Reliable Jitter and Shimmer Measurements In

Home , Jitter

Reliable Jitter and Shimmer Measurements in Voice Clinics: The Relevance of Vowel, Gender, Vocal Intensity, and Fundamental Frequency Effects in a Typical Clinical Task

*Meike Brockmann, †Michael J. Drinnan, ‡Claudio Storck, and §Paul N. Carding, *Zurich, Switzerland, yNewcastle upon Tyne, United Kingdom, zBasel, Switzerland, xNewcastle upon Tyne, United Kingdom

Summary: The aims of this study were to examine vowel and gender effects on jitter and shimmer in a typical clinical voice task while correcting for the confounding effects of voice sound pressure level (SPL) and fundamental frequency (F0). Furthermore the relative effect sizes of vowel, gender, voice SPL, and F0 were assessed, and recommendations for clinical measurements were derived. With this cross-sectional single cohort study, 57 healthy adults (28 women, 29 men) aged 20–40 years were investigated. Three phonations of /a/, /o/, and /i/ at ‘‘normal’’ voice loudness were analyzed using Praat (software). The effects of vowel, gender, voice SPL, and F0 on jitter and shimmer were assessed using descriptive and inferential (analysis of covariance) statistics. The effect sizes were determined with the eta-squared statistic. Vowels, gender, voice SPL, and F0, each had significant effects either on jitter or on shimmer, or both. Voice SPL was the most important factor, whereas vowel, gender, and F0 effects were comparatively small. Because men had systematically higher voice SPL, the gender effects on jitter and shimmer were smaller when correcting for SPL and F0. Surprisingly, in clinical assessments, voice SPL has the single biggest impact on jitter and shimmer. Vowel and gender effects were clinically important, whereas fundamental frequency had a relatively small influence. Phonations at a predefined voice SPL (80 dB minimum) and vowel (/a/) would enhance measurement reliability. Furthermore, gender-specific thresholds applying these guidelines should be established. However, the efficiency of these measures should be verified and tested with patients. Key Words: Vowel–Voice intensity–Gender–Fundamental frequency–Jitter–Shimmer–Voice assessments.

INTRODUCTION Why are jitter and shimmer measured? Measurements of voice frequency (jitter) and amplitude (shim- Small irregularities in the acoustic wave are considered as nor- mer) perturbation commonly form part of a comprehensive mal variation associated with physiologic body function and voice examination. These measures are used to provide supple- voice production.10,11 However, voice perturbation levels mentary information alongside visual laryngeal examination have been shown to considerably increase according to laryn- details and auditory perceptual voice assessment.1,2 Both geal pathology12–14 and to partially discriminate between func- perturbation parameters are obtained by computer analysis of tional voice disorder types.15,16 Investigations combining a prolonged vowel phonations or speech samples and quantify number of acoustic parameters and visual laryngeal assessment unintentional irregularity in the acoustic waves generated by methods suggest that jitter and shimmer may be valuable pre- the larynx. Therefore, voice perturbation analysis has been dictors of voice pathology.16,17 Additionally, correct recogni- described as an easily applicable, indirect, noninvasive mea- tion of pathologic voices has been reported to be as high as surement of laryngeal vibratory function, which can quantify above 95% when perturbation parameters are used in combina- the regularity and hence the stability of vocal fold vibration.3,4 tion with a mathematical pattern recognition model.18 Further- These measures may help describe and quantify pathological as more, investigations in ‘‘normal’’ sounding voices such as in well as healthy vocal fold vibration characteristics. The param- patients with laryngopharyngeal reflux or after thyroid surgery eters jitter and shimmer are widely used in clinical and scientific imply that perturbation parameters might even track subtle settings for diagnostic and descriptive purposes1,5,6 as well as to voice alterations not easily detectable by perceptual or visual document and evaluate voice treatment outcome.7–9 assessment methods.19,20 However, applications of jitter and shimmer analysis have Accepted for publication July 1, 2009. Conflicts of interest: All authors declare that they have no real or potential conflicts of been thwarted by unsatisfactory measurement reliability, sensi- interest influencing the presented research. From the *Speech Pathology Section, Department of Phoniatry and Speech Pathology, tivity, and specificity. This has been particularly true in the Clinic for Otorhinolaryngology, Head and Neck Surgery, University Hospital Zurich, Zur- analysis of ‘‘hoarse’’ voices with potentially severely aperiodic ich, Switzerland; yRegional Medical Physics Department, Newcastle General Hospital, sound signal structure.21–23 Yet even the analysis of mildly dys- Newcastle upon Tyne, United Kingdom; zDepartment of Phoniatry, University Hospital Basel, Basel, Switzerland; and the xDepartment of Speech, Voice and Swallowing, Free- phonic voices (with presumably more periodic sound signals) man Hospital, Newcastle upon Tyne, United Kingdom. 24 Address correspondence and reprint requests to Meike Brockmann, Head of Speech Pa- has been shown to have at best moderate test-retest reliability. thology Section, Department of Phoniatry and Speech Pathology, Clinic for Otorhinolaryn- Furthermore, pathology-specific norm values have not been gology, Head and Neck Surgery, University Hospital Zurich, Frauenklinikstrasse 24, 8091 3,4 Zurich, Switzerland. E-mail: [email protected] established to date. As a result, voice perturbation measure- Journal of Voice, Vol. 25, No. 1, pp. 44-53 ments have been cautiously interpreted. Measurement reliability 0892-1997/$36.00 Ó 2011 The Voice Foundation requires considerable improvement to enhance the potential of doi:10.1016/j.jvoice.2009.07.002 jitter and shimmer analysis in the voice clinic setting. Meike Brockmann, et al Vowel Influence on Jitter and Shimmer 45

