s−1)	20.22 ± 0.56	21.19 ± 0.48	21.55 ± 0.45	20.36 ± 0.37	20.43 ± 0.42
FMExtent (Hz)	51.78 ± 3.81	47.33 ± 5.15	48.02 ± 3.43	57.24 ± 5.39	63.65 ± 5.59
F0var (Hz/s)	1018.28 ± 69.17	948.40 ± 89.23	1015.39 ± 70.88	990.55 ± 54.89	1225.90 ± 97.21
F0AbsSlope (Hz/s)	1172.45 ± 74.69	1067.58 ± 100.65	1133.09 ± 76.71	1120.87 ± 63.52	1429.98 ± 115.16
Jitter%	7.48 ± 0.23	6.49 ± 0.24	6.94 ± 0.26	6.73 ± 0.20	6.91 ± 0.24
Shimmer%	20.94 ± 0.45	19.57 ± 0.50	21.07 ± 0.45	19.41 ± 0.38	19.72 ± 0.42
F1mean (Hz)	695.03 ± 18.20	624.71 ± 9.24	645.43 ± 9.06	764.90 ± 14.14	649.84 ± 12.37
F2mean (Hz)	1538.27 ± 17.99	1515.03 ± 14.33	1525.22 ± 11.83	1615.18 ± 17.15	1567.34 ± 12.52
F3mean (Hz)	2367.12 ± 38.00	2278.38 ± 23.14	2304.07 ± 21.07	2467.91 ± 22.71	2403.16 ± 17.71
F4mean (Hz)	3456.44 ± 26.64	3403.63 ± 17.43	3400.23 ± 21.90	3547.15 ± 19.76	3495.27 ± 14.83
AmpVar (dB/s)	116.58 ± 5.17	101.80 ± 3.49	108.72 ± 5.02	94.55 ± 2.63	98.94 ± 3.73
AMRate (s−1)	18.51 ± 0.47	18.58 ± 0.44	18.10 ± 0.49	19.01 ± 0.30	19.82 ± 0.42
AMExtent (dB)	6.51 ± 0.38	5.54± 0.17	6.27 ± 0.34	5.05 ± 0.14	4.97 ± 0.15

(Appendix A). Regarding PIC values, most of the estimated acoustic parameters showed considerable informational content (PIC > 1), indicating that they could potentially encode an individual vocal signature (

Table 2. Estimated PIC values (Potential Individual Coding) of the 19 raw acoustic parameters estimated for the ram calls within the five contexts.

Context	Dur	RMS	Q50	Fpeak	F0mean	F0AbSlope	F0Range	F0VAR	FMRate	FMExtent	F1mean	F2mean	F3mean	F4mean	Jitter%	Shimmer%	AmpVar	AMRate	AMExtent
1	1.36	1.56	1.48	1.26	1.81	1.51	1.44	1.37	1.12	1.52	1.05	1.16	1.57	1.18	1.09	1.05	1.83	1.32	2.01
2	1.24	1.23	1.24	1.13	1.48	1.38	1.05	1.37	1.21	1.34	1.09	1.30	1.35	1.10	1.25	1.40	1.41	1.28	1.08
3	1.27	1.30	1.25	1.21	1.60	1.26	1.23	1.27	1.16	1.20	1.19	1.28	1.26	1.18	1.20	1.19	2.11	1.46	1.91
4	1.25	1.29	1.26	1.74	1.23	1.25	1.07	1.22	0.99	1.41	1.58	1.98	1.99	1.76	1.23	1.10	1.17	1.35	1.22
5	1.23	1.36	1.32	1.24	1.18	1.24	0.92	1.29	1.17	1.30	1.04	1.32	1.28	1.11	1.35	1.23	1.85	1.25	1.40

). Depending on the context, the highest PIC values being the most promising across vocal signals were observed for (a) AMExtent and AmpVar (Contexts: 1, 2, and 5; (b) F0mean and AmpVar (Context 2); and (c) F3mean and F4mean (Context 4).

The 2^Hs information value was also computed for each call within each context as an estimate of how many individuals could potentially be classified for each context. According to

Table 3. Information content using the entropy calculation approach (2Hs; Beacher’s information statistic) for each determined acoustic parameter in the analyzed bleats of rams. Hs all represents the Hs summed over all variables in the data set, while Hs sig is the Hs summed over variables that differ significantly between individuals. Hs values indicate the signal capacity of a vocal cue, and not actual perception or recognition by the animals.

