Voice pitch modulation in human mate choice

Abstract

Inter-individual differences in human fundamental frequency (F0, perceived as voice pitch) predict mate quality and reproductive success, and affect listeners' social attributions. Although humans can readily and volitionally manipulate their vocal apparatus and resultant voice pitch, for instance, in the production of speech sounds and singing, little is known about whether humans exploit this capacity to adjust the non-verbal dimensions of their voices during social (including sexual) interactions. Here, we recorded full-length conversations of 30 adult men and women taking part in real speed-dating events and tested whether their voice pitch (mean, range and variability) changed with their personal mate choice preferences and the overall desirability of each dating partner. Within-individual analyses indicated that men lowered the minimum pitch of their voices when interacting with women who were overall highly desired by other men. Men also lowered their mean voice pitch on dates with women they selected as potential mates, particularly those who indicated a mutual preference (matches). Interestingly, although women spoke with a higher and more variable voice pitch towards men they selected as potential mates, women lowered both voice pitch parameters towards men who were most desired by other women and whom they also personally preferred. Between-individual analyses indicated that men in turn preferred women with lower-pitched voices, wherein women's minimum voice pitch explained up to 55% of the variance in men's mate preferences. These results, derived in an ecologically valid setting, show that individual- and group-level mate preferences can interact to affect vocal behaviour, and support the hypothesis that human voice modulation functions in non-verbal communication to elicit favourable judgements and behaviours from others, including potential mates.

1. Introduction

The human voice conveys evolutionarily and socially relevant information that affects the outcomes of mate choice and intrasexual competition. To date, most research in this area has focused on men, whose fundamental frequency (F0, perceived as pitch) appears to have been shaped by sexual selection to communicate masculinity, dominance and genetic quality [1]. Indeed, men with low-pitched voices typically have higher levels of pubertal [2] and circulating [3,4] testosterone and are perceived as more masculine and dominant than are men with higher-pitched voices ([5] for review). Low pitch variability, perceived as a monotone voice quality, also predicts men's formidability [6]. In turn, a large literature demonstrates that women generally prefer men with relatively low-pitched voices, particularly in the context of a short-term relationship, and that such men report more sexual partners and have more offspring than their higher-pitched counterparts [7]. In fact, men with low-pitched voices tend to have higher success in a range of social contexts, from mating to socioeconomic and political [5,7].

Among women, between-individual differences in voice pitch are understudied, but appear to indicate reproductive status (e.g. pre-pubertal and post-menopausal stages of fertility) and sexual receptivity [7,8]. Women with relatively high-pitched voices are typically perceived as more feminine, younger and more attractive than are women with low-pitched voices [9–12]. However, several studies suggest that a lower pitch in women is perceived as attractive or ‘sexy’ [13–16], while others fail to identify any relationship between female voice pitch and judgements of attractiveness [6]. By contrast, women with low-pitched voices are consistently judged as more dominant, competent and mature, and as better leaders than women with a higher voice pitch [12,17,18]. The trade-off implied by this dichotomy suggests that women may volitionally raise their voice pitch to signal youth and femininity, but lower their pitch in contexts where they wish to be taken seriously, or to indicate sexual interest or intimacy to a listener (see e.g. [13]).

Low-frequency vocalizations signal dominance, not only in humans, but in a wide range of animals. Many species vocalize with a lower voice pitch during agonistic interactions, thereby communicating aggression and threat, and in some species (e.g. red deer Cervus elaphus, and koalas Phascolarctos cinereus), males extend the length of their vocal tract by lowering their larynx (and thus their formant frequencies), thereby portraying a larger body size to other males in competitive contexts, as well as to potential female mates [19,20]. Importantly, such frequency changes observed in the calls of most nonhuman mammals, including primates, typically occur in response to physiological or environmental triggers. Humans, by contrast, are readily capable of volitionally (i.e. voluntarily) manipulating their vocal apparatus, owing to increased neural control [21,22]. Perhaps the most common form of volitional voice modulation in humans involves manipulating the articulators, including the lips, tongue and jaw, to produce different vowel sounds in articulated speech [23]. We can also actively modulate the pitch of our voices by adjusting the tension and the effective length of the vocal folds, or by increasing sub-glottal pressure (i.e. air flow from the lungs) [24]. Pitch modulation is important for intonation, prosody and emotional expression, and is also observed in acting and singing [22,25].

