Conceived and designed the experiments: JF SS OA. Performed the experiments: GF SS MH. Analyzed the data: JF SS. Wrote the paper: JF SS. Analysis of vowels: OA. Analysis of free speech: RJ. Performed the laryngoscopy: EK.
The authors have declared that no competing interests exist.
The human voice provides a rich source of information about individual attributes such as body size, developmental stability and emotional state. Moreover, there is evidence that female voice characteristics change across the menstrual cycle. A previous study reported that women speak with higher fundamental frequency (F0) in the high-fertility compared to the low-fertility phase. To gain further insights into the mechanisms underlying this variation in perceived attractiveness and the relationship between vocal quality and the timing of ovulation, we combined hormone measurements and acoustic analyses, to characterize voice changes on a day-to-day basis throughout the menstrual cycle. Voice characteristics were measured from free speech as well as sustained vowels. In addition, we asked men to rate vocal attractiveness from selected samples. The free speech samples revealed marginally significant variation in F0 with an increase prior to and a distinct drop during ovulation. Overall variation throughout the cycle, however, precluded unequivocal identification of the period with the highest conception risk. The analysis of vowel samples revealed a significant increase in degree of unvoiceness and noise-to-harmonic ratio during menstruation, possibly related to an increase in tissue water content. Neither estrogen nor progestogen levels predicted the observed changes in acoustic characteristics. The perceptual experiments revealed a preference by males for voice samples recorded during the pre-ovulatory period compared to other periods in the cycle. While overall we confirm earlier findings in that women speak with a higher and more variable fundamental frequency just prior to ovulation, the present study highlights the importance of taking the full range of variation into account before drawing conclusions about the value of these cues for the detection of ovulation.
An understanding of human sexuality is crucial for the evolutionary analysis of hominine morphology, life history, behavior and social organization
Despite the growing evidence that visual and olfactory modalities may be important as indicators of ovulation in humans, results regarding the acoustic modality, that is whether women provide cues of reproductive state in their voices, are scarce. Numerous studies revealed that voices are a rich source of information, for instance with regard to gender and age
Bryant and Haselton
Based on these previous studies assessing the hormonal influence on voice characteristics and the differential ratings of female voices during the menstrual cycle
The estrogen (E1G) and progestogen (PdG) profiles of the women in this study varied clearly and predictably across the cycle (
Average estrogen (E1G) and progestogen (PdG) values in the period of ten days preceding ovulation to ten days after the estimated day of ovulation. Values were compiled using data from 12 women for which we had complete hormone and acoustic data for the entire period, as well as from two women for which the first 2 days were missing.
In the free speech segments, the fundamental frequency varied marginally significantly between the different phases of the cycle (F13,273 = 1.56, P = 0.096, Partial Eta2 = 0.069, Observed Power = 0.842). The highest values of the fundamental frequency were observed in the days preceding ovulation, while the lowest values were observed during the period of ovulation (
Mean (and s.e.m.) of (A) fundamental frequency and (B) standard deviation of the fundamental frequency of free speech segments, in relation to the day of ovulation and at the onset of menstruation (m).