Context	Hs all	2Hs all	Hs sig	2Hs sig	Interpretation
1	1.01	2.01	1.11	2.16	Allows the distinction across context for at least 2 individuals
2	1.25	2.38	1.06	2.08	Allows the distinction across context for at least 2 individuals
3	1.06	2.08	0.68	1.60	Allows the distinction across context for at least 2 individuals
3	2.78	6.87	2.61	6.11	Allows the distinction across context for at least 6 individuals
5	1.32	2.50	1.01	2.01	Allows the distinction across context for at least 2 individuals

, calls in each context contained low information capacity since they could assist in distinguishing at least two individuals, with the exception of calls in Context 4, where the highest number of bits was achieved (2.61; distinguishing across the call at least six individuals).

The results of the DFA concerning the correct classification of rams’ calls in each examined context are shown in Figure 1. According to the results, vocalizations could be correctly classified by an individual within each context better than by chance (DFA; relative cross-classification level > 3.2, p < 0.0001 in all contexts examined;

Table 4. Results of discriminant function analysis (DFA; cross-clarified) on rams’ vocalizations within each context.

Context	Correctly Classified (%)	Chance Level (%)	Relative Value	Significance
1	80.91	25.40	3.19	p < 0.0001
2	59.42	13.09	5.54	p < 0.0001
3	76.19	15.65	4.87	p < 0.0001
4	71.74	10.26	6.99	p < 0.0001
5	64.44	13.43	4.80	p < 0.0001

; Figure 1). The correct percentage classification was 80.95%, 59.42%, 76.19%, 71.74%, and 64.44% for each of the examined contexts, respectively (Contexts 1–5). In Figure A1 (Appendix A) are also presented the percentages (%) of correct classification of each ram to itself based on their vocalizations within each context (Contexts 1–5). Furthermore, parameters related mainly to amplitude and F0 contour and higher formants were mostly correlated with the first two (LD1, LD1) discriminating roots (Appendix A:

Table A4. Correlation coefficients between individual acoustic parameters and the first two linear discriminant functions (LD1 and LD2) across all examined behavioral contexts (Context 1–5). These values reflect the relative contribution of each parameter to the discrimination of vocalizations among individuals within each context.

Parameter	Context 1		Context 2		Context 3		Context 4		Context 5
	LD1	LD2	LD1	LD2	LD1	LD2	LD1	LD2	LD1	LD2
Dur	−0.25	0.38	0.47	0.36	0.34	−0.57	0.08	0.51	0.67	−0.14
RMS	0.42	0.35	0.07	−0.32	0.01	0.26	−0.50	−0.45	−0.03	−0.40
Q50	0.54	−0.40	0.31	−0.43	0.51	0.14	−0.62	−0.05	−0.44	−0.07
Fpeak	0.58	−0.20	0.22	−0.36	0.28	0.19	−0.51	−0.11	−0.10	−0.23
F0mean	−0.60	−0.46	0.37	0.04	0.71	−0.02	0.11	−0.12	0.38	0.17
F0AbSlope	−0.59	−0.47	0.16	0.46	0.48	−0.32	0.18	0.02	0.53	0.08
F0Range	−0.33	−0.56	0.41	0.30	0.45	−0.32	0.29	−0.01	0.52	0.13
F0VAR	−0.53	−0.35	0.18	0.46	0.49	−0.28	0.17	0.04	0.54	0.12
FMRate	0.06	0.29	0.17	0.12	0.19	−0.16	−0.20	0.17	0.03	0.37
FMExtent	−0.51	−0.48	0.08	0.37	0.42	−0.27	0.12	−0.08	0.53	0.04
F1mean	0.21	−0.56	0.20	−0.31	0.29	−0.16	−0.02	−0.22	−0.19	−0.01
F2mean	−0.50	−0.07	−0.62	0.26	−0.46	−0.19	0.83	−0.31	−0.31	0.72
F3mean	−0.67	−0.15	−0.33	0.50	−0.12	−0.38	0.71	0.08	0.08	0.41
F4mean	−0.30	0.16	−0.10	0.50	0.31	−0.45	0.42	0.10	0.18	−0.17
Jitter%	−0.17	0.23	0.48	0.45	0.38	−0.51	0.08	0.51	0.76	−0.07
Shimmer%	−0.22	0.13	0.57	0.33	0.45	−0.20	0.37	0.41	0.75	0.03
AmpVar	−0.80	−0.26	0.50	0.23	0.86	0.20	0.29	0.42	0.82	0.04
AMRate	−0.32	0.43	0.62	0.34	0.25	−0.72	0.34	0.51	0.68	−0.08
AMExtent	−0.49	−0.47	0.08	0.02	0.57	0.47	0.04	0.09	0.56	0.19

).