Although the capacity for voice pitch modulation in speech production is well documented, whether humans exploit this capacity to adjust the non-verbal dimensions of their voices during social interactions has not yet been systematically investigated. The ability to volitionally change the pitch of our voice may be evolutionarily advantageous. During social and sexual communication, such vocal modulation could function to honestly communicate one's motivations and emotions, but may also be used to favourably manipulate the perceptions and behaviours of others, including potential mates [22]. Indeed, men and women can volitionally lower their voice pitch when instructed to sound more masculine [26] or physically larger [27], and both sexes modulate their pitch when instructed to sound confident, dominant and intelligent [14]. Thus, not only are humans capable of vocal control ‘on demand’, but also of fine-tuning voice modulations to mimic real or perceived associations between physical and psychological traits of a speaker and their voice.

Studies examining voice modulation in mating contexts have, to date, only been conducted in the laboratory using widely varied methodologies and have produced mixed results [13,28–31]. Some studies have found that men speak with a lower and less variable voice pitch [13,29] or a lower minimum voice pitch [31], whereas others report an increase in men's mean pitch and pitch variability [28,31] when speaking with a hypothetical female mate. Likewise, women have been shown to increase [30], decrease [13,14] or not alter their voice pitch [31] when speaking to hypothetical male mates. In these studies, dating contexts were mocked (e.g. participants were asked to leave a voice message for an unknown person whose photo or video they viewed), and the attractiveness of hypothetical dating partners was either not controlled or was pre-rated by another group of participants.

Here, we recorded the voices of 30 adult men and women (aged 20–40) taking part in real speed-dating events and examined within-individual changes in their voice pitch parameters as a function of their date's overall desirability (i.e. number of successful speed dates), and importantly, their own personal mate choice preference for each dating partner. The present study is the first to examine human voice modulation in a real-world mate choice scenario. With this high degree of ecological validity and multiple mate preference measures (individual and group-level), this study addresses key limitations of past work that likely contributed to inconsistent findings.

2. Material and methods

(a). Participants

Thirty participants were recruited by a professional speed-dating company via posters and online adverts announcing local speed-dating events for single men and women aged 20–40. Participants were then assigned to one of two age groups to limit age differences between dating partners, as is typical in speed dating. Ages ranged from 20 to 33 in the younger group (mean 22 ± 1.7 in women, 28.3 ± 1.0 in men) and 25 to 40 in the older group (mean 29 ± 4.1 in women, 29.0 ± 5.6 in men). Although men were on average 3.5 years older than women (28.5 versus 24.9 years), the age difference between pairs in speed dates averaged less than a year (0.07 ± 6.9 years) and was not significantly different from zero (t₂₁₃ = −0.137, p = 0.89). Before confirming their participation, interested individuals were informed that their voices would be recorded during the event for subsequent acoustic analysis as part of a research study. All participants provided signed consent before taking part in the study.

Questionnaires completed after the event confirmed that all participants were single. The reported amount of time since their previous relationship ranged from 1 to 60 months and did not differ significantly between the sexes (mean 19.3 months among women, 11.9 months among men; t₂₄ = 1.08, p = 0.29). Half of the women and a quarter of the men reported not previously having participated in a speed-dating event. Among those who had, participation rates ranged from 1 to 2 (women) and 1 to 15 (men) previous events. Men in our sample therefore reported greater experience with speed dating than did women (t₂₆ = −2.34, p = 0.03, equal variances not assumed).

(b). Procedure

We held two speed-dating events co-organized and co-directed by an experienced speed-dating host, held in a dedicated room at a local cafe. Participants arrived individually to the cafe where they were greeted by the host and researchers. Upon arrival, the host explained the speed-dating procedures to participants in small groups, and each participant was given a name tag and booklet in which to mark their personal mate choice preferences after each speed date (i.e. ‘yes’ or ‘no’). After providing informed consent, participants were fitted with a portable voice recorder and headset, and seated at one of several designated tables.

