- Email: [email protected]
Acoustic and Long-Term Average Spectrum Measures to Detect Vocal Aging in Women
Acoustic and Long-Term Average Spectrum Measures to Detect Vocal Aging in Women *Paula Torres da Silva, †Suely Master, *Solange Andreoni, ‡Paulo Pontes, and *Luiz R. Ramos, )yzSa˜o Paulo, Brazil Summary: Along the normal aging process, voice tends to become weak, breathy, and loses projection, which may interfere in the communication process. One reliable way to evaluate voice quality is through acoustical analysis using, for instance, the long-term average spectrum (LTAS). The aim of this study was to identify acoustic measures, particularly LTAS’s, which characterize vocal aging in women without vocal complaints. For this purpose, 30 elderly and 30 young women were included in this study. All spoke standard Portuguese and none had a history of vocal and laryngeal alterations or respiratory diseases. On the basis of the reading task, in habitual and loud levels, the following parameters were assessed: the equivalent sound level (Leq), the speaking fundamental frequency (SFF) and, at the LTAS window, the difference between the levels of the regions of the first formant and fundamental frequency F0 (L1 L0), alpha ratio, and the amplitude levels obtained at equal intervals of 160 Hz, ranging from 0 to 8 kHz. There were significant differences between young and old voices for SFF and Leq in both levels. In the LTAS window, amplitude levels were higher for young voices, comprising all frequencies except those in the regions between 4.6–6.7 and 4.8–6.5 kHz, in habitual and loud levels, respectively. There were also significant differences regarding L1 L0 and alpha ratio between groups, in both levels.The observed differences in LTAS’s slopes, L1 L0 measures, and even Leq and SFF measures, may be attributed, to some extent, to lower subglottal pressure or a glottal setting providing a slower glottal closing speed for the elderly group. Key Words: Vocal aging–Long-term average spectrum–Acoustic analysis–Voice quality. INTRODUCTION Increasing attention has been given to the problems of the elderly population, which is growing fast all over the world as a result of impressive increases in life expectancy. Age-related laryngeal and vocal alterations are associated with other body functional declines and may be seen as a part of normal senescence and do not necessarily interfere with the communication process and, therefore, with interpersonal relationships.1 Many studies described histological as well as anatomical and functional changes of the phonatory organ as a function of age.2–9 These modifications may result in some glottic characteristics, such as vocal fold bowing and atrophy of the intrinsic laryngeal musculature, with a high incidence of edema in older women.10 The vibration pattern of the vocal folds may present alterations because of these glottic configurations that might vary as a function of the muscles affected.11–13 Although these findings are usually not related to voice complaints,2,14 when evaluated by perceptual analysis, changes in voice quality, such as lower vocal pitch, increased harshness or hoarseness, increased strain, higher incidence of voice breaks, vocal tremor, increased breathiness, and reduced loudness, are frequently perceived in aged voices.11,15–17 Perceived breathiness and reduced loudness, for instance, are generally regarded to be caused by glottal air leakage, a common finding in laryngoscopy studies of the elderly people.18,19
Accepted for publication April 6, 2010. From the *Department of Preventive Medicine, Universidade Federal de Sa˜o Paulo, Sa˜o Paulo, Brazil; yDepartment of Scenic Arts, Arts Institute, Universidade Estadual de Sa˜o Paulo, Sa˜o Paulo, Brazil; and the zDepartment of Otorrinolaryngology and Head and Neck Surgery, Universidade Federal de Sa˜o Paulo, Sa˜o Paulo, Brazil. Address correspondence and reprint requests to Paula Torres da Silva, Department of Preventive Medicine, Universidade Federal de Sa˜o Paulo, Rua Borges Lagoa, 1341— 2nd floor, Sa˜o Paulo, SP 04038-034, Brazil. E-mail: [email protected] Journal of Voice, Vol. 25, No. 4, pp. 411-419 0892-1997/$36.00 Ó 2011 The Voice Foundation doi:10.1016/j.jvoice.2010.04.002
Although vocal quality has been evaluated by different methods of perceptual analysis, they present a considerable degree of subjectivity. Computerized acoustic analysis, on the other hand, provides objective data for describing voice quality. The quantification of acoustic parameters of voice makes it easier to compare different voices for research purposes.20–23 Among the many possibilities of sound spectrum analysis, the long-term average spectrum (LTAS) has been successfully used to evaluate voice quality. This acoustic measurement provides information about the contribution of the source and the filter to the voice quality, pointing the differences between gender,24 age,25–27 profession,28–31 and dysphonic voices. 20–22,32 Several measures of the LTAS were successfully used to discriminate voice qualities, especially the ones that are related to aging voice, such as breathiness and perceived hypofunction.21,32 One of these measures, the difference between the levels of the regions of F1 and F0 (L1 L0), provides information on both, the vocal loudness and the glottic source, and can be related to voice quality. Thus, stronger F0 than F1 correlated with hypofunctional voices perceived as breathy or weak, whereas stronger F1 than F0 was correlated with hyperfunctional and loud voices perceived as resonant, tight, or strong.20–22,31,32 Another LTAS measure is the alpha ratio, which is the ratio of energy between 0–1 and 1–5 kHz.20 The level differences between frequency ranges in the LTAS reflect the slope of the source spectrum, which, in turn, is related to the glottal closing speed.33 The shape of the transglottal airflow during the closing phase is a determining factor for the amplitude of the higher harmonics of the voice source, and therefore, if the glottis closes slowly or insufficiently during phonation, the phase of vocal fold closing may present alterations, and thus, the spectral level in the upper formant region will be reduced. That is why the overall slope is also affected by changes of vocal loudness34,35 and by different types of phonations, which in turn, were strongly correlated with voice quality.32,36,37
412 Comparisons between gender using LTAS24 showed lower spectral tilt and greater levels of aspiration noise in the region corresponding to the third formant (F3) among women, which causes a more breathy quality for female voices. This acoustic clue with high amplitude values of F0 and a bandwidth increase of the lower formant frequencies seems to contribute to the perception of breathiness.38 In an LTAS study, to verify age-related changes in the source characteristics of dynamic speech, elderly women in comparison with young ones presented a tendency toward increased frequency levels at 160 Hz, significantly higher spectral amplitude levels at the frequencies of 320 Hz and 6080–6720 Hz, and significantly lower levels at the frequencies of 3040 and 3200 Hz. These events were related to the perception of breathy voice. Young women demonstrated higher spectral amplitude levels in the frequency region of 3 kHz and a tendency toward lower spectral tilt values. Elderly women presented higher amplitude levels in the region of F0 and in the frequency region of more than 6 kHz, which were related, respectively, to increased open quotient and generation of turbulent aspiration noise at the glottal opening.26 LTAS was used to investigate resonance characteristics of dynamic speech in young adulthood and old age.27 Measurements of the first three formants revealed significant lowering of their central frequencies for elderly women. A lowering in all formant frequency peaks in aging voices is consistent with anatomic data, suggesting that the vocal tract also suffers modifications and gets longer with the larynx lowering in the neck.39 Besides LTAS, other acoustic data related to aging voice have been largely investigated. Among them, lower values of the fundamental frequency (F0) was found in women’s voices and higher F0 in men in association with anatomical and functional characteristics of the vocal folds.16,17,40 Furthermore, although young women tend to present higher formant frequencies because of the smaller length of the vocal tract, all formant frequencies were found to be lower in elderly women, as the vocal tract gets longer with the larynx lowering.26,30,39 Changes in male vocal intensity with aging, unlike the female voices, have also been described in a number of studies that found that the elderly men compared with the young men presented decreased sound pressure levels (SPL) as a result of the weakening of the respiratory and laryngeal muscles.41–45 Despite numerous studies of aging voice, most of them neither controlled the SPL nor analyzed the loudness variation in female vocal aging. This study aimed to identify acoustic measures, LTAS in particular, that significantly differ between young and older women, with no vocal complaints, in different levels of vocal effort, controlling for the intensity of the voice.