Recording and measurement technique effects of a number of influencing factors such as (1) an unequal re- The limited measurement reliability might partially be ex- cording equipment between studies not complying to current plained by a number of technical confounding factors such as guidelines and/or (2) the use of both male and female subjects microphone type and placement, recording distance, analysis in the same experiments29,30 and (3) measurements at different method, and background noise. Also, measurement reliability voice intensities and fundamental frequencies5,26 and/or (4) an decreases with increasing voice irregularity because the method unequal recording distance (Table 1). However, all these factors depends on exact sound pressure level (SPL) and F0 recogni- have to be considered when determining the relative influence tion.5,17,25,26 To minimize confounding effects, guidelines of different vowels on jitter and shimmer measurements. defining an appropriate acoustic assessment equipment and From a theoretical perspective, various hypotheses have been setup and also suitable voice types have been published.5 proposed to explain or reject the possibility of vowel effects on However, from a clinical perspective, also the recording in- voice intensity and fundamental frequency perturbation. Ac- structions and interindividual differences in performing the cording to the source-filter theory, jitter and shimmer should given voice tasks might account for the high measurement var- not be affected by vowel articulation influencing the overtone iability. In clinical assessments, patients are usually instructed to spectrum through alterations of the vocal tract shape (filter) sustain the vowels /a/, /u/, /o/, or /i/ ‘‘atcomfortable loudness and because this does not alter the signal of the sound source pitch.’’1 This voice task is used to avoid potentially confounding (larynx).31 This is supported by electroglottographic (EGG) intensity and pitch change effects on acoustic measurements as measurements showing no distinct vowel effects.29 However, well as a possible influence of articulatory movements.1,5,27 the physical linkage hypothesis proposes an interrelation be- However, it is unclear whether these instructions are sufficiently tween vocal tract and laryngeal movements. Investigations in rigorous to produce maximum reliability in clinical practice. the position of the hyoid-larynx complex32 or the head and tongue33 showed associations with changes in fundamental Vowel influence frequency and also the amount of jitter and shimmer. Vowel effects in measurements of jitter and shimmer have been investigated by a number of researchers and are summarized in Gender differences Table 1. The results range from the description of distinct vowel Based on the results in Table 1, it might be assumed that phys- differences28 to no vowel effects at all.29 Highest jitter has been iologic gender differences were the strongest confounding fac- found in /u/, /i/, or /a/ and lowest shimmer in /i/ or /u/ (Table 1). tor in current investigations regarding vowel effects. However, These contradictory results clearly show that the influence of the evidence regarding gender differences also is contradictory vowel articulation has not been fully understood to date. in the published literature hindering a clear attribution. From a methodological standpoint, the reported results appear Comparing two similar designed studies examining 31 men not comparable or transferable into clinical practice because and 20 women,34,35 the authors concluded that female voices