Following typical speed-dating procedures, each dating round lasted 6 min after which men rotated to an adjacent table. During the brief interlude between dates, both sexes indicated their preference for their previous date in a personalized booklet by marking ‘yes’ or ‘no’ beside that date's ID. This process continued until all men and women had dated one another. Following the final round, participants were given additional time to indicate their mate choice preferences. As is customary at speed-dating events, participants who were ‘matched’ (i.e. who indicated a mutual preference for one another) were informed by email within 24 h and given one another's contact information. Participants also completed a short questionnaire after the event, in which they indicated their age and sex, and were asked to provide demographic information and details regarding previous speed-dating experience.

(c). Acoustic recording and analysis

Participants’ voices were recorded throughout the entire duration of the speed-dating event using portable Tascam DR-05 recorders and lightweight (12 g), discreet cardioid condenser headset microphones at a sampling rate of 96 kHz and 24-bit amplitude quantization. Recordings were stored onto microSDHC media cards as uncompressed WAV files and later transferred to a laptop computer for editing and analysis. This method allowed us to obtain high-quality, directional voice recordings that would otherwise be difficult to obtain in a noisy environment using a stationary microphone.

Acoustic editing and analysis were performed in Praat v. 6.0.21 [32]. Fragments of silence, acute noise, non-verbal vocalizations (e.g. laughter) and multi-voicing (e.g. the voice of the dating partner) were first manually removed from audio files. Recordings were then segmented into multiple parts each corresponding to a given participant and a single speed date. We further split each sound file into three equal time segments (beginning, middle and end of the date; mean segment duration 50.6 ± 23 s), resulting in a total of 726 voice clips for acoustic analysis.

We used a batch-processing script to measure five parameters of fundamental frequency: mean (F0 mean), range (F0 min and F0 max), variability and contour, including standard deviation (F0 s.d.) and the coefficient of variation (F0 CV; given by F0 s.d./F0 mean [33]). All F0 parameters were measured using Praat's autocorrelation algorithm with a search range of 60–600 Hz and a time step of 0.01 s. Spurious octave jumps were manually corrected (see [34]). Perceptually, F0 mean, min and max represent the average, lower and upper ranges of a speaker's voice pitch, respectively, with relatively lower values sounding ‘deeper’. By contrast, F0 s.d. represents the absolute degree of voice pitch variability around the mean across an utterance, and F0 CV adjusts this variation to the magnitude of F0 thereby controlling for the nonlinear perception of voice pitch. Thus, F0 CV more reliably represents the perceptual salience of this F0 variability.

(d). Preference scores

We computed three types of preference score for each participant. Desirability scores represent how ‘desired’ the participant was by others and were computed by dividing the number of dates who marked the participant as ‘yes’ by the total number of dates, giving the proportion of dating partners who indicated a preference for the participant. Choosiness scores represent how ‘choosy’ the participant was and were computed by dividing the number of dates whom the participant marked as ‘yes’ by the total number of dates. Finally, match scores indicate the number of two-way preferences, that is, the number of dates on which both participants indicated a mutual preference for one another.

(e). Statistical analysis

We used linear mixed models (LMMs) with maximum-likelihood estimation to test for within-individual variation in F0 parameters (voice pitch modulation). We ran separate models for each sex due to non-independence in female and male data (pairing in speed dating), and because vocal parameters and preferences differ between the sexes.

We examined F0 modulation as a function of the personal mate choice preferences of both dating partners (i.e. indicating one another as a ‘yes’ or ‘no’), the speaker's overall choosiness score, and the date's overall desirability score. Preference variables were included as fixed factors (chose date, the participant marked that respective date as a ‘yes’; chosen by date, the date marked the participant as a ‘yes’) or fixed covariates (choosiness score of speaker; desirability score of date). Participant identity was included as a random subject variable in all models, and the age difference between each man and woman on each speed date was included as a random covariate. We additionally included time segment as a fixed factor to examine whether voice changes were more likely to occur at the beginning, middle or end of a date. These factors were first examined in a fully factorial model. There were no main or interaction effects of choosiness score of speaker on vocal parameters for either sex, therefore this variable was excluded from the final models. Time segment, although not significant, was retained in final models to avoid pseudo-replication; its inclusion did not affect the pattern of results. The final model can be described with the equation:

$$y_{ij}=(b_0+u_{0j})+\hspace{0.17em}{b}_1{X}_{ij}+\hspace{0.17em}{b}_2{X}_{ij}+b_3{X}_{ij}+b_4{X}_{ij}+(b_5{X}_{ij}+u_{5j})X_{ij}+{ϵ}_{ij},$$

where b₁ = chose date; b₂ = chosen by date; b₃ = time segment; b₄ = date's overall desirability; and b₅ = age difference between each man and woman on each speed date. The terms u_0j and u_5j estimate the variability in intercepts and slopes, respectively.