METHOD Sixty women were included in this study, 30 young and 30 elderly. Young speakers ranged in age from 20 to 35 years (mean ¼ 26.8) and elderly speakers ranged from 60 to 82 years (mean ¼ 69.57). The elderly sample was drawn from an ongoing cohort of elderly women living in the community in a large urban center and randomly selected for the study.46 The young
Journal of Voice, Vol. 25, No. 4, 2011
sample was drawn from the students of the university. They were all standard Portuguese speaking, and none had a history of previous or current vocal alterations, surgery, head- and neck-related treatments, or respiratory diseases. Subjects were recorded during a read-aloud task of a 250word text, in habitual and loud levels. They were trained by the researcher to talk both in normal and strong loudness. In strong loudness, they were told to speak ‘‘as loudly as possible’’ and were prompt to repeat until adequate efforts were achieved. Our interest was to obtained different levels of intensities and not the louder emission possible. The elderly women who were unable to modulate the intensity satisfactorily were excluded from the initial sample according to the researcher’s perceptual evaluation. The duration of each recording was about 200 seconds. A Sony digitalaudio tape (DAT) recorder, model TDC-D8 and a professional unidirectional cardioid dynamic microphone (Shure, model SM58) were used to capture the voice samples. The microphone was positioned 15 cm away from the speakers who remained seated on a stool. SPL was controlled with an SPL meter, placed together, at the same distance of the sound source. A pure tone was generated, measured, and recorded for later comparisons. For the purposes of analyzing the data, we will use equivalent sound level (Leq)35 as the measure of vocal loudness, because it gives us an average over a long time window while SPL is computed over a short time window, and instead of F0, we will use speaking fundamental frequency (SFF),11 because it means that the values were obtained in long speech samples. Following the procedure outlined by other authors24,27 and, thus, facilitating comparisons between studies, these parameters of the LTAS were selected. In the LTAS window, the acoustical variables in this study were the level difference between the F1 and F0 regions (L1 L0), that is, the level difference between 300–800 Hz and 50–300 Hz, which provides information on the mode of phonation; and the alpha ratio, that is, the level difference between 0–50 Hz and 1– 5 kHz, which provides information on the spectral slope declination and the amplitudes values obtained at intervals of 160 Hz throughout the frequency range of 0–8 kHz, totalizing 50 measurements per emission. The LTAS spectra for each subject and the average spectrum for each group, as well as the other acoustic measures were all obtained automatically by Praat program (version 5.0.23).47 The frequency range used in the acoustic analysis was 0–8 kHz, with Hanning window with a time resolution of 40 milliseconds and bandwidth of 160 Hz. Only the central 40 seconds of each sample were analyzed, which is sufficient to produce an LTAS that is independent of the text.21,22 To perform the LTAS, unvoiced sounds and pauses were automatically eliminated from the registers by the program. To facilitate measurements, the spectra were normalized.48 The statistical analyses were performed taking a multivariate approach with respect to the multiple possibly correlated dependent variables (Leq, alpha, SFF, and L1 L0). This doubly multivariate design was analyzed using a multivariate profile
413
13.69 11.93 10.05 8.83 3.61 9.98 2.34 20.25 15.28 13.85 12.09 8.68 13.95 2.29 18.65 15.88 14.15 12.81 9.34 14.28 2.34 21.20 20.16 17.73 16.01 12.07 17.67 2.68 0.14 3.04 4.05 5.00 8.77 4.10 2.03 1.19 0.85 1.82 4.25 6.57 2.35 2.06 6.31 2.01 0.24 1.46 5.96 0.05 2.84 5.91 2.99 1.54 0.06 2.15 1.60 2.17 178.18 214.70 233.34 258.61 361.75 238.75 33.46 183.23 197.98 215.55 235.54 283.75 221.30 26.71 159.92 184.29 194.95 219.96 255.38 202.50 25.19 162.76 178.73 187.53 196.07 238.61 188.84 17.88 72.11 76.37 78.84 84.77 89.77 80.13 5.19 65.98 73.54 76.63 79.41 84.78 76.20 4.44 51.09 68.28 71.55 76.69 84.24 71.96 6.47 62.01 67.82 70.59 72.73 78.68 70.63 4.06 Elderly (n ¼ 30)
Minimum 1st quartile Median 3rd quartile Maximum Mean SD Minimum 1st Quartile Median 3rd quartile Maximum Mean SD Young (n ¼ 30)
Alpha Habitual (dB) L1 L0 Loud (dB) L1 L0 Habitual (dB) SFF Loud (Hz) SFF Habitual (Hz) Leq Loud (dB) Leq Habitual (dB) Statistical Measures
RESULTS Table 1 shows quartiles and minimum, maximum, mean, and standard deviation (SD) values for acoustic measures in habitual and loud levels by age group. The Leq mean values at habitual level were 70.63 and 71.96 dB and, at loud level, were 76.20 and 80.13 dB, respectively, for the elderly and the young subjects. The variation of intensity, from habitual to loud, was significant for both groups. The SFF mean values at habitual level were 188.84 and 202.50 Hz, and at loud level, 221.30 and 238.75 Hz, respectively, for the elderly and young speakers. With the variation of intensity, from habitual to loud, both groups presented significant differences with higher values for the SFF, and in both intensities, the elderly women showed significantly lower values than the young women. The mean L1 L0 values at habitual level were 1.60 and 0.05 dB, respectively, and at loud level, the values were 2.35 and 4.10 dB, respectively, for the elderly and young women. With the variation of intensity, from habitual to loud, both groups presented significant differences, and in both intensities, elderly women showed significantly lower values than the young ones. The mean alpha ratios at habitual level were 17.67 and 14.28 dB, and at loud level, 13.95 and 9.98 dB, respectively, for the elderly and young subjects. With the variation of intensity, from habitual to loud, both groups presented significant differences with higher values of this measure, and in both intensities, the elderly women showed significantly lower values than the young ones. The multivariate profile analysis showed no interaction between level and age group (the level profiles are parallel or the variables change in a similar way between the groups, P ¼ 0.1948), with significant differences because of age groups (P < 0.0001) and level (P < 0.0001). Table 2 shows the comparisons between age groups and between levels. Regarding the amplitude levels measured every 160 Hz, in habitual level, the young showed significantly higher amplitudes values in the regions of 480–4640 and 6720–8000 Hz (P < 0.005), whereas in loud level, the amplitude values were stronger for the young, in the regions of 480–4800 and
Age Group
model, with between-individual effect being age group (young and elderly) and within-individual effects being level (habitual, loud) and the interaction between level and age group. Under this setting, the interaction effect of age group and level is tested first. If the interaction term turned out to be not statistically significant (ie, the age groups are parallel with respect to level-profile means), the multivariate test on age group was calculated as a pool of the age group effects averaged across levels, and similarly, the multivariate test on levels would be based on the pool of the level effects averaged over age groups. Then, if the age group and/or the level effects are significant, the comparisons among their average levels are reported. Using Pearson’s correlation, the means for the acoustic variables were compared between the two groups in habitual and loud levels regarding age. The computer program Statistical Package for the Social Sciences (SPSS)49 was used to perform the statistical analysis with a significance level of 5%.
Alpha Loud (dB)
Acoustic and LTAS Measures to Detect Vocal Aging
TABLE 1. Quartiles and Minimum, Maximum, Mean, and SD Values for Acoustic Measures in Habitual and Loud Levels by Age Group
Paula Torres da Silva, et al
414
Journal of Voice, Vol. 25, No. 4, 2011
TABLE 2. Summary of the Pairwise Averaged Comparisons Between Age Groups and Between Levels With Respect to the Acoustic Variables
Comparison Between age groups Elderly and young
Between levels Loud and habitual
95% Confidence Interval for Difference*
Measure
Averaged Mean Difference
Standard Error
P*
Lower Bound
Upper Bound
Leq SFF L1 L0 Alpha
2.63 15.56 1.70 3.68
1.21 6.23 0.55 0.56
0.0337 0.0154 0.0033 <0.0001
5.06 28.04 2.81 4.81
0.21 3.08 0.59 2.55
Leq SFF L1 L0 Alpha
6.87 34.35 4.00 4.01
0.53 2.76 0.21 0.27
<0.0001 <0.0001 <0.0001 <0.0001
5.80 28.84 3.57 3.47
7.93 39.87 4.43 4.55
Based on estimated marginal means. * Adjustment for multiple comparisons: least significant difference (equivalent to no adjustments).
6560–8000 Hz (P < 0.005). Figures 1 to 4 show the mean LTAS for both groups in habitual and loud levels. Table 3 shows the correlations between the acoustic variables for better understanding of the findings so far. DISCUSSION The purpose of this study was to explore acoustic and LTAS measures able to discriminate between recorded voices of elderly and young women, with no voice complaints, reading a text in habitual and loud levels. It is well established in the literature that there is a correlation between vocal changes that result from the natural aging process and anatomical-physiological alterations of the vocal mechanism,2–13 and that such characteristics may be better evaluated in acoustic studies, which would be complementary to perceptual analysis of voice quality.20–27 Equivalent sound level and speaking fundamental frequency Changes in vocal intensity with aging have been widely described in a number of studies, most of them with male voices.42–45 In general, they showed lower levels of SPL
among the aged, because of the weakening of the respiratory and laryngeal muscles, causing smaller lung pressure and smaller peak airflow, with greater open quotient.43 According to Baker et al,44 old individuals generate lower levels of SPL when compared with young individuals but modulate loudness levels in a manner similar to that of the young subjects. In our study, mean Leq values were significantly affected by aging, in both levels, and by modulating from habitual to loud, in both age groups. Vocal intensity is known to vary with subglottic pressure.36,42–45 The increased subglottic pressure occurs along with the action of the laryngeal muscles, which promotes strong adduction and tension of the vocal folds, providing high values of F0.12,36,44 Vocal intensity has been found to increase with increasing fundamental frequency at any given level of subglottic pressure.45 According to the data obtained by Hodge et al43, the SPL differences between elderly and young men could be partially explained by the difference of fundamental frequency. In fact, with the variation of intensity, our results showed that both groups presented significant differences for the SFF, with higher values at loud level. At habitual
FIGURE 1. Mean LTAS for the elderly and young groups in habitual level.