TABLE 1. Evidence Regarding Vowel Differences Highest Lowest Highest Lowest Reference Jitter Jitter Shimmer Shimmer Participants Recording Conditions Kiliçetal28 /u/ and /y/ /a/, /e/, /u/, /y/ 26 men ‘‘Comfortable loudness and pitch’’; recorded at 15 cm distance Dwire & /u/ 25 women/ ‘‘Comfortable loudness and pitch’’; McCauley38 24 men patient held microphone; recorded at 2.5–3.5 cm distance 48 Gelfer /i/ and /a/ equal /a/ /i/ 29 women Speaking F0 and speaking F0 +1 octave; 60, 70, and 80 dB; recorded at 12 inches distance Sussmann & /a/ and /i/ /u/ 10 women/ ‘‘Comfortable loudness and pitch’’; Sapienza40 10 men recorded at 15 cm distance Sorensen & /i/ /a/ /a/ /u/ 20 women ‘‘Comfortable F0’’; between 70 and Horii35 80 dB; recorded at 15 cm distance 34 Horii /u/ 20 young/ ‘‘Comfortable F0’’; between 70 and 20 aged men 80 dB; recorded at 15 cm distance Milenkovic30 /a/ /u/ and /i/ /a/ /u/ 3 women/ ‘‘Comfortable loudness and pitch’’; 3 men recorded at unspecified distance Orlikoff29 No vowel effects 10 women/ Women: 220 Hz, men 110 Hz ± 0.5 10 men semitones; 74 ± 4 dB; recorded at 30 cm distance Notes: Research in vowel differences summarized according to highest and lowest jitter and shimmer values. Results are clearly contradictory. This might be partly because of the different recording conditions. 46 Journal of Voice, Vol. 25, No. 1, 2011 normally display less shimmer but more jitter than male voi- through acoustic assessments and the clinical usefulness of these ces.35 This was supported by a number of later works.36–38 How- measurements. ever, also smaller absolute jitter values were observed in women compared with men.39,40 In contrast to this, similar jitter was Study aims found when comparing 6 women and 6 men under control of Theaims ofthis study were (1) to examine the effects of vowel and a 41 voice SPL and F0. gender on jitter and shimmer in a typical clinical voice task while Again the reported results are difficult to compare because of correcting for the confounding effects of vocal intensity and small numbers of participants, phonations at unequal voice SPL fundamental frequency and (2) to assess the relative effect sizes and F0 levels, or the use of different recording and analysis of the four factors, vowel, gender, voice SPL, and F0, and thereby tools. Furthermore, gender effects appear to be strongly linked recommend suitable protocols for clinical voice assessments. 41,42 to the influences of the individuals voice SPL and F0. METHODS Voice intensity effects Inclusion and exclusion criteria Research in healthy and dysphonic voices has shown a decrease Seventy volunteers from the University Hospital Zurich, 35 of jitter and shimmer with increasing vocal intensity irrespec- women and 35 men between 20 and 40 years of age, were tive of voice task.42–45 Furthermore, in an earlier work, we recruited for this study. The native language (usually Swiss were able to show that jitter and shimmer dramatically increase German or German) and smoking habits of all participants below a critical threshold of 80 dB, when adults are asked to were noted. Participants were excluded if they met one or phonate at ‘‘soft,’’‘‘normal,’’and ‘‘loud’’ individual vocal inten- more of the following criteria: sity.42 It is clear that when healthy adults are asked to phonate ‘‘at comfortable loudness and pitch,’’ voice SPL varies consid- 1. had a hoarse voice on the day of recording; erably between individuals.46 Across several speaking condi- 2. reported recent voice problems or a voice disorder history; tions, a range of 63–75 dB between healthy adults should be 3. had any previous formal voice training or voice therapy; expected, whereas ‘‘comfortable’’ voice intensity of both 4. were taking medication or had a condition that might af- women and men lies below the above-described critical thresh- fect normal voice function; old of 80 dB.42,46 Also women tend to phonate systematically 5. were intubated recently for any surgical intervention; softer in the same voice task, which might explain some of 6. had undergone surgery in the torso, head, and neck region the jitter and shimmer differences reported between women in the last 18 months; and men.35,40 In addition to this, if different vowels are natu- 7. were unable to phonate 5 seconds of /a/, /o/, or /i/ after 10 rally phonated at different vocal intensity, then voice SPL minutes of training. would become a major separate confounding factor. Furthermore, three recorded phonations of /a/ were rated by two independent experts (one speech-language pathologist, Fundamental frequency influence one phoniatrician) using the grade, roughness, breathiness, Based on studies investigating gender effects, it was often con- aesthenia, strain (GRBAS) scale, an auditory-perceptual voice 37,47 49 cluded that higher F0 is associated with higher jitter. How- analysis method. Voice recordings were excluded from objec- ever, investigations in women and men changing their F0 show tive acoustic analysis if the mean of the two expert ratings was 1 that jitter and shimmer are highest in lower frequencies.41,48 or higher for any GRBAS characteristic. Furthermore, lowering F0 led to a more intense decrease of jitter and shimmer in men than in women.41 Because few studies Subjects studied 3,4 have investigated F0 effects, Baken and Orlikoff concluded Of the original 70 volunteers, 13 were excluded: 12 people with that the influence of F0 on jitter and shimmer has not been fully a mean GRBAS scale score of >1 in one voice characteristic and understood to date. However, based on the current evidence, we 1 participant was unable to phonate for 5 seconds. In total, have to expect that an influence of F0 might underlie reported therefore, 57 participants were included in the study: 28 women vowel and gender effects. (mean age: 28;8 years) and 29 men (mean age: 28;11 years). Of the 57 participants, 9 were smokers and 12 no native speakers.