Following this, we examined significant main effects of categorical variables in the LMM using pairwise tests with Šidák correction for multiple comparisons and examined significant main effects of, or interactions with, continuous variables (i.e. date's desirability) using linear regression to illustrate the direction and strength of these relationships. For interactions between continuous covariates and fixed factors, we averaged vocal parameters within each relevant fixed category (e.g. chose date, did not choose date,) and plotted separate lines of best fit for each fixed factor.

Sex differences, effects of participant age or effects of the age difference between dating pairs on choosiness, desirability and match scores or on personal mate choice preferences were tested using one-way analyses of variance or linear regression. All tests were two-tailed at an alpha level of 0.05.

3. Results

(a). Desirability, choosiness and matches

We found that women were on average significantly choosier than men (F_1,29 = 6.74, p = 0.01): while women indicated a preference for 30% (range 0–67%) of their dates, men indicated a preference for 51% (range 13–88%). In turn, women's desirability scores were significantly higher than men's (F_1,29 = 6.72, p = 0.01), averaging 52% in women (range 0–91%) compared to 28% in men (range 0–67%). Both partners indicated a mutual personal preference for one another on 14% of dates; individual success rates for matches ranged from 0% (no match) to 33% (women) and 38% (men), and did not differ significantly between the sexes (F_1,29 = 0.61, p = 0.81). Desirability and choosiness scores correlated negatively in both sexes; however, this relationship only reached statistical significance among women (r = −0.60, p = 0.03; men: r = −0.24, p = 0.35), indicating that women who were more desirable were also choosier.

Participant age did not significantly predict desirability scores (men: F_1,16 = 1.7, p = 0.21; women: F_1,11 = 3.8, p = 0.08), choosiness scores (men: F_1,16 = 0.05, p = 0.82; women: F_1,11 = 4.4, p = 0.06) or match scores (men: F_1,16 = 0.38, p = 0.55; women: F_1,11 = 0.004, p = 0.95) in either sex, though older women showed non-significant trends towards lower desirability scores, and higher choosiness scores, compared with younger women.

(b). Voice pitch modulation

LMMs tested for within-individual changes in voice pitch parameters across speed dates (see Statistical analysis). Table 1 reports significant effects of models conducted separately for each sex.

Table 1.

Within-individual F0 modulation. Linear mixed models examining relationships among speed-dating preferences, date's overall desirability and within-individual modulation of voice pitch parameters in women's and men's voices, controlling for the age difference between dating partners. Only significant (p < 0.05) effects are reported here; for full model outputs and estimated marginal means see electronic supplementary material, tables S1 and S2. Model syntax is provided in electronic supplementary material. See also figure 1.

voice parameter	model source	d.f.1, d.f.2	F	p-value
women
F0 mean	intercept	1, 321	9411.2	<0.001
	chose date	1, 321	18.7	<0.001
	date's desirability	1, 321	19.1	<0.001
	chose date × date's desirability	1, 321	9.8	0.002
F0 min	intercept	1, 321	595.2	<0.001
	chosen by date	1, 321	4.6	0.034
F0 max	intercept	1, 321	6237.0	<0.001
	chose date	1, 321	4.0	0.047
F0 s.d.	intercept	1, 321	1571.9	<0.001
	chose date	1, 321	16.4	<0.001
	date's desirability	1, 321	14.8	<0.001
	chose date × date's desirability	1, 321	6.9	0.009
F0 CV	intercept	1, 321	2013.1	<0.001
	chose date	1, 321	7.0	0.009
	date's desirability	1, 321	5.5	0.02
	chose date × chosen by date × date's desirability	1, 321	3.9	0.048
men
F0 mean	intercept	1, 321	1089.6	<0.001
	chose date	1, 321	4.6	0.033
	chose date × chosen by date × date's desirability	1, 321	4.4	0.036
F0 min	intercept	1, 321	633.0	<0.001
	date's desirability	1, 321	6.6	0.01
F0 max	intercept	1, 321	340.2	<0.001
F0 s.d.	intercept	1, 321	173.2	<0.001
F0 CV	intercept	1, 321	257.9	<0.001