Paula Torres da Silva, et al
415
Acoustic and LTAS Measures to Detect Vocal Aging
FIGURE 2. Mean LTAS for the elderly and young groups in loud level. and loud level, the mean SFF was significantly lower for the elderly group. F0 change with age has been well documented in both men and women. In women, F0 appears to remain fairly constant until menopause when a drop in fundamental frequency occurs, probably as a result of hormonal changes during menopause when the superficial layer of the vocal folds may become swollen.2,6,10,13 Our results are in agreement with the literature. According to Hodge et al43, it is possible that the SPL diferences between elderly and young men could be partially explained by the difference of fundamental frequency. In our study, lower values of SFF for the elderly women may also have contributed to the lower values of Leq. The correlation between these variables for the elderly group, although weak, was significant in contrast to the young group (Table 3). Other factors that may explain these results for the elderly group are the resonance or the articulatory factors. It is known that opening the mouth during loud emissions, in acoustic analysis, provides tuning between F0 and F1, thus reinforcing SPL. It may be possible that our elderly group presents limitation in this movement because of modifications at the oral structures, well described in the literature.16 According to Gauffin and Sundberg, the SPL of speech sounds is highly affected not only by vocal loudness but also by the frequency distance between the strongest spectrum partial and the first formant. Moreover, the intensity level greater than 1000 Hz is strongly influenced by the frequencies of
formants 1 and 2.33 But this subject is beyond the scope of this study. Difference L1 L0 Changes in vocal intensity would also be caused by changes in glottal resistance. Vocal fold bowing and atrophy plus the weakness of the laryngeal muscles with aging may lead to smaller glottal resistance to air flow, thus compromising the increase in subglottic pressure.42–45 The findings of Gauffin and Sundberg33 about the relationships between waveform and spectrum of the pulsating transglottal airflow helped to clarify acoustic aspects involved in this situation. By inverse filtering, the pressure or airflow signal at the mouth during voicing, a reasonable representation of the airflow waveform at the glottis, was obtained. Furthermore, it was found that the peak-topeak amplitude measure of the flow glottogram pulses—or waveform—is closely related to the amplitude of the source spectrum fundamental (L0), which in turn, varies considerably during phonation, presumably depending on the degree of glottal ab/adduction or the mode of phonation. Breathy phonation, in contrast to pressed phonation, shows higher values of L0 as an effect of the glottal leakage.26,30,33,39 In our study, there were significant differences between groups regarding L1 L0, with elderly women presenting lower values of this acoustic measure, even at loud intensity, when Leq
Sound pressure level (dB / Hz)
0
-10 ------ Habitual ____ Loud
-20
-30
-40
-50 0
1000
2000
3000
4000 5000 Frequency (Hz)
6000
7000
8000
FIGURE 3. Mean LTAS for the elderly group in habitual and loud levels.
416
Journal of Voice, Vol. 25, No. 4, 2011 0
Sound pressure level (dB / Hz)
-10
----- Habitual ____ Loud
-20
-30
-40
-50 0
1000
2000
3000
4000 5000 Frequency (Hz)
6000
7000
8000
FIGURE 4. Mean LTAS for the young group in habitual and loud levels. was significantly lower compared with that in young ones. With the variation of intensity, from habitual to loud, both groups presented significant higher values of L1 L0, meaning that the degree of glottal resistance increased significantly with raised loudness.23,31,37 Low levels of the F0 and high level of the F1 was largely related in literature to hypofunctional and breathy voices, whereas stronger F1 than F0 was correlated with hyperfunctional and loud voices perceived as resonant, tight, or strong.32–37 Because of elderly vocal fold configuration, glottis probably closes slowly or insufficiently, and the vibratory cycle may present alterations. In Gauffin and Sundberg’s study, when varying the type of phonation, the waveform of the glottal airflow can be changed within wide limits, and great differences can be found in the length of the closed phase, that is, the closed quotient.33 These changes were observed by measuring maximal flow declination rate. On the other hand, as closing decreases with decreased vocal intensity and breathy phonation, open quotient increases. Amplitude of higher overtones of the source fundamental frequency depends on closing rate. According to Hodge et al,43 the open quotient measure of vocal folds during vibration was found to be high for the elderly people at different loud conditions. Alpha ratio If glottis is no longer completely closed during the vibratory cycle or if sound is produced with lower SPL, it might affect the spectrum radiated from the lip openings, too. In our study, balance between low and high regions were measured by alpha ratio. LTAS of hypofunctional and breathy voices were characterized by high L0 and steeper fall above the first formant region, that is, low amplitude in the upper formant region. LTAS of hyperfunctional and loud voices, in turn, were characterized by high levels of F1 and a less steep fall above this region, that is high amplitude in the upper formant region.20,31,32,37 Regarding alpha ratio, in both levels, the young women presented a more subtle decline of LTAS curve than the elderly ones, and with a variation of intensity, from habitual to loud, both groups presented significant differences with high values of this measure. To Nordenberg and
Sundberg,34 high levels of loudness strengthen highfrequency region of LTAS. Linville and Rens’ results, however, showed a lower spectral tilt values for the ratio of energy between 0–1.6 and 1.6–5 kHz for the young female group.26 By analyzing amplitude measurements obtained at regular intervals of 160 Hz in LTAS at habitual level, significant differences between groups were found in the regions of 480–4640 and 6720–8000 Hz, with weaker amplitude values for the elderly group. In loud level, amplitude values were also weaker for the elderly group and comprised ranges between 480–4800 and 6560–8000 Hz (Figures 1–4). There were no differences regarding the region of 0–320 Hz, which is the region of F0, but the elderly women presented high values for this variable. This finding is in agreement with Linville and Rens’26 results. Correlations between acoustic data For the elderly group, in habitual level, there were significant positive correlations between Leq and L1 L0, Leq and alpha ratio, and L1 L0 and alpha ratio, and there was a positive correlation between these variables and the frequencies ranging from 320 to 4460 Hz. The harmonic energy in this frequency range, to some extent, appears to depend on Leq and L1 L0, which may explain the alpha ratio value for the elderly group. In loud level, there was a significant positive correlation between Leq and frequencies ranging from 480 to 4160 Hz. Frequency levels measured along this frequency range appear to depend mainly on Leq, which may explain the alpha ratio value. At the end of the spectrum, instead, there was negative correlation between L1 L0 and frequencies ranging from 5120 to 7040 Hz. Hence, high L1 L0 values, meaning strong amplitude of L1 and weak amplitude of L0 region, correspond to weaker amplitude of these frequencies, or, if glottic resistance decreases with SPL or mode of phonation, the amplitude value of this region decreased too. In some studies, this increase in energy spectrum of elderly women voices was related to breathy voice quality and observed at different regions, namely, at 5– 6, 6, or 6.5–10 kHz.21,24,27 In Lo¨fqvist’s21 study, this finding was closely related to the presence of glottic noise in hypofunctional voices. Linville27 found higher amplitude values at 6 kHz in the elderly, people which was assigned to increased vocal
Abbreviation: NS, not significant.
Alpha/frequency points
160 and 320; 5120– 7040 Hz (P < 0.005) 320 Hz; 800–4640 Hz (P < 0.005) 160 Hz; 320–4460 Hz (P < 0.005) 480–4640 Hz; 7200–7680 Hz (P < 0.005) L1 L0/frequency points
480–2080 Hz (P < 0.005) 2720–4320 Hz (P < 0.005) 320 Hz (P ¼ 0.005) SFF/frequency points
NS r ¼ 0.389 (P ¼ 0.034) r ¼ 0.505 (P ¼ 0.004) NS NS r ¼ 0.617 (P ¼ 0.000) NS r ¼ 0.398 (P ¼ 032) r ¼ 0.560 (P ¼ 0.002) NS NS r ¼ 0.574 (P ¼ 0.001) NS 480–4160 Hz (P < 0.005) r ¼ 0.441 (P ¼ 0.015) NS NS r ¼ 0.432 (P ¼ 0.017) r ¼ 0.642 (P ¼ 0.000) r ¼ 0.546 (P ¼ 0.002) 480–3680 Hz (P < 0.005) Leq/SFF SFF/alpha SFF/L1 L0 Leq/L1 L0 Leq/alpha L1 L0/alpha Leq/frequency points
Habitual Loud Habitual
Elderly
TABLE 3. Pearson’s Correlation and P Values for Acoustic Variables in Habitual and Loud Levels
160 Hz; 320–3680 Hz (P < 0.005) 480–4800 Hz; 5760–8000 Hz (P < 0.005)
NS r ¼ 0.650 (P ¼ 0.000) r ¼ 0.537 (P ¼ 0.002) NS NS r ¼ 0.630 (P ¼ 0.000) 320 Hz; 6080–8000 Hz (P < 0.005) 160 and 320 Hz; 3200– 3520 Hz (P < 0.005) 160 and 320 Hz; 3200– 3520 Hz (P < 0.005) 160 and 320 Hz; 800– 7520 Hz (P < 0.005)
Acoustic and LTAS Measures to Detect Vocal Aging
Young
Loud
Paula Torres da Silva, et al
417
fold opening quotient because of the presence of glottic chink, and perceptual breathiness. However, looking at these results, it is difficult to say if, on the basis of LTAS, this high concentration of energy in the upper region of the spectrum is a harmonic or nonharmonic spectral component.32,34 Furthermore, as we have no reference about SPL values in these studies, it is difficult to compare the results. The Leq values for the elderly group in the present research were 70.62 dB (SD ¼ 4.06 dB) and 76.21 dB (SD ¼ 4.52 dB), respectively, in habitual and loud levels, and our findings suggest a relationship between glottal resistance and amplitude of harmonics rather than glottal noise. For young in habitual level, there were moderate correlations between L1 L0 and alpha ratio, L1 L0 and the frequencies ranging from 320 to 3680 Hz, and alpha ratio and the region between 320 Hz and 4800 Hz. This frequency range, whose amplitude depends primarily on L1 L0, responded to some extent for the alpha ratio value found for this group at 71.96 dB (SD ¼ 6.47 dB). High amplitude values in LTAS were observed in the third formant region (F3) in both levels, for the young. In loud level, at 80.13 dB (SD ¼ 5.19 dB), there was also a significant positive correlation between L1 L0 and frequencies ranging from 3200 to 3520 Hz, and a positive correlation between Leq and the amplitude levels of frequencies ranging from 6080 to 8000 Hz (P < 0.005). Thus, if glottic resistance increases with SPL or mode of phonation, the amplitude value of these regions also increases. This region was found to be correlated with breathy voice quality because of the presence of glottic chink in some studies.27,38 But again, we are not sure about the SPL level at which these measures were obtained; hence, it is difficult to compare the results. In the elderly group, there was another correlation between the amplitudes ranging from 2720 to 4320 Hz and SFF, Leq, and alpha ratio. Therefore, as Leq increases, SFF gets higher, and the amplitude in this region of the spectrum (F3) increases too. This, once again, reinforces the hypothesis that this increased amplitude around F3 would not be related to glottal noise or air escape during phonation. In this study, among the LTAS parameters, L1 L0 and alpha ratio seem to be the most reliable measures to indicate agerelated changes of the voice. Furthermore, the acoustic measures SFF and Leq can be used to differentiate young from elderly women voices. All these measures are related to the glottic source. The observed differences between groups regarding alpha ratio and L1 L0 measure, and even Leq and SFF, may be attributable, to some extent, to lower subglottal pressure or a glottal setting providing a slower glottal-closing speed for the elderly group.