day relative to ovulation | −6 | −5 | −4 | −3 | −2 | −1 | 0 | 1 | 2 | 3 | 4 | 5 | 6 | M |
−6 | 0.641 | 0.518 | 0.225 | 0.598 | 0.980 | 0.054 | 0.170 | 0.290 | 0.837 | 0.251 | 0.742 | 0.720 | 0.256 | |
−5 | 0.816 | 0.850 | 0.096 | 0.326 | 0.658 | 0.149 | 0.372 | 0.555 | 0.506 | 0.497 | 0.900 | 0.412 | 0.511 | |
−4 | 0.166 | 0.110 | 0.070 | 0.251 | 0.534 | 0.223 | 0.497 | 0.696 | 0.402 | 0.633 | 0.759 | 0.322 | 0.653 | |
−3 |
|
|
0.614 | 0.501 | 0.216 |
|
|
|
0.320 |
|
0.134 | 0.392 |
|
|
−2 | 0.078 | 0.048 | 0.739 | 0.865 | 0.581 |
|
|
0.119 | 0.750 | 0.099 | 0.404 | 0.863 | 0.100 | |
−1 | 0.205 | 0.138 | 0.869 | 0.488 | 0.607 | 0.057 | 0.178 | 0.302 | 0.818 | 0.261 | 0.761 | 0.701 | 0.267 | |
0 | 0.332 | 0.467 |
|
|
|
|
0.576 | 0.406 |
|
0.459 | 0.125 |
|
0.426 | |
1 | 0.788 | 0.974 | 0.100 |
|
|
0.125 | 0.483 | 0.776 | 0.118 | 0.846 | 0.319 | 0.084 | 0.812 | |
2 | 0.346 | 0.481 |
|
|
|
|
0.995 | 0.496 | 0.212 | 0.930 | 0.483 | 0.160 | 0.959 | |
3 | 0.538 | 0.401 | 0.435 | 0.186 | 0.254 | 0.523 |
|
0.378 | 0.126 | 0.182 | 0.600 | 0.882 | 0.184 | |
4 | 0.661 | 0.837 | 0.075 |
|
|
0.094 | 0.610 | 0.860 | 0.622 | 0.301 | 0.431 | 0.135 | 0.969 | |
5 | 0.184 | 0.276 |
|
|
|
|
0.694 | 0.285 | 0.696 | 0.057 | 0.380 | 0.500 | 0.443 | |
6 | 0.377 | 0.268 | 0.592 | 0.282 | 0.371 | 0.701 | 0.064 | 0.249 | 0.072 | 0.795 | 0.193 |
|
0.136 | |
M | 0.402 | 0.289 | 0.561 | 0.261 | 0.347 | 0.666 | 0.071 | 0.269 | 0.079 | 0.831 | 0.209 |
|
0.963 |
Above diagonal: results for F0; below diagonal results for F0 SD. M: day 2 after onset of menstruation. Significant differences at the corrected level are indicated in bold.
For the sustained vowels, the fundamental frequency was highest at the beginning of the cycle, dropped somewhat in the ovulatory period and then increased again (
White triangles: mean fundamental frequency (F0); black squares: degree of unvoiceness (DUV); for the vowel ‘a’.
Cycle Phase | |||||||||
Acoustic Variable | Follicular | Pre-Ovul | Ovulatory | Luteal | Menstr. | F | P | Partial Eta2 | Observed Power |
Fundamental frequency | 213±3 | 212±3 | 211±3 | 214±3 | 214±4 | 0.90 | 0.464 | 0.011 | 0.285 |
F0 maximum | 252±4 | 251±4 | 249±5 | 254±4 | 256±5 | 1.63 | 0.165 | 0.020 | 0.502 |
F0 minimum | 185±3 | 185±3 | 185±3 | 187±3 | 185±3 | 0.35 | 0.840 | 0.005 | 0.131 |
F0 SD | 5.3±0.3 | 5.4±0.3 | 5.0±0.2 | 5.5±0.3 | 5.6±0.4 | 1.32 | 0.262 | 0.017 | 0.412 |
Degree of unvoiceness | 8.2±0.9 | 8.0±0.9 | 7.2±0.6 | 7.1±0.8 | 9.1±1.1 | 2.89 | 0.022 | 0.036 | 0.779 |
Noise-to-harmonic ratio | 0.139±0.004 | 0.138±0.004 | 0.132±0.003 | 0.141±0.005 | 0.147±0.007 | 2.50 | 0.042 | 0.031 | 0.710 |
Shimmer | 4.3±0.3 | 4.6±0.3 | 4.2±0.2 | 4.3±0.3 | 4.5±0.3 | 0.86 | 0.487 | 0.011 | 0.274 |
Jitter | 2.1±0.1 | 2.1±0.2 | 1.9±0.1 | 2.0±0.1 | 2.1±0.2 | 1.85 | 0.119 | 0.023 | 0.559 |
Mean and s.e.m for N = 23 women and results of the statistical analysis the effect of cycle phase on different acoustic variables.
GLM with subject and vowel as random factor, and menstrual cycle phase as fixed factor; df = 4,313 for all analyses. Partial Eta2 provides an overestimate of the actual effect size; observed power gives the probability that the test statistic is greater than the critical value under the alternative hypothesis.
In the perceptual preference tests, British heterosexual males who did not speak German rated the free speech samples recorded three days prior to estimated ovulation to be, marginally significantly, more attractive than those recorded in the middle of the ovulatory period (t27 = 2.662, p = 0.013). There were no significant differences between attractiveness ratings during the period of ovulation and at the onset of menstruation (t27 = −0.377, p = 0.709; see
Samples recorded on the estimated day of ovulation (‘Ovul’) were compared with those recorded 3 days prior to this day (‘Pre’), and with those recorded on the second day of the menstrual cycle.