Women modulated their average voice pitch (F0 mean), maximum voice pitch (F0 max) and pitch variability (F0 s.d. and CV) as a function of their personal mate choice preferences across speed dates (figure 1a). Pairwise tests with Šidák correction showed that, overall, women spoke with a significantly higher F0 mean (p < 0.001, d.f. = 321, 95% CI for difference = 6.5 to 19.3), F0 max (p = 0.01, d.f. = 321, 95% CI = 4.2 to 36.5) and F0 variability (F0 s.d., p < 0.001, d.f. = 321, 95% CI = 4.0 to 11.5; F0 CV, p = 0.006, d.f. = 321, 95% CI = 0.005 to 0.029) towards men they chose as potential mates (marked as ‘yes’) than towards men they did not choose (marked as ‘no’). The average magnitude of women's mean voice pitch modulation (13 Hz) exceeds perceptual discrimination thresholds from vowel sounds by almost three-fold [9].

Figure 1.

Women's and men's F0 modulation across speed dates. (a,c) Main effects of personal mate choice preferences on voice pitch modulation. (b,d) Interactions between personal preferences and date's overall desirability. Columns in bar graphs represent estimated marginal means from LMMs (table 1; electronic supplementary material, table S2), where error bars represent standard errors of the means. Markers in scatterplots represent individual speed dates; markers for distinct categories are minimally offset along the x-axis to avoid overplotting and to improve visualization. Asterisks' indicate statistical significance of pairwise comparisons following Šidák correction, where ***p < 0.001, **p < 0.01, *p < 0.05 and ^nsp > 0.05. Correlation coefficients (r) are given beside each linear regression line. See embedded legend for additional details.

Importantly, these main effects were qualified by interactions between women's personal mate choice preferences and men's overall desirability scores. As illustrated in figure 1b, women raised the F0 mean and absolute variability (F0 s.d.) of their voices on dates with men they personally preferred, but only towards men with desirability scores below 0.50 (i.e. men chosen by fewer than half of their dating partners), for whom women raised their pitch by more than 20 Hz on average. By contrast, women marginally lowered these pitch parameters towards men they preferred and whose desirability scores were among the highest. On dates with men they preferred, 14% and 12% of the modulation in women's F0 mean and F0 s.d. was explained by the man's desirability score, respectively, whereas the desirability of non-preferred men did not predict women's voice pitch modulation (figure 1b). While women spoke with a less monotone voice (higher F0 CV) towards men they preferred, the magnitude of this voice change also decreased as men's overall desirability increased and was absent on dates in which neither the woman nor the man showed a personal preference for one another.

Variation in women's minimum pitch was not qualified by the date's overall desirability nor women's own preferences but rather was predicted solely by men's preferences. Women spoke with a lower F0 min on dates in which their dating partner chose them, lowering their F0 min by an average of 15.8 Hz compared to dates on which they were not chosen (p = 0.007, d.f. = 321. 95% CI = −27.3 to −4.3; figure 1a).

Men also modulated their voices across speed dates. Although the LMM indicated that men spoke with a lower F0 mean towards women they chose (marked as ‘yes’) than towards those they did not choose (table 1), pairwise tests showed that this main effect did not reach significance following Šidák correction (p = 0.26, d.f. = 321, 95% CI for difference = −11.0 to −2.9; figure 1c). Indeed, as in women, this effect was qualified by a significant interaction, indicating that men lowered their F0 mean more towards women with low than high desirability scores, and only on dates with women they chose as potential mates, particularly if those women chose them in return (figure 1d). On such dates, men lowered their F0 mean by more than 20 Hz, or approximately four times the just-noticeable difference (in vowels [9]), towards ‘matched’ women whose desirability scores were among the lowest. Like female participants, the desirability of non-preferred women did not predict men's mean voice pitch modulation (figure 1d). Linear regression further showed that men lowered their F0 min when speaking to women with relatively higher overall desirability scores, regardless of personal preference (figure 1c).