CONCLUSION The purpose of this study was to explore the utility of the acoustic and LTAS measures to discriminate between voice quality of elderly and young women, speaking at habitual and loud levels. The sample size was calculated to have the power to detect differences using representative samples of Portuguese speakers of elderly people living in the community in a large urban
418 center and young students of the university. The data found can only be generalized to a similar population. The long-term goal, regarding future research, is to provide a reliable measure of whether observed age-associated speech phonatory changes depart from an average, suggesting a pathologic event. Thus, among LTAS parameters, L1 L0 and alpha ratio would be the most reliable measure to indicate age-related changes of the voice in habitual and loud levels. Furthermore, the acoustic measures, SFF and Leq, can be used to differentiate the young from the elderly women voices. The observed differences between groups regarding alpha ratio and L1 L0 measure, and even Leq and SFF, may be attributable, to some extent, to lower subglottal pressure or a glottal setting providing a slower glottal closing speed for the elderly group. Because the elderly population is continuously growing and will certainly comprise a significant segment in economy, society, and culture, it is important to produce scientific data to understand normal aging and ensure quality of life in old age. Further studies are needed to evaluate whether these acoustic variables would help differentiate age-associated speech phonatory changes from pathologic events affecting the elderly subjects’ voices. REFERENCES 1. Boone DR, Bayles KA, Koopmann CFJR. Communicative aspects of aging. Otolaryngol Clin North Am. 1982;15:313–327. 2. Pontes P, Brasolotto A, Behlau M. Glottic characteristics and voice complaint in the elderly. J Voice. 2005;19:84–94. 3. Sato K, Hirano M. Age-related changes of elastic fibers in the superficial layer of the lamina propria of vocal folds. Ann Otol Rhinol Laryngol. 1997;106:44–48. 4. Honjo I, Isshiki N. Laryngoscopic and voice characteristics of aged persons. Arch Otolaryngol. 1980;106:149–150. 5. Kahane JC. A survey of age-related changes in the connective tissues of the human adult larynx. In: Bless DM, Abbs JH, eds. Vocal Fold Physiology: Contemporary Research and Clinical Issues. San Diego, CA: College Hill Press; 1981:44–49. 6. Hirano M, Kurita S, Nakashima T. Growth, development and aging of human vocal folds. In: Bless D, ed. Vocal Fold Physiology: Contemporary Research and Clinical Issues. San Diego CA: College Hill Press; 1983: 22–43. 7. Mueller PB, Sweeney RJ, Baribeau LJ. Acoustic and morphologic study of the senescent voice. Ear Nose Throat J. 1984;63:292–295. 8. Kahane JC. Connective tissue changes in the larynx and their effects on voice. J Voice. 1987;1:27–30. 9. Hirano M, Kurita S, Sakaguchi S. Ageing of the vibratory of human vocal folds. Acta Otolaryngol (Stockh). 1989;107:428–433. 10. Close LG, Woodson GE. Common upper airway disorders in the elderly and their management. Geriatrics. 1989;44:67–71. 11. Linville SE. The sound of senescence. J Voice. 1996;10:190–200. 12. Sulter AM, Schutte HK, Miller DG. Standardized laryngeal videostroboscopic rating: differences between untrained and trained male and female subjects, and effects of varying sound intensity, fundamental frequency and age. J Voice. 1996;10:175–189. 13. Bloch I, Behrman A. Quantitative analysis of videostroboscopic images in presbylarynges. Laryngoscope. 2001;111:2022–2027. 14. Polido AM, Martins MASUR, Hanayama EH. Perception of aging voice. Rev CEFAC. 2005;7:241–251. 15. Ptacek P, Sander E. Age recognition from voice. J Speech Hear Res. 1966; 9:273–277. 16. Behlau M. Presbifonia: envelhecimento vocal inerente a` idade (Presbiphonia: vocal aging related to age). In: Russo ICP, ed. Intervenc¸a˜o Fonoaudiolo´gica na Terceira Idade (Speech Therapy intervation in elderly people). Rio de Janeiro, Brazil: Revinter; 1999:25–50.
Journal of Voice, Vol. 25, No. 4, 2011 17. Camargo ZA, Vilarim GS, Cukier S. Voice quality in elderly: long-term acoustic parameters of male and female voices. Rev CEFAC. 2004;6: 189–196. 18. Frohlich M, Michaelis D, Strube HW. Acoustic ‘‘breathiness measures’’ in the description of pathologic voices. In: Proceedings of International Conference on Acoustics. Seattle, WA; 1998;2:937–940. 19. So¨dersten M, Lindestad P. Glottal closure and perceived breathiness during phonation in normally speaking subjects. J Speech Hear Res. 1990;33: 601–611. 20. Frokjaer-Jensen B, Prytz S. Registration of voice quality. Bruel Kjaer Technol Rev. 1976;3:3–17. 21. Lo¨fqvist A. The long time average spectrum as a tool in voice research. J Phon. 1986;14:471–475. 22. Kitzing P. LTAS criteria pertinent to the measurement of voice quality. J Phon. 1986;14:477–482. 23. Master S, Biased NG, Chiari BM, Pedrosa V. The long-term average spectrum in research and in the clinical practice of speech therapists. Pro´-Fono R Atual Cient. 2006;18:111–120. 24. Mendoza E, Valencia N, Munoz J, Trujillo H. Differences in voice quality between men and women: use of long-term average spectrum (LTAS). J Voice. 1996;10:59–66. 25. White P. Long-term average spectrum analysis of sex- and genderrelated differences in children’s voice. Logo Phon Vocol. 2001;26: 97–101. 26. Linville SE, Rens J. Vocal tract resonance analysis of aging voice using long-term average spectra. J Voice. 2001;15:323–330. 27. Linville SE. Source characteristics of aged voice assessed from long-term average spectra. J Voice. 2002;16:472–479. 28. Cleveland TF, Sundberg J, Stone RE. Long-term average spectrum characteristics of country singers during speaking and singing. J Voice. 2001;1: 54–60. 29. Barrichelo VO, Heuer JR, Dean CM, Sataloff RT. Comparison of singer’s formant, speaker’s ring, and LTAS among classical singers and untrained speakers. J Voice. 2001;3:344–350. 30. Laukkanen AM, Sundberg J, Bjo¨rkner E. Acoustic study of the ‘‘throaty’’ voice quality. TMH-QPSR KTH. 2004;46:14–24. 31. Master S, Biase NG, Chiari BM, Laukkanen AM. Acoustic and perceptual analyses of Brazilian male actors’ and nonactors’ voices: longterm average spectrum and the ‘‘actor’s formant’’. J Voice. 2008;22: 146–154. 32. Hammarberg B, Fritzell B, Gauffin J, Sundberg J, Wedin L. Perceptual and acoustic correlates of abnormal voice qualities. Acta Otolaryngol. 1980;90: 441–451. 33. Gauffin J, Sundberg J. Spectral correlates of glottal voice source waveform characteristics. J Speech Hear Res. 1989;32:556–565. 34. Nordenberg M, Sundberg J. Effects on LTAS of vocal loudness variation. Logoped Phoniatr Vocol. 2004;29:183–191. 35. Sundberg J, Nordemberg M. Effects of vocal loudness variation on spectrum balance as reflected by the alpha measure of long-term-average spectra of speech. J Acoust Soc Am. 2006;120:453–457. 36. Sundberg J. The Science of the Singing Voice. Decal, IL: Northern Illinois University Press; 1987. 37. Bele I. The speaker’s formant. J Voice. 2006;20:555–578. 38. Klatt D, Klatt L. Analysis, synthesis, and perception of voice quality variations among female and male talkers. J Acoust Soc Am. 1990;87: 820–857. 39. Xue SA, Hao GJ. Changes in the human vocal tract due to aging and the acoustic correlates of speech production: a pilot study. J Speech Lang Hear Res. 2003;46:689–701. 40. Colton RH, Casper JK. Understanding Voice Problems: A Physiological Perspective for Diagnosis and Treatment. 2nd ed. Baltimore, MD: Lippincott Williams & Wilkins; 1996. 41. Morrison MD, Gore-Hickman P. Voice disorders in the elderly. J Otolaryngol. 1986;15:231–234. 42. Morris RJ, Brown WS Jr. Age-related differences in speech intensity among adult females. Folia Phoniatr Logop. 1994;46:64–69. 43. Hodge FS, Colton RH, Kelley RT. Vocal intensity characteristics in normal and elderly speakers. J Voice. 2001;15:503–511.