Regarding the correct classification of individuals across all vocalizations, the percentage of individuals correctly classified across all contexts was 29.76% (pDFA; relative cross-classification level = 3.59, p = 0.001,

Table 5. Results of permuted discriminant function analysis (pDFA) for (A), across all vocal signals of rams, and (B), between the five contexts.

A. Across All Calls
No of individuals	13
Total no of calls	402
Correctly classified %	71.56
Chance level%	48.01
p-value for classified	0.002
Correctly cross-classified %	29.76
Chance level for cross-classified %	8.27
Relative cross-classification level	3.59
p value for cross-classified	0.001
B. Between Contexts
No of contexts	5
No of individuals	13
Total no of calls	402
Correctly classified %	45.18
Chance level%	31.2
p-value for classified	0.001
Correctly cross-classified %	38.79
Chance level for cross-classified %	23.83
Relative cross-classification level	1.63
p value for cross-classified	0.001

). Figure 2 presents examples of vocal individuality variation in the examined contexts. Further, according to Table 5, 38.79% of the cross-validated rams’ calls could be classified in the correct context by the pDFA above chance levels (p < 0.001).

The VTLmax did not significantly differentiate between the contexts (F = 0.61, p > 0.05; Context 1: 25.57 ± 0.91 cm; Context 2: 24.79 ± 0.81 cm; Context 3: 24.50 ± 0.83 cm; Context 4: 24.61 ± 0.74 cm; Context E: 24.61 ± 0.79 cm). Since VTLmax did not vary significantly between contexts, it was not included as a control factor in the LMMs that were conducted for the acoustic parameters. Consequently, the F0mean was not found to differ significantly between contexts (F = 1.61, p > 0.05; LLM log-transformed scale: Context 1: 2.14 ± 0.03; Context 2: 2.17 ± 0.03 Hz; Context 3: 2.16 ± 0.03 Hz; Context 4: 2.16 ± 0.02 Hz; Context 5: 2.19 ± 0.03; for interpretability purposes the corresponding back-transformed values were: Context 1 = 143.31 ± 12.85 Hz, Context 2 = 163.26 ± 11.02 Hz, Context 3 = 153.15 ± 11.37 Hz, Context 4 = 155.31 ± 9.38 Hz, and Context 5 = 169.64 ± 10.46 Hz). Contrary to this, formant dispersion (Fd) was found to significantly differ between contexts (F = 6.88, p < 0.001, Figure 3a). Vocalizations in Context 5 had a higher formant dispersion (970.71 ± 12.71 Hz) compared to rams’ vocalizations in Context 2 (p < 0.05; 948.78 ± 12.64 Hz), Context 3 (p < 0.05; 946.41 ± 12.84 Hz), and Context 4 (p < 0.001; 936.28 ± 11.74 Hz). Regarding AmpVar, it was significantly different between contexts (F = 8.76, p < 0.001, Figure 3b).

Specifically, rams’ vocalizations in Context 1 had significantly (p < 0.01) higher AmpVar values (111.55 ± 6.93 dB/s) compared to those of Context 4 (91.87 ± 5.70 dB/s) and Context 5 (93.79 ± 6.01 dB/s). The same was noticed when the respective values of Context 3 (113.39 ± 6.40 dB/s) were compared to those of Context 4 (91.87 ± 5.70 dB/s; p < 0.001) and Context 5 (93.79 ± 6.01dB/s; p < 0.001).

4. Discussion

This study examined whether high-frequency vocalizations produced by rams in different contexts (positive or negative) convey individual-specific acoustic signatures. Analysis revealed that multiple vocal parameters contribute to the acoustic distinctiveness among individuals. Furthermore, the degree of vocal individuality exhibited was found to be context dependent. To our knowledge, this is the first evidence demonstrating that male sheep preserve vocal identity markers across diverse husbandry conditions. These findings advance the current understanding of vocal communication in Ovis aries and suggest further exploration for potential applications of non-invasive welfare monitoring in farming environments.