Reliable measures in the voice clinic Voice recordings In summary, it is clear that the effects of vowels, gender, voice During a training phase of 10 minutes maximum, participants SPL, and F0 have not been satisfactorily investigated to date. were asked to ‘‘sustain /a/ for 5 seconds at habitual pitch and This is especially true in a clinical voice task where the com- loudness.’’ This procedure was repeated with the vowels /o/ bined effects of these variables are not controlled. It is reason- and /i/. When they were able to comply comfortably, the record- able to assume that this contributes to the unsatisfactory ings were made in a randomized order. Each participant made 9 22,24 measurement reliability in voice clinics to date. This clearly individual phonations (3 of each vowel: /a/, /o/, /i/), which gave reduces the diagnostic and descriptive information obtained a total of 513 vowel phonations. aVoice SPL refers to ‘‘vocal intensity’’ measured in dB SPL. In this work Recordings were made according to European and American 1,2 voice SPL was determined using comparison method. assessment guidelines. All participants were recorded in Meike Brockmann, et al Vowel Influence on Jitter and Shimmer 47 a soundproof room using a head-mounted off-axis positioned mi- We also assessed the first-order interactions between the crophone (C444; AKG Acoustics GmbH, Vienna, Austria) with factors; this would indicate, for example, whether vowel effects 10-cm microphone-mouth distance and a portable DAT-Recorder are different in women and men. (TCD-D8; Sony, Tokyo, Japan) at a sampling rate of 48 000 Hz Subjects (1–57) were included as a random factor within gen- with 16 bits per sample. The recording system was calibrated der. That is to say, we assume the male and female subjects are using the comparison method.50 Calibration was performed be- simply random samples of the population being studied. The fore voice recording using speech-weighted noise,51 recorded three repeats of each vowel, which under ideal conditions with 10 cm distance to sound source at 50, 65, 85, and 90 dB SPL. would give indentical jitter (shimmer) values, allow us to estimate the variability due to measurement error. Editing of voice recordings Finally, we estimated the effect size of each factor using the Each phonation was edited into an individual file and labeled eta-squared statistic. Eta-squared is simply the proportion of the anonymously using Audacity 1.2.4b (Free Software Foundation overall variance in jitter (or shimmer) that can be attributed to Europe, Hamburg, Germany).52 Acoustic analysis was conducted each of the factors. Each factor has its own eta-squared value, with Praat (Paul Boersma, University of Amsterdam, The and the sum of all the eta-squared values (including the mea- Netherlands),53 using seconds 0.5 to 3.5 from voice onset to surement error) is exactly 1. Therefore, eta-squared gives exclude the known variability of the voice onset and offset phase. a pragmatic indication of which factors affect jitter (or shimmer) the most and therefore are the most important. Main outcome measures The main outcome measures were the Praat parameters ‘‘local jitter’’ (in %) and ‘‘shimmer dB’’ (in dB). These measure rela- RESULTS tive jitter and shimmer, which is normalized for an individual’s Description of jitter and shimmer measurements 3,4 F0 and voice SPL. In Figure 1, the effects of voice SPL, gender, and vowel on jitter and shimmer measurements are shown. Even without resort to Statistical analysis statistics, it is clear that voice SPL has a dramatic confounding First, the relationships of jitter and shimmer with voice SPL effect on jitter and particularly shimmer. The effects of gender were plotted. From these graphs and from a residuals plot during and vowel are not so apparent. our preliminary modeling work, it was clear that jitter and shim- Table 2 gives descriptive statistics for the acoustic variables. mer both showed a log-normal distribution and so a logarithmic In every case, the men phonated systematically louder than the transform was applied to each before further statistical analysis. women with lower fundamental frequency. Equally, the men al- Thereafter, we used an analysis of covariance (ANCOVA) ways had lower jitter and shimmer values (more periodic voice model, which is an extension of linear regression. In linear signals). It is also clear from Figure 1 that women are system- regression, jitter (or shimmer) is modeled as the sum of effects atically quieter in their phonation than men and naturally have because of continuous explanatory factors. In this work, these substantially different fundamental frequency. We hypothesize were voice SPL and F0. ANCOVA allows us to add in the fixed that these differences can explain some of the reported differ- factors because of vowel (/a/, /o/, /i/) and gender (female/male). ences in jitter and shimmer between men and women.