Paula Torres da Silva, et al
Acoustic and LTAS Measures to Detect Vocal Aging
44. Baker KK, Ramig LO, Sapir S, Luschei ES, Smith ME. Control of vocal loudness in young and old adults. J Speech Lang Hear Res. 2001;44:297–305. 45. Titze IR, Sundberg J. Vocal intensity in speakers and singers. J Acoust Soc Am. 1992;91:2936–2946. 46. Ramos LR, Neto JN, Cendoroglo MS, et al. Two-year follow-up study of elderly residents in Sa˜o Paulo, Brazil: methodology and preliminary results. J Public Health. 1998;32:397–407.
419
47. Boersma P, Weenink D. Praat Manual: doing phonetics by computer (V. 5.0.23). [Computer program]. Available at: http://www.praat.org/. Last retrieved September 3, 2008. 48. Boersma B, Kovacic G. Spectral characteristics of three styles of Croatian folk singing. J Acoust Soc Am. 2006;119:1805–1816. 49. SPSS Incorporation. SPSS for Windows. Statistical Package for the Social Sciences (SPSS) for Windows 13. SPSS Coop., Chicago, IL, 2004.
Accepted for publication April 6, 2010. From the *Department of Preventive Medicine, Universidade Federal de Sa˜o Paulo, Sa˜o Paulo, Brazil; yDepartment of Scenic Arts, Arts Institute, Universidade Estadual de Sa˜o Paulo, Sa˜o Paulo, Brazil; and the zDepartment of Otorrinolaryngology and Head and Neck Surgery, Universidade Federal de Sa˜o Paulo, Sa˜o Paulo, Brazil. Address correspondence and reprint requests to Paula Torres da Silva, Department of Preventive Medicine, Universidade Federal de Sa˜o Paulo, Rua Borges Lagoa, 1341— 2nd floor, Sa˜o Paulo, SP 04038-034, Brazil. E-mail: [email protected] Journal of Voice, Vol. 25, No. 4, pp. 411-419 0892-1997/$36.00 Ó 2011 The Voice Foundation doi:10.1016/j.jvoice.2010.04.002
Although vocal quality has been evaluated by different methods of perceptual analysis, they present a considerable degree of subjectivity. Computerized acoustic analysis, on the other hand, provides objective data for describing voice quality. The quantification of acoustic parameters of voice makes it easier to compare different voices for research purposes.20–23 Among the many possibilities of sound spectrum analysis, the long-term average spectrum (LTAS) has been successfully used to evaluate voice quality. This acoustic measurement provides information about the contribution of the source and the filter to the voice quality, pointing the differences between gender,24 age,25–27 profession,28–31 and dysphonic voices. 20–22,32 Several measures of the LTAS were successfully used to discriminate voice qualities, especially the ones that are related to aging voice, such as breathiness and perceived hypofunction.21,32 One of these measures, the difference between the levels of the regions of F1 and F0 (L1 L0), provides information on both, the vocal loudness and the glottic source, and can be related to voice quality. Thus, stronger F0 than F1 correlated with hypofunctional voices perceived as breathy or weak, whereas stronger F1 than F0 was correlated with hyperfunctional and loud voices perceived as resonant, tight, or strong.20–22,31,32 Another LTAS measure is the alpha ratio, which is the ratio of energy between 0–1 and 1–5 kHz.20 The level differences between frequency ranges in the LTAS reflect the slope of the source spectrum, which, in turn, is related to the glottal closing speed.33 The shape of the transglottal airflow during the closing phase is a determining factor for the amplitude of the higher harmonics of the voice source, and therefore, if the glottis closes slowly or insufficiently during phonation, the phase of vocal fold closing may present alterations, and thus, the spectral level in the upper formant region will be reduced. That is why the overall slope is also affected by changes of vocal loudness34,35 and by different types of phonations, which in turn, were strongly correlated with voice quality.32,36,37
412 Comparisons between gender using LTAS24 showed lower spectral tilt and greater levels of aspiration noise in the region corresponding to the third formant (F3) among women, which causes a more breathy quality for female voices. This acoustic clue with high amplitude values of F0 and a bandwidth increase of the lower formant frequencies seems to contribute to the perception of breathiness.38 In an LTAS study, to verify age-related changes in the source characteristics of dynamic speech, elderly women in comparison with young ones presented a tendency toward increased frequency levels at 160 Hz, significantly higher spectral amplitude levels at the frequencies of 320 Hz and 6080–6720 Hz, and significantly lower levels at the frequencies of 3040 and 3200 Hz. These events were related to the perception of breathy voice. Young women demonstrated higher spectral amplitude levels in the frequency region of 3 kHz and a tendency toward lower spectral tilt values. Elderly women presented higher amplitude levels in the region of F0 and in the frequency region of more than 6 kHz, which were related, respectively, to increased open quotient and generation of turbulent aspiration noise at the glottal opening.26 LTAS was used to investigate resonance characteristics of dynamic speech in young adulthood and old age.27 Measurements of the first three formants revealed significant lowering of their central frequencies for elderly women. A lowering in all formant frequency peaks in aging voices is consistent with anatomic data, suggesting that the vocal tract also suffers modifications and gets longer with the larynx lowering in the neck.39 Besides LTAS, other acoustic data related to aging voice have been largely investigated. Among them, lower values of the fundamental frequency (F0) was found in women’s voices and higher F0 in men in association with anatomical and functional characteristics of the vocal folds.16,17,40 Furthermore, although young women tend to present higher formant frequencies because of the smaller length of the vocal tract, all formant frequencies were found to be lower in elderly women, as the vocal tract gets longer with the larynx lowering.26,30,39 Changes in male vocal intensity with aging, unlike the female voices, have also been described in a number of studies that found that the elderly men compared with the young men presented decreased sound pressure levels (SPL) as a result of the weakening of the respiratory and laryngeal muscles.41–45 Despite numerous studies of aging voice, most of them neither controlled the SPL nor analyzed the loudness variation in female vocal aging. This study aimed to identify acoustic measures, LTAS in particular, that significantly differ between young and older women, with no vocal complaints, in different levels of vocal effort, controlling for the intensity of the voice.
METHOD Sixty women were included in this study, 30 young and 30 elderly. Young speakers ranged in age from 20 to 35 years (mean ¼ 26.8) and elderly speakers ranged from 60 to 82 years (mean ¼ 69.57). The elderly sample was drawn from an ongoing cohort of elderly women living in the community in a large urban center and randomly selected for the study.46 The young
Journal of Voice, Vol. 25, No. 4, 2011
sample was drawn from the students of the university. They were all standard Portuguese speaking, and none had a history of previous or current vocal alterations, surgery, head- and neck-related treatments, or respiratory diseases. Subjects were recorded during a read-aloud task of a 250word text, in habitual and loud levels. They were trained by the researcher to talk both in normal and strong loudness. In strong loudness, they were told to speak ‘‘as loudly as possible’’ and were prompt to repeat until adequate efforts were achieved. Our interest was to obtained different levels of intensities and not the louder emission possible. The elderly women who were unable to modulate the intensity satisfactorily were excluded from the initial sample according to the researcher’s perceptual evaluation. The duration of each recording was about 200 seconds. A Sony digitalaudio tape (DAT) recorder, model TDC-D8 and a professional unidirectional cardioid dynamic microphone (Shure, model SM58) were used to capture the voice samples. The microphone was positioned 15 cm away from the speakers who remained seated on a stool. SPL was controlled with an SPL meter, placed together, at the same distance of the sound source. A pure tone was generated, measured, and recorded for later comparisons. For the purposes of analyzing the data, we will use equivalent sound level (Leq)35 as the measure of vocal loudness, because it gives us an average over a long time window while SPL is computed over a short time window, and instead of F0, we will use speaking fundamental frequency (SFF),11 because it means that the values were obtained in long speech samples. Following the procedure outlined by other authors24,27 and, thus, facilitating comparisons between studies, these parameters of the LTAS were selected. In the LTAS window, the acoustical variables in this study were the level difference between the F1 and F0 regions (L1 L0), that is, the level difference between 300–800 Hz and 50–300 Hz, which provides information on the mode of phonation; and the alpha ratio, that is, the level difference between 0–50 Hz and 1– 5 kHz, which provides information on the spectral slope declination and the amplitudes values obtained at intervals of 160 Hz throughout the frequency range of 0–8 kHz, totalizing 50 measurements per emission. The LTAS spectra for each subject and the average spectrum for each group, as well as the other acoustic measures were all obtained automatically by Praat program (version 5.0.23).47 The frequency range used in the acoustic analysis was 0–8 kHz, with Hanning window with a time resolution of 40 milliseconds and bandwidth of 160 Hz. Only the central 40 seconds of each sample were analyzed, which is sufficient to produce an LTAS that is independent of the text.21,22 To perform the LTAS, unvoiced sounds and pauses were automatically eliminated from the registers by the program. To facilitate measurements, the spectra were normalized.48 The statistical analyses were performed taking a multivariate approach with respect to the multiple possibly correlated dependent variables (Leq, alpha, SFF, and L1 L0). This doubly multivariate design was analyzed using a multivariate profile
413