Our analysis revealed higher than average F0 values in the free speech segments in the pre-ovulatory period, and a significant drop around the day of ovulation. Similarly, women produced speech with greater variability in F0 in the days preceding ovulation, compared to the time around ovulation. The analysis of the sustained vowels revealed no significant variation in fundamental frequency across the cycle. The only two parameters that exhibited significant variation were the degree of unvoiceness (DUV) and the noise-to-harmonic ratio (NHR), which revealed significantly higher values during menstruation.
The finding that F0 is higher prior to ovulation mirrors the results of a previous study, which reported an increase in fundamental frequency in free speech in the pre-ovulatory period
Despite a lack of clear cues to the pre-ovulatory period in the sustained vowels, we found significant changes in voice characteristics during the time of menstruation, particularly in noise-to-harmonic ratio (NHR) and degree of unvoiceness (DUV). Both variables are associated with vocal fold vibration patterns, which appear to be less stable during the time of menstruation. This leads to a less harmonic vocal pattern, which is reflected by the NHR parameter; and a higher degree of voice breaks, which is reflected by the DUV parameter. While these results indicate a relationship between endocrine milieu and voice characteristics, the underlying physiological mechanisms remain to be explored in more detail. It was previously shown that breast tissue, for example, increases in water content with rising progesterone levels, with water content peaking about 2–4 d after the progesterone peak
Our perceptual experiments revealed a marginal preference for the free speech samples recorded in the period with the highest conception risk. One problem with using free speech is that content could be a major confound. This was at least partly alleviated by the fact that German speech was presented to British listeners who did not speak German. Free speech certainly varies more strongly in relation to changes in mood and motivation than sustained vowels, and perhaps also more than standardized sentences. The lack of preference for sustained vowels recorded in any of the three periods suggests that listeners indeed pay more attention to variation at the level of the entire sentence, i.e. prosody, than to changes that become evident at the level of the phoneme, such as the degree of unvoiceness and noise-to-harmonic ratio. The effect of the mode of speech production warrants further attention, as it was shown that reading scripted texts results in a higher F0 than spontaneous speech
In previous experiments
Overall, the results regarding the relation between pitch and attractiveness indicate that the pattern may be more complicated than initially suggested. A recent study by Hughes and colleagues
Our study contributes to the growing body of evidence that a number of cues may vary during the female cycle
Taken together, our study corroborates earlier findings regarding the increase of the fundamental frequency just prior to ovulation
The study was approved by the ethics committee of the Medical faculty of the Georg-August-University Göttingen (application number 16/2/05). Subjects provided written informed consent regarding their participation in this study.
We recruited young female candidates for participation in the study through advertisement at university switchboards and word-of-mouth propaganda in a middle-sized German university town. Potential candidates were asked to fill in an anamnesis questionnaire, in which the following information was collected: typical menstrual cycle length [days], birth year, height [cm] and weight [kg], history of voice or speech training or singing lessons, vocal activity (e.g. working in a call centre), overall physical activity and health status, smoking habits and drug use. The following criteria led to exclusion from the study: reported history of speech or voice disorders; prior history of neurological disease; recent infections of the respiratory tract, asthma or regular medication; use of hormonal contraceptives during the last three months before study onset; medication containing steroid compounds; irregular menstrual cycles; smoking or working in an environment with heavy smoking; drug abuse; extreme BMI. The remaining candidates were examined by a laryngologist using a KS 4200/S laryngo-stroboscope (Labor Dr. Timcke Hamburg, Germany), and an rpSzene video documentation system (Rehder und Partner, Hamburg, Germany). All participants had normal larynges, with both vocal folds presenting normal and symmetric vibratory pattern with no pathological findings and with complete glottal closure during phonation. All voices had a normal hoarseness diagram (Göttingen hoarseness diagram using the software lingWAVES, Forchheim, Germany)
Of the 33 women who eventually entered the study, 10 had to be excluded because they had anovulatory cycles during the sampling period (N = 7), became pregnant (N = 1), dropped out due to personal reasons (N = 1) or did not collect the required daily urine samples during the peri-ovulatory period, such that timing of ovulation in them could not be assessed reliably (N = 1). The final study set, thus, comprised 23 women with a mean age of 22 years (range 18–26 years), a mean BMI of 21.0 (18.0–25.0) and a reportedly regular menstrual cycle length of 29.1 d (24–32 d). After completion of the data collection, participants were paid € 100 (plus a € 50 bonus, see below).