All the examined acoustic variables showed the potential to effectively convey in rams’ vocalizations information about the context in which they find themselves (PIC > 1), likely reflecting distinct affective valences. Higher individuality was noted mainly in parameters related to source or amplitude characteristics. These findings align with previous studies indicating that fundamental frequency (F0) and amplitude parameters are key contributors to vocal distinctiveness in lambs and ewes [25,26,27]. Moreover, the observed expression patterns are consistent with vocal individuality levels observed across mammals. Source-related acoustic parameters (e.g., fundamental frequency, F0) and spectral-related features have been widely recognized as reliable cues of individual identity in various mammalian species, including deer [9,10,45], giant pandas [14,46], sows [43,47], and ruminants [11,24,25,26,27,48]. Interestingly, the higher formant frequencies (F3 and F4) revealed higher encoded information related to individual identity in rams under the positive context of feed anticipation, which, to the best of our knowledge, has not previously been documented in ungulates. High values of the higher formants have been reported previously under positive contexts compared to negative grunts in pigs [43]. In heifers experiencing similar feed anticipation contexts, vocal individuality was primarily linked to amplitude-related parameters [11]. In our case, the enhanced individuality conveyed through F3 and F4 may be functionally significant in rams, particularly in relation to direct access to feed resources. This could reflect an adaptive need to communicate individual identity over longer distances. Higher frequencies, such as those in F3 and F4, are known to propagate more effectively in open environments, which may be advantageous in natural settings, where searching and locating food is a competitive and spatially challenging task.

We further examined the information content of vocalizations using Beecher’s information statistic (Hs), considered one of the most robust and reliable indices for vocal individuality [36]. The Hs index reflects the entropy embedded in a vocal cue and estimates the number of bits of information conveyed by a signal, that is, how many individuals can be reliably discriminated based on their acoustic features [12]. Species that live in large social groups are generally expected to produce vocalizations with high information content, resulting in elevated Hs values [2]. However, this expectation was not met in our study, where Hs values ranged from 0.68 to 2.61 across the vocal cues analyzed. Accordingly, the noted values are comparable to those reported for Myiopsitta monachus (Hs up to 2.77; [49]), but notably lower than those observed in various sciurid rodent species (Hs = 4.89–7.76; [2]), Alle alle (Hs = 3.58–5.39; [34]), and Tursiops truncatus (Hs = 13.7; [50]). A profound explanation for the relatively low Hs values in our data set could be the limited vocal repertoire of sheep, which predominantly consists of two types of bleats—low-pitched and high-pitched. Further, a study on zebra finches demonstrated that vocalizations used in specific contexts, like feed anticipation, showed reduced acoustic variability, supporting the notion that emotionally uniform contexts can lead to simplified vocal patterns [51]. Therefore, when vocalizations are produced in uniform contexts such as feed anticipation, individuals may rely on a restricted set of acoustic patterns, further reducing entropy and overall information content.

The analyzed vocalizations exhibited, also, individual-specific acoustic features, allowing for accurate classification of signals to specific individuals within each context. The noted variation in the accuracy of classification (59–80%) can be attributed to the context, reflecting the need for higher individual distinctiveness, mainly under negative contexts. Importantly, we observed that vocal distinctiveness was retained across different contexts, indicating a degree of acoustic stability. This suggests that identity cues are not merely context-dependent but may reflect intrinsic vocal characteristics. Previous studies have also reported vocal individuality in lambs at different postpartum time points, suggesting that even young animals exhibit distinct acoustic signatures early in life [24,25,27]. Additionally, ewes have been shown to modulate their vocalizations depending on social and environmental contexts, i.e., dry season or weaning period [26,27], further supporting the notion that vocal cues encode both identity and situational information. However, when call types from all contexts were pooled, classification accuracy declined (~30%). This reduction likely reflects increased acoustic variability due to different contexts, which can obscure individual-specific features and lead to overlap between individuals. Maintaining vocal individuality across different signal types is not a universal trait among animals. Among mammals, vocal individuality has been documented in species such as bottlenose dolphins (Tursiops truncatus), whose signature whistles are highly distinctive and stable over time [50], and in primates like baboons and macaques, where vocal cues convey identity and social status [52]. Ungulates also demonstrate vocal individuality, though it can be influenced by emotional context. Domestic goats (Capra hircus) and goitred gazelles (Gazella subgutturosa) show increasing vocal distinctiveness with age and social experience [8,53]. However, individuality in vocalizations may diminish under high arousal or negative emotional states, as shown across ungulate species including cattle, pigs, horses, and wild boars [11,43,54,55]. Future studies should explore the extent to which these identity cues are perceptible to conspecifics and whether they influence social interactions or group dynamics. Additionally, investigating the neural or physiological mechanisms underlying vocal consistency across contexts could provide deeper insight into the evolution of individual vocal signatures.

In addition to individual identity, the permutated discrimination analysis revealed a moderate classification accuracy (39%; 1.65 times better than random classification) of rams’ vocalizations, indicating that the discriminant functions captured meaningful structure in the data beyond chance expectations and further that rams’ vocal cues could encode some information about their affective state. This is in accordance with previous studies indicating context-specific vocal modulation in other species, including cattle [48], pigs [43,55], and primates [52]. In addition, our results support the notion that animal vocalizations are not solely identity markers but can simultaneously convey emotional states or motivational contexts. The ability to discriminate between contexts based on acoustic structure suggests that rams modulate their vocal output in response to situational factors, such as feed anticipation or social isolation. Understanding the dual encoding of identity and emotion in vocal signals may offer deeper insights into the evolution of complex communication systems in social mammals.