13.69 11.93 10.05 8.83 3.61 9.98 2.34 20.25 15.28 13.85 12.09 8.68 13.95 2.29 18.65 15.88 14.15 12.81 9.34 14.28 2.34 21.20 20.16 17.73 16.01 12.07 17.67 2.68 0.14 3.04 4.05 5.00 8.77 4.10 2.03 1.19 0.85 1.82 4.25 6.57 2.35 2.06 6.31 2.01 0.24 1.46 5.96 0.05 2.84 5.91 2.99 1.54 0.06 2.15 1.60 2.17 178.18 214.70 233.34 258.61 361.75 238.75 33.46 183.23 197.98 215.55 235.54 283.75 221.30 26.71 159.92 184.29 194.95 219.96 255.38 202.50 25.19 162.76 178.73 187.53 196.07 238.61 188.84 17.88 72.11 76.37 78.84 84.77 89.77 80.13 5.19 65.98 73.54 76.63 79.41 84.78 76.20 4.44 51.09 68.28 71.55 76.69 84.24 71.96 6.47 62.01 67.82 70.59 72.73 78.68 70.63 4.06 Elderly (n ¼ 30)
Minimum 1st quartile Median 3rd quartile Maximum Mean SD Minimum 1st Quartile Median 3rd quartile Maximum Mean SD Young (n ¼ 30)
Alpha Habitual (dB) L1 L0 Loud (dB) L1 L0 Habitual (dB) SFF Loud (Hz) SFF Habitual (Hz) Leq Loud (dB) Leq Habitual (dB) Statistical Measures
RESULTS Table 1 shows quartiles and minimum, maximum, mean, and standard deviation (SD) values for acoustic measures in habitual and loud levels by age group. The Leq mean values at habitual level were 70.63 and 71.96 dB and, at loud level, were 76.20 and 80.13 dB, respectively, for the elderly and the young subjects. The variation of intensity, from habitual to loud, was significant for both groups. The SFF mean values at habitual level were 188.84 and 202.50 Hz, and at loud level, 221.30 and 238.75 Hz, respectively, for the elderly and young speakers. With the variation of intensity, from habitual to loud, both groups presented significant differences with higher values for the SFF, and in both intensities, the elderly women showed significantly lower values than the young women. The mean L1 L0 values at habitual level were 1.60 and 0.05 dB, respectively, and at loud level, the values were 2.35 and 4.10 dB, respectively, for the elderly and young women. With the variation of intensity, from habitual to loud, both groups presented significant differences, and in both intensities, elderly women showed significantly lower values than the young ones. The mean alpha ratios at habitual level were 17.67 and 14.28 dB, and at loud level, 13.95 and 9.98 dB, respectively, for the elderly and young subjects. With the variation of intensity, from habitual to loud, both groups presented significant differences with higher values of this measure, and in both intensities, the elderly women showed significantly lower values than the young ones. The multivariate profile analysis showed no interaction between level and age group (the level profiles are parallel or the variables change in a similar way between the groups, P ¼ 0.1948), with significant differences because of age groups (P < 0.0001) and level (P < 0.0001). Table 2 shows the comparisons between age groups and between levels. Regarding the amplitude levels measured every 160 Hz, in habitual level, the young showed significantly higher amplitudes values in the regions of 480–4640 and 6720–8000 Hz (P < 0.005), whereas in loud level, the amplitude values were stronger for the young, in the regions of 480–4800 and
Age Group
model, with between-individual effect being age group (young and elderly) and within-individual effects being level (habitual, loud) and the interaction between level and age group. Under this setting, the interaction effect of age group and level is tested first. If the interaction term turned out to be not statistically significant (ie, the age groups are parallel with respect to level-profile means), the multivariate test on age group was calculated as a pool of the age group effects averaged across levels, and similarly, the multivariate test on levels would be based on the pool of the level effects averaged over age groups. Then, if the age group and/or the level effects are significant, the comparisons among their average levels are reported. Using Pearson’s correlation, the means for the acoustic variables were compared between the two groups in habitual and loud levels regarding age. The computer program Statistical Package for the Social Sciences (SPSS)49 was used to perform the statistical analysis with a significance level of 5%.
Alpha Loud (dB)
Acoustic and LTAS Measures to Detect Vocal Aging
TABLE 1. Quartiles and Minimum, Maximum, Mean, and SD Values for Acoustic Measures in Habitual and Loud Levels by Age Group
Paula Torres da Silva, et al
414
Journal of Voice, Vol. 25, No. 4, 2011
TABLE 2. Summary of the Pairwise Averaged Comparisons Between Age Groups and Between Levels With Respect to the Acoustic Variables
Comparison Between age groups Elderly and young
Between levels Loud and habitual
95% Confidence Interval for Difference*
Measure
Averaged Mean Difference
Standard Error
P*
Lower Bound
Upper Bound
Leq SFF L1 L0 Alpha
2.63 15.56 1.70 3.68
1.21 6.23 0.55 0.56
0.0337 0.0154 0.0033 <0.0001
5.06 28.04 2.81 4.81
0.21 3.08 0.59 2.55
Leq SFF L1 L0 Alpha
6.87 34.35 4.00 4.01
0.53 2.76 0.21 0.27
<0.0001 <0.0001 <0.0001 <0.0001
5.80 28.84 3.57 3.47
7.93 39.87 4.43 4.55
Based on estimated marginal means. * Adjustment for multiple comparisons: least significant difference (equivalent to no adjustments).
6560–8000 Hz (P < 0.005). Figures 1 to 4 show the mean LTAS for both groups in habitual and loud levels. Table 3 shows the correlations between the acoustic variables for better understanding of the findings so far. DISCUSSION The purpose of this study was to explore acoustic and LTAS measures able to discriminate between recorded voices of elderly and young women, with no voice complaints, reading a text in habitual and loud levels. It is well established in the literature that there is a correlation between vocal changes that result from the natural aging process and anatomical-physiological alterations of the vocal mechanism,2–13 and that such characteristics may be better evaluated in acoustic studies, which would be complementary to perceptual analysis of voice quality.20–27 Equivalent sound level and speaking fundamental frequency Changes in vocal intensity with aging have been widely described in a number of studies, most of them with male voices.42–45 In general, they showed lower levels of SPL
among the aged, because of the weakening of the respiratory and laryngeal muscles, causing smaller lung pressure and smaller peak airflow, with greater open quotient.43 According to Baker et al,44 old individuals generate lower levels of SPL when compared with young individuals but modulate loudness levels in a manner similar to that of the young subjects. In our study, mean Leq values were significantly affected by aging, in both levels, and by modulating from habitual to loud, in both age groups. Vocal intensity is known to vary with subglottic pressure.36,42–45 The increased subglottic pressure occurs along with the action of the laryngeal muscles, which promotes strong adduction and tension of the vocal folds, providing high values of F0.12,36,44 Vocal intensity has been found to increase with increasing fundamental frequency at any given level of subglottic pressure.45 According to the data obtained by Hodge et al43, the SPL differences between elderly and young men could be partially explained by the difference of fundamental frequency. In fact, with the variation of intensity, our results showed that both groups presented significant differences for the SFF, with higher values at loud level. At habitual
FIGURE 1. Mean LTAS for the elderly and young groups in habitual level.
Paula Torres da Silva, et al
415
Acoustic and LTAS Measures to Detect Vocal Aging
FIGURE 2. Mean LTAS for the elderly and young groups in loud level. and loud level, the mean SFF was significantly lower for the elderly group. F0 change with age has been well documented in both men and women. In women, F0 appears to remain fairly constant until menopause when a drop in fundamental frequency occurs, probably as a result of hormonal changes during menopause when the superficial layer of the vocal folds may become swollen.2,6,10,13 Our results are in agreement with the literature. According to Hodge et al43, it is possible that the SPL diferences between elderly and young men could be partially explained by the difference of fundamental frequency. In our study, lower values of SFF for the elderly women may also have contributed to the lower values of Leq. The correlation between these variables for the elderly group, although weak, was significant in contrast to the young group (Table 3). Other factors that may explain these results for the elderly group are the resonance or the articulatory factors. It is known that opening the mouth during loud emissions, in acoustic analysis, provides tuning between F0 and F1, thus reinforcing SPL. It may be possible that our elderly group presents limitation in this movement because of modifications at the oral structures, well described in the literature.16 According to Gauffin and Sundberg, the SPL of speech sounds is highly affected not only by vocal loudness but also by the frequency distance between the strongest spectrum partial and the first formant. Moreover, the intensity level greater than 1000 Hz is strongly influenced by the frequencies of
formants 1 and 2.33 But this subject is beyond the scope of this study. Difference L1 L0 Changes in vocal intensity would also be caused by changes in glottal resistance. Vocal fold bowing and atrophy plus the weakness of the laryngeal muscles with aging may lead to smaller glottal resistance to air flow, thus compromising the increase in subglottic pressure.42–45 The findings of Gauffin and Sundberg33 about the relationships between waveform and spectrum of the pulsating transglottal airflow helped to clarify acoustic aspects involved in this situation. By inverse filtering, the pressure or airflow signal at the mouth during voicing, a reasonable representation of the airflow waveform at the glottis, was obtained. Furthermore, it was found that the peak-topeak amplitude measure of the flow glottogram pulses—or waveform—is closely related to the amplitude of the source spectrum fundamental (L0), which in turn, varies considerably during phonation, presumably depending on the degree of glottal ab/adduction or the mode of phonation. Breathy phonation, in contrast to pressed phonation, shows higher values of L0 as an effect of the glottal leakage.26,30,33,39 In our study, there were significant differences between groups regarding L1 L0, with elderly women presenting lower values of this acoustic measure, even at loud intensity, when Leq
Sound pressure level (dB / Hz)
0
-10 ------ Habitual ____ Loud
-20
-30
-40
-50 0
1000
2000
3000
4000 5000 Frequency (Hz)
6000
7000
8000
FIGURE 3. Mean LTAS for the elderly group in habitual and loud levels.