Ten women began sampling on the first day of menstruation; six began about a third into the cycle (mean: 9.5 days after last onset of menstruation) and seven at the beginning of the last third of the cycle (19.4 days after last onset of menstruation). Data collection took place in subjects' homes. Subjects were requested to collect urine first thing in the morning (see below). Next, they were instructed to drink one glass of water and then read aloud ten lines of text given to them, to control the amount of vocal activity before voice recording began. They then started the voice recordings using a Marantz PMD 660 Flashcard-Recorder or a Mayah Flashman Harddisk-Recorder, with a sampling rate for recording of 48 kHz (16 bits). Sennheiser pc150 headset microphones were used to ensure a fixed distance of approximately 5 cm between the microphone and the corner of the mouth. Before the beginning of the sampling period, the women were instructed how to use the recording equipment and in particular how to adjust the recording level to avoid saturation and clipping of the sound file. Women were instructed to perform the recordings alone, and no other voices or sounds were to be heard on any of the recordings. As an incentive, we informed them that a € 50 cash bonus would be paid if the recording quality met the requirements (all did). After the first three days of sampling, the research assistant checked the quality of the recordings. The women recorded the following speech samples: (1) one min of free speech (comments on weather, plans for the day etc.; of these, ten 4 s segments were extracted and the mean F0 and the SD across the segment were analyzed); (2) sustained production of the vowels /a/ (as in ‘bar’), /i/ as in ‘bee’, and /u/ (as in ‘you’) (
Spectrographic displays of the vowels /a/, /u/, and /i/ for two women in the preovulatory (pre-) and peri-ovulatory (ovul) period. Upper panel: this subject revealed no major variation in the two phases. Lower panel: this subject revealed a drop in F0 in the ovulatory period of about 20 Hz. Frequency on the y-axis and time on the x-axis. Sampling frequency 3200 Hz, FFT settings: length: 512 pt, frame 100%, Hamming window, frequency resolution 6 Hz, temporal resolution 40 ms (75% overlap).
Prior to performing the acoustic analysis, all recordings were normalized for amplitude. Then, F0 contours within each recorded sample were examined manually, by displaying the sound signal on the computer monitor, with simultaneous auditory and visual evaluation of the signal, to ensure that no octave errors occurred during the preliminary stage of the analysis.
Firstly, we determined variation in F0 and STD of F0 in a 4 s segment of free speech for all 23 subjects for the week preceding and following the estimated day of ovulation (N = 13 days) plus one day after the onset of menstruation. The mean F0 and the SD of these segments were measured using Avisoft-SASLab Pro Version 4.52 (Avisoft Bioacoustics, Berlin, Germany). For this purpose, spectrograms were created (sampling frequency of 2.2 kHz, Hamming window and 98.43% overlap) with a 1.1 kHz frequency range, a time resolution of 7 ms, and a frequency resolution of 2 Hz, from which the F0 was measured by hand every 0.2 s using the Free reticule cursor.
Secondly, from the sustained vowels, the following acoustic parameters were measured using the software package Multidimensional Voice Profile MDVP (Kay Elemetrics Corporation, Lincoln Park, NJ): mean fundamental frequency (F0) [Hz], maximum and minimum F0 [Hz], standard deviation of F0 [Hz], degree of unvoiceness (DUV), i.e. the ratio of the unvoiced part of the signal to the whole signal; noise-to-harmonic ratio (NHR), i.e. the ratio of non-harmonic components of the voice signal (noise) to the harmonic components; Jitter (cycle-to-cycle frequency perturbation), and Shimmer (cycle-to-cycle amplitude perturbation).
The values obtained from the individual recorded segments (i.e., spoken sentences or isolated vowels) were averaged by type of signal, within each recording day. This was performed to obtain a reliable representation of the voice signal. For example, the acoustic measurements obtained from the six repeated productions of the vowel /a/ within each recording day were averaged, for each participant. This yielded a mean value for each acoustic parameter, which was then submitted to the statistical analyses.