The comparative analysis of acoustic parameters across contexts revealed that rams’ vocalizations under negative emotional states—such as feed frustration and nocturnal isolation—were characterized by greater formant dispersion and increased amplitude variation compared to those produced during positive contexts (e.g., feed anticipation). These findings suggest that negative affective states lead to deeper and more unstable calls. Formant dispersion may reflect physiological changes in vocal tract configuration due to heightened arousal, while amplitude variability could indicate reduced vocal control or urgency in signal transmission. Although the estimated values of VTLmax did not differ significantly between the examined contexts, similar values (e.g., 21.3 ± 0.2 cm) have been reported in adult goats, a species closely related to sheep, suggesting changes in tract configuration under different contexts [56]. In sheep, craniometric measurements of basal skull length in Romanov male rams [57] imply a lower respective vocal tract length (17.6 ± 1.88 cm) in the resting (static) position or higher (24.6 ± 2.1 cm) in Barbados Black Belly sheep [58]. To the best of our knowledge, direct anatomic evidence of laryngeal retraction (i.e., video validation to assess dynamic vocal tract adjustments) is currently lacking. Therefore, any potential implication of tract elongation during vocalization requires further investigation to assess its plausibility in relation to breed and formant dispersion. However, physiological changes in vocal tract configuration have been observed in other ruminants [9,45,53] and other species, including pigs and cattle, where vocal cues shift in response to emotional intensity [43,48]. In primates, emotional arousal has been linked to changes in pitch and temporal dynamics [52]. However, previous studies in sheep revealed that amplitude modulation and jitter were the key acoustic parameters contributing to individual vocal identity in ewes and lambs during a short separation among them in an early or later postpartum period [25,27]. Sibling pairs also demonstrated unique vocal signatures, particularly through variations in fundamental frequency (F0) contour parameters. Notably, the individuality of ewes’ vocalizations was not limited to the suckling phase; it was also observed during the dry season, indicating that vocal distinctiveness persists across different reproductive stages [27]. In rams, the observed acoustic modulations may serve adaptive functions—such as signaling distress or soliciting social support. Considering the presence of individuality, they may also reduce the clarity of individual vocal signatures, as emotional modulation can mask identity-related features. This dual encoding of affect and identity presents a complex dynamic in vocal communication, opening avenues for further investigation into how conspecifics interpret and respond to such signals. For example, beyond individual recognition, these vocal cues may play a pivotal role in shaping group cohesion and social structure. In highly social species such as sheep, the ability to discern emotional states and individual identities through vocalizations could influence affiliative behaviors, stress responses, and collective decision-making. For instance, the vocal expression of distress or contentment by a single member may elicit corresponding behavioral adjustments in nearby individuals, potentially amplifying emotional states across the flock. This, potentially, raises questions about the extent to which vocal signals contribute to emergent group-level phenomena such as synchronization, leadership, and conflict resolution. Investigating these dynamics could provide deeper insight into the mechanisms of social modulation in animal groups and inform welfare practices by identifying vocal markers of group stability or disruption.

While vocalizations often contain a range of acoustic parameters, not all of these are necessarily meaningful or detectable to receivers in social species [59]. Depending on the context and species, certain parameters may contribute more significantly to vocal distinctiveness and/or transmission of context information. To determine which features truly function as markers of individual identity or contextual situations, playback experiments using bleats with altered acoustic structures would be essential. Such experiments could help clarify whether specific parameters are perceived and utilized by the receiver, as the presence of encoded information does not a priori guarantee their vocal signature.

5. Conclusions

In conclusion, our findings revealed clear individual distinctiveness in rams’ vocalizations, which varied depending on the emotional context. This individuality was most pronounced in negative contexts, particularly during social isolation or feed frustration. We also observed that vocal distinctiveness was retained across different contexts, indicating a degree of acoustic stability. Moreover, vocalizations were correctly classified according to context above chance levels, suggesting that rams can convey some information about their affective state through their acoustic signals. Among the examined parameters, formant dispersion and amplitude variation emerged as key features distinguishing vocal output between negative and positive emotional states. The dual encoding of affect and identity within these vocalizations reflects a complex dynamic in animal communication, suggesting the presence of a particularly reliable vocal recognition system, and highlights the need for further research into how conspecifics perceive such acoustic signals.