416
Journal of Voice, Vol. 25, No. 4, 2011 0
Sound pressure level (dB / Hz)
-10
----- Habitual ____ Loud
-20
-30
-40
-50 0
1000
2000
3000
4000 5000 Frequency (Hz)
6000
7000
8000
FIGURE 4. Mean LTAS for the young group in habitual and loud levels. was significantly lower compared with that in young ones. With the variation of intensity, from habitual to loud, both groups presented significant higher values of L1 L0, meaning that the degree of glottal resistance increased significantly with raised loudness.23,31,37 Low levels of the F0 and high level of the F1 was largely related in literature to hypofunctional and breathy voices, whereas stronger F1 than F0 was correlated with hyperfunctional and loud voices perceived as resonant, tight, or strong.32–37 Because of elderly vocal fold configuration, glottis probably closes slowly or insufficiently, and the vibratory cycle may present alterations. In Gauffin and Sundberg’s study, when varying the type of phonation, the waveform of the glottal airflow can be changed within wide limits, and great differences can be found in the length of the closed phase, that is, the closed quotient.33 These changes were observed by measuring maximal flow declination rate. On the other hand, as closing decreases with decreased vocal intensity and breathy phonation, open quotient increases. Amplitude of higher overtones of the source fundamental frequency depends on closing rate. According to Hodge et al,43 the open quotient measure of vocal folds during vibration was found to be high for the elderly people at different loud conditions. Alpha ratio If glottis is no longer completely closed during the vibratory cycle or if sound is produced with lower SPL, it might affect the spectrum radiated from the lip openings, too. In our study, balance between low and high regions were measured by alpha ratio. LTAS of hypofunctional and breathy voices were characterized by high L0 and steeper fall above the first formant region, that is, low amplitude in the upper formant region. LTAS of hyperfunctional and loud voices, in turn, were characterized by high levels of F1 and a less steep fall above this region, that is high amplitude in the upper formant region.20,31,32,37 Regarding alpha ratio, in both levels, the young women presented a more subtle decline of LTAS curve than the elderly ones, and with a variation of intensity, from habitual to loud, both groups presented significant differences with high values of this measure. To Nordenberg and
Sundberg,34 high levels of loudness strengthen highfrequency region of LTAS. Linville and Rens’ results, however, showed a lower spectral tilt values for the ratio of energy between 0–1.6 and 1.6–5 kHz for the young female group.26 By analyzing amplitude measurements obtained at regular intervals of 160 Hz in LTAS at habitual level, significant differences between groups were found in the regions of 480–4640 and 6720–8000 Hz, with weaker amplitude values for the elderly group. In loud level, amplitude values were also weaker for the elderly group and comprised ranges between 480–4800 and 6560–8000 Hz (Figures 1–4). There were no differences regarding the region of 0–320 Hz, which is the region of F0, but the elderly women presented high values for this variable. This finding is in agreement with Linville and Rens’26 results. Correlations between acoustic data For the elderly group, in habitual level, there were significant positive correlations between Leq and L1 L0, Leq and alpha ratio, and L1 L0 and alpha ratio, and there was a positive correlation between these variables and the frequencies ranging from 320 to 4460 Hz. The harmonic energy in this frequency range, to some extent, appears to depend on Leq and L1 L0, which may explain the alpha ratio value for the elderly group. In loud level, there was a significant positive correlation between Leq and frequencies ranging from 480 to 4160 Hz. Frequency levels measured along this frequency range appear to depend mainly on Leq, which may explain the alpha ratio value. At the end of the spectrum, instead, there was negative correlation between L1 L0 and frequencies ranging from 5120 to 7040 Hz. Hence, high L1 L0 values, meaning strong amplitude of L1 and weak amplitude of L0 region, correspond to weaker amplitude of these frequencies, or, if glottic resistance decreases with SPL or mode of phonation, the amplitude value of this region decreased too. In some studies, this increase in energy spectrum of elderly women voices was related to breathy voice quality and observed at different regions, namely, at 5– 6, 6, or 6.5–10 kHz.21,24,27 In Lo¨fqvist’s21 study, this finding was closely related to the presence of glottic noise in hypofunctional voices. Linville27 found higher amplitude values at 6 kHz in the elderly, people which was assigned to increased vocal
Abbreviation: NS, not significant.
Alpha/frequency points
160 and 320; 5120– 7040 Hz (P < 0.005) 320 Hz; 800–4640 Hz (P < 0.005) 160 Hz; 320–4460 Hz (P < 0.005) 480–4640 Hz; 7200–7680 Hz (P < 0.005) L1 L0/frequency points
480–2080 Hz (P < 0.005) 2720–4320 Hz (P < 0.005) 320 Hz (P ¼ 0.005) SFF/frequency points
NS r ¼ 0.389 (P ¼ 0.034) r ¼ 0.505 (P ¼ 0.004) NS NS r ¼ 0.617 (P ¼ 0.000) NS r ¼ 0.398 (P ¼ 032) r ¼ 0.560 (P ¼ 0.002) NS NS r ¼ 0.574 (P ¼ 0.001) NS 480–4160 Hz (P < 0.005) r ¼ 0.441 (P ¼ 0.015) NS NS r ¼ 0.432 (P ¼ 0.017) r ¼ 0.642 (P ¼ 0.000) r ¼ 0.546 (P ¼ 0.002) 480–3680 Hz (P < 0.005) Leq/SFF SFF/alpha SFF/L1 L0 Leq/L1 L0 Leq/alpha L1 L0/alpha Leq/frequency points
Habitual Loud Habitual
Elderly
TABLE 3. Pearson’s Correlation and P Values for Acoustic Variables in Habitual and Loud Levels
160 Hz; 320–3680 Hz (P < 0.005) 480–4800 Hz; 5760–8000 Hz (P < 0.005)
NS r ¼ 0.650 (P ¼ 0.000) r ¼ 0.537 (P ¼ 0.002) NS NS r ¼ 0.630 (P ¼ 0.000) 320 Hz; 6080–8000 Hz (P < 0.005) 160 and 320 Hz; 3200– 3520 Hz (P < 0.005) 160 and 320 Hz; 3200– 3520 Hz (P < 0.005) 160 and 320 Hz; 800– 7520 Hz (P < 0.005)
Acoustic and LTAS Measures to Detect Vocal Aging
Young
Loud
Paula Torres da Silva, et al
417
fold opening quotient because of the presence of glottic chink, and perceptual breathiness. However, looking at these results, it is difficult to say if, on the basis of LTAS, this high concentration of energy in the upper region of the spectrum is a harmonic or nonharmonic spectral component.32,34 Furthermore, as we have no reference about SPL values in these studies, it is difficult to compare the results. The Leq values for the elderly group in the present research were 70.62 dB (SD ¼ 4.06 dB) and 76.21 dB (SD ¼ 4.52 dB), respectively, in habitual and loud levels, and our findings suggest a relationship between glottal resistance and amplitude of harmonics rather than glottal noise. For young in habitual level, there were moderate correlations between L1 L0 and alpha ratio, L1 L0 and the frequencies ranging from 320 to 3680 Hz, and alpha ratio and the region between 320 Hz and 4800 Hz. This frequency range, whose amplitude depends primarily on L1 L0, responded to some extent for the alpha ratio value found for this group at 71.96 dB (SD ¼ 6.47 dB). High amplitude values in LTAS were observed in the third formant region (F3) in both levels, for the young. In loud level, at 80.13 dB (SD ¼ 5.19 dB), there was also a significant positive correlation between L1 L0 and frequencies ranging from 3200 to 3520 Hz, and a positive correlation between Leq and the amplitude levels of frequencies ranging from 6080 to 8000 Hz (P < 0.005). Thus, if glottic resistance increases with SPL or mode of phonation, the amplitude value of these regions also increases. This region was found to be correlated with breathy voice quality because of the presence of glottic chink in some studies.27,38 But again, we are not sure about the SPL level at which these measures were obtained; hence, it is difficult to compare the results. In the elderly group, there was another correlation between the amplitudes ranging from 2720 to 4320 Hz and SFF, Leq, and alpha ratio. Therefore, as Leq increases, SFF gets higher, and the amplitude in this region of the spectrum (F3) increases too. This, once again, reinforces the hypothesis that this increased amplitude around F3 would not be related to glottal noise or air escape during phonation. In this study, among the LTAS parameters, L1 L0 and alpha ratio seem to be the most reliable measures to indicate agerelated changes of the voice. Furthermore, the acoustic measures SFF and Leq can be used to differentiate young from elderly women voices. All these measures are related to the glottic source. The observed differences between groups regarding alpha ratio and L1 L0 measure, and even Leq and SFF, may be attributable, to some extent, to lower subglottal pressure or a glottal setting providing a slower glottal-closing speed for the elderly group.