In addition to controlling for age difference in LMMs, we ran additional analyses of variance that confirmed that the age difference between sexes in each speed-dating pair did not significantly predict the mate choices of either sex in either age group (younger women: F_1,76 = 0.02, p = 0.90, older women: F_1,29 = 1.18, p = 0.29; younger men: F_1,76 = 3.74, p = 0.06, older men: F_1,29 = 0.51, p = 0.48). Moreover, our LMM showed no significant effect of time segment on voice pitch modulation (electronic supplementary material, table S1), indicating that pitch modulation emerged early in the speed date and persisted at a similar magnitude throughout.

(c). Relationships between voice pitch and desirability or choosiness

In addition to examining within-individual fluctuations in voice pitch (that is, how participants modulated their voices from date to date), we tested whether between-individual differences in pitch predicted desirability and choosiness scores. Average pitch parameters were computed for each participant by averaging F0 parameters across all dates within individuals, and regressed against their overall desirability and choosiness scores.

We found that women's F0 min explained 44–55% of the variance in their desirability scores (r = −0.66, p = 0.014, n = 13), indicating that women with lower pitch minima were more desired by men. The strength of this relationship increased after controlling for women's choosiness scores (r_p = −0.74, p = 0.006, d.f. = 10). Although relationships between women's desirability and their F0 mean (r_p = −0.45) and F0 CV (r_p = 0.37) were moderate in strength, these between-subject analyses were underpowered, thus no other voice pitch parameter explained a significant amount of variance in the desirability or choosiness scores of either sex (electronic supplementary material, table S3).

4. Discussion

The results of this study support the hypothesis that women and men modulate their voices in real-life mate choice contexts based on personal mate choice preferences and the mate quality of a potential partner. Women spoke in a higher-pitched and less monotone voice on speed dates with men they chose as potential mates; however, only if those men also had a relatively lower overall desirability score (i.e. were preferred by fewer than 50% of other women). Men lowered their mean pitch on dates with women they personally preferred, but here too, the magnitude of men's voice pitch modulation decreased as their preferred dates' overall desirability increased.

Previous studies examining voice modulation in mock dating contexts have produced conflicting results [13,28–31], possibly because dating partners were hypothetical, and their desirability was based solely on pre-rated attractiveness of photos or videos rather than on participants’ personal preferences. Here, we show that personal mate choice preferences can deviate from, interact or even conflict with group-based desirability scores. Indeed, although women generally spoke with a higher pitch (in line with [30]), and men with a lower pitch (in line with [13,31]), towards dating partners they marked as ‘yes’, this was driven largely by preferred dating partners with lower desirability scores, towards whom both sexes altered their voice pitch by more than 20 Hz (1.5 semitones in women, 2.7 in men) on average. In fact, women spoke with a lower voice pitch towards men who were both highly desired by other women and whom they personally preferred. This finding supports at least two other studies [13,14] that report that women lower, rather than raise, their voice pitch in a mating context. Our results thus highlight the need to consider both group-level and individual mate preferences in future work as they interact to influence voice modulation. Indeed, while a date's overall desirability proved to be important, it had minimal to no effect on men's and women's voice pitch modulation if they did not also show a personal preference for that dating partner.

A large body of research indicates that men with relatively low-pitched voices are preferred by women as mates and have high mate value [5,7]. Thus, the observation that men lowered their voice pitch in response to women they preferred, particularly if those women also preferred them, suggests that voice modulation in men may function to increase their reproductive success. By contrast, our results contradict the prediction that women ubiquitously feminize their voices towards preferred potential mates, as this was not the case on dates with highly desirable men. Moreover, in our sample of speed daters, women spoke with lower pitch minima towards men who preferred them, and when comparing across women, men preferred women who spoke with a lower minimum pitch, such that women's minimum pitch explained up to 55% of the variance in how desired they were by men.