In order to determine the day of ovulation, the women collected daily first-morning urine specimens (from the mid-stream) for one full cycle. A 2 ml aliquot of the urine sample was transferred into a (bisphenol A-free) Eppendorf cup and stored frozen until hormone analysis. The research assistant collected the samples regularly and transferred them to the laboratory for analysis. The day of ovulation was determined based on the distinct changes in estrogen and progesterone secretion during the peri-ovulatory phase. These changes are reflected in the concentrations of the urinary metabolites of these hormones, namely estrone-3-glucuronide (E1G) and pregnanediol-3-glucuronide (PdG), respectively
Concentrations of urinary E1G and PdG were measured directly in unextracted samples using enzymeimmunoassays as described by Heistermann and Hodges
The free speech data were analyzed using a Linear Mixed Model, with subject as random and cycle day as fixed factor, applying the restricted maximum likelihood estimation (REML). For 8 subjects, only 13 of the 14 repeated measures were available and for one subject only 12 data points. For the remaining 14, all 14 data points were available.
For the sustained vowels, we used two approaches. Firstly, we defined specific periods in the cycle and compared the average value for those periods. The period of days 7 to 5 before ovulation was defined as the follicular phase, the following three days (4 to 2 before ovulation) as pre-ovulatory, the three days around ovulation as ovulatory, the days 5–7 post ovulation as post-ovulatory (luteal phase), and the days 1–3 after onset of menstruation as menstruation (N = 5 phases). We then calculated the mean values for each variable and subject for each of the three vowels and phase (N = 23 subjects). We used Linear Mixed Models to assess the influence of the cycle stage on acoustic features, with phase as fixed factor and subject ID and vowel category as random factors, applying the restricted maximum likelihood estimation (REML). Secondly, we examined the day-to-day acoustic variation from the period of 10 days before to 10 days after ovulation. This allowed us to explore the link between hormone levels and putative acoustic variation in more detail. Because of the sampling scheme, we were not able to collect continuous data for all of the women in the study so that this analysis is restricted to 14 subjects. To minimize the effect of random fluctuations, we applied a running mean procedure (window size = 3), including days −11 and 11, so that we obtained smoothed values for both the predictor and the dependent variables for the days −10 to 10. We then used the Generalized Estimating Equations procedure, with subject as random factor, PdG, E1G and recording day as co-variates and the respective acoustic variables as dependent variables. All statistical analyses were conducted with PASW 18 (SPSS Inc.). We refrained from post hoc power analyses of non-significant results because these do not provide more information than the p value itself
Voice samples used in these tests were recorded from ten different women at one of three time periods: ovulation (i.e. on the estimated day of ovulation), pre-ovulation (3 days before the estimated day of ovulation) or the beginning of the cycle (the second day of the menstrual cycle). Because the estimated day of ovulation has a margin of error of one day, the selection of the samples for the first two of these periods should be considered as representative of the ‘ovulatory period’ and the ‘pre-ovulatory period’, respectively. Two different sample types were used: the vowel /a/ and a 5 s section of free speech. For each female subject, four exemplars of each sample type for each time period were used in order to reduce issues of pseudoreplication. During a testing session, there were four choice tests, comparing (1) vowel /a/, ovulation vs. pre-ovulation, (2) vowel /a/, ovulation vs. beginning of the menstrual cycle (3) 5 s of free speech, ovulation vs. pre-ovulation, (4) 5 s of free speech, ovulation vs. beginning of the menstrual cycle.
Twenty eight heterosexual male English subjects were recruited for the study from the student population at Roehampton University. They were informed prior to testing that they would be taking part in a study of vocal attractiveness, but no further information on the nature of the study was provided. The subjects did not speak or understand German. During testing, subjects sat alone in a small, windowless room and listened to voices played from a Dell Optiplex desktop computer through Sennheiser HD280 PRO closed headphones. Within each of the four choice tests, there were ten trials (one for each female from whom voice recordings were used in the study). In each trial, the subject heard two voices, presented one after the other with a gap of 5 seconds in-between; the voices in each trial pair were from the same woman. The order of presentation of the two voices for each woman was decided at random, and the order in which voices of the ten women were presented within tests was also randomized. For each trial, subjects were asked to mark on a paper check sheet which of the two voices in the pair they found more attractive, i.e. this was a forced choice design.
For each subject in each of the four choice tests, the proportion of trials in which they scored the ‘ovulation voice’ as more attractive was then calculated. For each choice test, the mean of this proportion was compared to the chance value of 0.5 in a one sample t-test. As there were four choice tests carried out, a Bonferroni correction was made, such that the critical level of significance was set at 0.0125.
We thank Andrea Heistermann for support with the hormone analyses and Dr Emily Bethell for assistance with the preference tests.