CONCLUSION The purpose of this study was to explore the utility of the acoustic and LTAS measures to discriminate between voice quality of elderly and young women, speaking at habitual and loud levels. The sample size was calculated to have the power to detect differences using representative samples of Portuguese speakers of elderly people living in the community in a large urban
418 center and young students of the university. The data found can only be generalized to a similar population. The long-term goal, regarding future research, is to provide a reliable measure of whether observed age-associated speech phonatory changes depart from an average, suggesting a pathologic event. Thus, among LTAS parameters, L1 L0 and alpha ratio would be the most reliable measure to indicate age-related changes of the voice in habitual and loud levels. Furthermore, the acoustic measures, SFF and Leq, can be used to differentiate the young from the elderly women voices. The observed differences between groups regarding alpha ratio and L1 L0 measure, and even Leq and SFF, may be attributable, to some extent, to lower subglottal pressure or a glottal setting providing a slower glottal closing speed for the elderly group. Because the elderly population is continuously growing and will certainly comprise a significant segment in economy, society, and culture, it is important to produce scientific data to understand normal aging and ensure quality of life in old age. Further studies are needed to evaluate whether these acoustic variables would help differentiate age-associated speech phonatory changes from pathologic events affecting the elderly subjects’ voices. REFERENCES 1. Boone DR, Bayles KA, Koopmann CFJR. Communicative aspects of aging. Otolaryngol Clin North Am. 1982;15:313–327. 2. Pontes P, Brasolotto A, Behlau M. Glottic characteristics and voice complaint in the elderly. J Voice. 2005;19:84–94. 3. Sato K, Hirano M. Age-related changes of elastic fibers in the superficial layer of the lamina propria of vocal folds. Ann Otol Rhinol Laryngol. 1997;106:44–48. 4. Honjo I, Isshiki N. Laryngoscopic and voice characteristics of aged persons. Arch Otolaryngol. 1980;106:149–150. 5. Kahane JC. A survey of age-related changes in the connective tissues of the human adult larynx. In: Bless DM, Abbs JH, eds. Vocal Fold Physiology: Contemporary Research and Clinical Issues. San Diego, CA: College Hill Press; 1981:44–49. 6. Hirano M, Kurita S, Nakashima T. Growth, development and aging of human vocal folds. In: Bless D, ed. Vocal Fold Physiology: Contemporary Research and Clinical Issues. San Diego CA: College Hill Press; 1983: 22–43. 7. Mueller PB, Sweeney RJ, Baribeau LJ. Acoustic and morphologic study of the senescent voice. Ear Nose Throat J. 1984;63:292–295. 8. Kahane JC. Connective tissue changes in the larynx and their effects on voice. J Voice. 1987;1:27–30. 9. Hirano M, Kurita S, Sakaguchi S. Ageing of the vibratory of human vocal folds. Acta Otolaryngol (Stockh). 1989;107:428–433. 10. Close LG, Woodson GE. Common upper airway disorders in the elderly and their management. Geriatrics. 1989;44:67–71. 11. Linville SE. The sound of senescence. J Voice. 1996;10:190–200. 12. Sulter AM, Schutte HK, Miller DG. Standardized laryngeal videostroboscopic rating: differences between untrained and trained male and female subjects, and effects of varying sound intensity, fundamental frequency and age. J Voice. 1996;10:175–189. 13. Bloch I, Behrman A. Quantitative analysis of videostroboscopic images in presbylarynges. Laryngoscope. 2001;111:2022–2027. 14. Polido AM, Martins MASUR, Hanayama EH. Perception of aging voice. Rev CEFAC. 2005;7:241–251. 15. Ptacek P, Sander E. Age recognition from voice. J Speech Hear Res. 1966; 9:273–277. 16. Behlau M. Presbifonia: envelhecimento vocal inerente a` idade (Presbiphonia: vocal aging related to age). In: Russo ICP, ed. Intervenc¸a˜o Fonoaudiolo´gica na Terceira Idade (Speech Therapy intervation in elderly people). Rio de Janeiro, Brazil: Revinter; 1999:25–50.
Journal of Voice, Vol. 25, No. 4, 2011 17. Camargo ZA, Vilarim GS, Cukier S. Voice quality in elderly: long-term acoustic parameters of male and female voices. Rev CEFAC. 2004;6: 189–196. 18. Frohlich M, Michaelis D, Strube HW. Acoustic ‘‘breathiness measures’’ in the description of pathologic voices. In: Proceedings of International Conference on Acoustics. Seattle, WA; 1998;2:937–940. 19. So¨dersten M, Lindestad P. Glottal closure and perceived breathiness during phonation in normally speaking subjects. J Speech Hear Res. 1990;33: 601–611. 20. Frokjaer-Jensen B, Prytz S. Registration of voice quality. Bruel Kjaer Technol Rev. 1976;3:3–17. 21. Lo¨fqvist A. The long time average spectrum as a tool in voice research. J Phon. 1986;14:471–475. 22. Kitzing P. LTAS criteria pertinent to the measurement of voice quality. J Phon. 1986;14:477–482. 23. Master S, Biased NG, Chiari BM, Pedrosa V. The long-term average spectrum in research and in the clinical practice of speech therapists. Pro´-Fono R Atual Cient. 2006;18:111–120. 24. Mendoza E, Valencia N, Munoz J, Trujillo H. Differences in voice quality between men and women: use of long-term average spectrum (LTAS). J Voice. 1996;10:59–66. 25. White P. Long-term average spectrum analysis of sex- and genderrelated differences in children’s voice. Logo Phon Vocol. 2001;26: 97–101. 26. Linville SE, Rens J. Vocal tract resonance analysis of aging voice using long-term average spectra. J Voice. 2001;15:323–330. 27. Linville SE. Source characteristics of aged voice assessed from long-term average spectra. J Voice. 2002;16:472–479. 28. Cleveland TF, Sundberg J, Stone RE. Long-term average spectrum characteristics of country singers during speaking and singing. J Voice. 2001;1: 54–60. 29. Barrichelo VO, Heuer JR, Dean CM, Sataloff RT. Comparison of singer’s formant, speaker’s ring, and LTAS among classical singers and untrained speakers. J Voice. 2001;3:344–350. 30. Laukkanen AM, Sundberg J, Bjo¨rkner E. Acoustic study of the ‘‘throaty’’ voice quality. TMH-QPSR KTH. 2004;46:14–24. 31. Master S, Biase NG, Chiari BM, Laukkanen AM. Acoustic and perceptual analyses of Brazilian male actors’ and nonactors’ voices: longterm average spectrum and the ‘‘actor’s formant’’. J Voice. 2008;22: 146–154. 32. Hammarberg B, Fritzell B, Gauffin J, Sundberg J, Wedin L. Perceptual and acoustic correlates of abnormal voice qualities. Acta Otolaryngol. 1980;90: 441–451. 33. Gauffin J, Sundberg J. Spectral correlates of glottal voice source waveform characteristics. J Speech Hear Res. 1989;32:556–565. 34. Nordenberg M, Sundberg J. Effects on LTAS of vocal loudness variation. Logoped Phoniatr Vocol. 2004;29:183–191. 35. Sundberg J, Nordemberg M. Effects of vocal loudness variation on spectrum balance as reflected by the alpha measure of long-term-average spectra of speech. J Acoust Soc Am. 2006;120:453–457. 36. Sundberg J. The Science of the Singing Voice. Decal, IL: Northern Illinois University Press; 1987. 37. Bele I. The speaker’s formant. J Voice. 2006;20:555–578. 38. Klatt D, Klatt L. Analysis, synthesis, and perception of voice quality variations among female and male talkers. J Acoust Soc Am. 1990;87: 820–857. 39. Xue SA, Hao GJ. Changes in the human vocal tract due to aging and the acoustic correlates of speech production: a pilot study. J Speech Lang Hear Res. 2003;46:689–701. 40. Colton RH, Casper JK. Understanding Voice Problems: A Physiological Perspective for Diagnosis and Treatment. 2nd ed. Baltimore, MD: Lippincott Williams & Wilkins; 1996. 41. Morrison MD, Gore-Hickman P. Voice disorders in the elderly. J Otolaryngol. 1986;15:231–234. 42. Morris RJ, Brown WS Jr. Age-related differences in speech intensity among adult females. Folia Phoniatr Logop. 1994;46:64–69. 43. Hodge FS, Colton RH, Kelley RT. Vocal intensity characteristics in normal and elderly speakers. J Voice. 2001;15:503–511.
Paula Torres da Silva, et al
Acoustic and LTAS Measures to Detect Vocal Aging
44. Baker KK, Ramig LO, Sapir S, Luschei ES, Smith ME. Control of vocal loudness in young and old adults. J Speech Lang Hear Res. 2001;44:297–305. 45. Titze IR, Sundberg J. Vocal intensity in speakers and singers. J Acoust Soc Am. 1992;91:2936–2946. 46. Ramos LR, Neto JN, Cendoroglo MS, et al. Two-year follow-up study of elderly residents in Sa˜o Paulo, Brazil: methodology and preliminary results. J Public Health. 1998;32:397–407.
419
47. Boersma P, Weenink D. Praat Manual: doing phonetics by computer (V. 5.0.23). [Computer program]. Available at: http://www.praat.org/. Last retrieved September 3, 2008. 48. Boersma B, Kovacic G. Spectral characteristics of three styles of Croatian folk singing. J Acoust Soc Am. 2006;119:1805–1816. 49. SPSS Incorporation. SPSS for Windows. Statistical Package for the Social Sciences (SPSS) for Windows 13. SPSS Coop., Chicago, IL, 2004.