This apparent conflict between the functional relevance of high versus low voice pitch in women may be resolved by considering that indexical cues to static speaker traits (e.g. age, sex) may function differently than more dynamic, social or sexual cues. Thus, while a relatively high voice pitch in women can signal youth, femininity and reproductive fecundity [7], by dynamically lowering her voice pitch a woman might be signalling sexual interest and intimacy to a man [13–15]. Alternatively, or simultaneously, she might lower her pitch to communicate social dominance or a confident and mature person, as people with low-pitched voices are often attributed traits such as competence, trustworthiness and leadership [5,17]. Indeed, recent studies suggest that young women are increasingly using a very low-pitched register, resulting in vocal fry or a ‘creaky’ quality, in professional work contexts [35,36].

Dynamic voice modulation may be especially functional when multiple modalities are available to the receiver, as they were in our study and typically are in social interactions. In a mating context, for instance, a man may gauge a woman's age, femininity and fecundity from visual and olfactory cues [37], while simultaneously interpreting social and sexual information from the dynamic properties of the woman's voice. While it is possible that a woman's age may influence the direction and degree to which she modulates her voice pitch towards potential mates, our results showed no effect of age, or age differences between dating partners, on voice modulation. Age also did not predict choosiness, desirability or successful dating matches, though it must be noted that between-subject comparisons were underpowered.

In line with evolutionary models that implicate women as the ‘choosier’ sex [38], both choosiness and desirability scores were higher among women than men, with women showing a personal preference for only one-third of their dates (compared with one-half in men). Women also modulated more vocal parameters in response to preferred dating partners than did men, including their maximum pitch, and most notably, variability in their pitch, which is known to communicate masculinity and physical formidability among men [6]. This sex difference could be tied to ability or effort. Women have been shown to more effectively modulate their vocal attractiveness ‘on demand’ than men [14]. Alternatively, for women, dates marked as ‘yes’ were likely to include men whom women were maximally attracted to and may have thus exerted maximal vocal effort towards, whereas men's longer list of potential mates probably included marginally preferred partners towards whom men might have displayed less vocal effort.

Studying voice modulation in real-life dating offers high ecological validity, yet the lack of experimental manipulation does not allow for causal inferences regarding the role of various vocal parameters, or different modalities, on mate choice preferences. This may be investigated in future work using resynthesized speech as uni-modal and within multi-modal stimuli. Given our results, experiments are now also clearly needed to gauge the role of own versus group-level mate preferences on voice modulation. Our small sample size, while adequate for capturing dynamic within-individual modulation across speed dates, limits the extent to which inferences can be made about between-individual differences in vocal parameters and behaviour. Individual difference factors, such as the influence of past dating experience, should thus be investigated in replication studies using larger samples. Finally, future studies may examine the influence of stress (e.g. due to a first date or the realization of being audio-recorded) and other emotions on voice modulation in a dating context.

Although there is some recent evidence for behavioural and contextual flexibility in the vocalizations of other mammals and great apes [22], the capacity for volitional vocal control in humans is unprecedented in its complexity, and thus in its potential breadth of functionality. Indeed, this study shows that, while integral to human speech production, voice modulation also affects the non-verbal dimensions of vocal communication during mate choice. The capacity for women and men to dynamically alter their voice pitch therefore has the potential to affect reproductive success, but beyond this, it may function to manipulate the perceptions and behaviours of others in a wide range of social, economic and political contexts [22].

Ethics

The study was reviewed and approved by the Sciences and Technology Cross-Schools Research Ethics Committee (C-REC) of the University of Sussex (ER-REBY-3, ER-KP292-3/4).

Data accessibility

The dataset supporting this article is available in the electronic supplementary material.

Authors' contributions

K.P. and D.R. designed the investigation. K.P., M.G. and A.O. collected the data. J.P prepared audio files and questionnaire data for analysis. K.P. performed acoustic analysis. K.P. and D.R. performed statistical analyses. K.P. wrote the manuscript and created the figures. The manuscript was reviewed, edited and approved by all authors, who agree to be accountable for the work.

Competing interests

We declare we have no competing interests.

Funding

This work was supported by the European Commission through a Marie Skłodowska-Curie individual fellowship to K.P. (H2020-MSCA-IF-2014-655859).