FIGURE 1. The dramatic relationship of (left) jitter and (right) shimmer with gender, vowel, and voice SPL. It is clear that voice SPL represents the single biggest source of variability in the measurements. The effect of fundamental frequency is not considered in these ﬁgures. 48 Journal of Voice, Vol. 25, No. 1, 2011

TABLE 2. Means and Confidence Intervals for Voice SPL, F0, Jitter, and Shimmer Men Women Acoustic Parameters /a/ /i/ /o/ /a/ /i/ /o/ Voice SPL (dB) 78.5 (77.6–79.3) 77.5 (76.6–78.4) 80.2 (79.4–81.1) 73.2 (72.3–74.1) 72.3 (71.4–73.2) 75.0 (74.1–75.9) F0 (Hz) 127 (123–132) 143 (139–148) 129 (124–133) 214 (209–218) 230 (225–234) 215 (211–220) Jitter (%) 0.30 (0.28–0.33) 0.26 (0.24–0.29) 0.27 (0.24–0.29) 0.37 (0.33–0.40) 0.32 (0.29–0.35) 0.32 (0.29–0.35) Shimmer (dB) 0.31 (0.27–0.35) 0.44 (0.39–0.49) 0.38 (0.34–0.43) 0.46 (0.41–0.51) 0.65 (0.58–0.73) 0.57 (0.50–0.64) Notes: Mean values and 95% confidence intervals for each acoustic parameter with respect to vowel and gender. Note that the confidence intervals for jitter and shimmer are not symmetrical; this is a consequence of the logarithmic transformation.

Statistical significance of confounding factors biggest sources of variability. By contrast, the combined effects on jitter and shimmer of F0, gender, vowel, and the interactions between the factors The results in Table 3 use an ANCOVA model to assess the con- are relatively small. founding effects of continuous variables voice SPL and F0 and fixed factors of gender and vowel. We note that all the factors voice SPL, F0, gender, and vowel The effect of correcting for the confounding factors have a statistically significant effect on either jitter or shimmer, In the left part of Figure 3 are displayed the original jitter and or both. For the interactions, there is no clear picture with some shimmer measurements from Table 2, by gender and vowel. significant effects. Nevertheless, statistical significance simply On the right side is shown the effect of correcting each jitter indicates that an effect is probably not attributable to chance and shimmer measurement according to its own voice SPL alone. It gives no indication of how big the effect is, and there- and fundamental frequency. fore, we must consider separately which of these effects are There are a number of key observations to be made from clinically important. Figure 3, moving from uncorrected (left) to corrected (right) measurements. First, there is less spread in the measurements Relative importance of confounding factors because the variability introduced by voice SPL and F0 has on jitter and shimmer been removed. In Figure 2, the eta-squared measures of effect size taken from Second, in every case, the gap between the genders has Table 3 are shown. These sum to 1 and give a good impression changed in favor of women appearing to have relatively less of the relative contributions of the various sources of error. irregular voice signals. For jitter, after correction, the difference In each case, the voice SPL, the intersubject differences and has become smaller but is still statistically significant. For shim- random (unexplained) measurement error are clearly the mer, there is no evidence of any difference between the genders.

TABLE 3. Results of ANCOVA Assessing Main Effects and Interactions of the Confounding Factors Jitter Shimmer

Confounding Factors DF P Eta-Squared P Eta-Squared Main effects Voice SPL 1 <<0.001* 0.24 <<0.001* 0.62 F0 1 0.02* 0.03 0.02* 0.02 Gender 1 0.002* 0.04 0.4 0.00 Vowel 2 0.1 0.00 <<0.001* 0.06 Subject 55 <<0.001* 0.33 <<0.001* 0.18 Interactions Gender with voice SPL 1 0.8 0.00 0.7 0.00 Gender with F0 2 0.01* 0.01 0.05* 0.00 Vowel with voice SPL 1 <0.001* 0.01 0.9 0.00 Vowel with F0 2 0.3 0.01 <0.001* 0.00 Gender with vowel 2 0.8 0.00 0.06 0.00 Residual variance not explained by the model Error 0.34 0.11 Notes: Results of the ANCOVA model. For each effect are given the degrees of freedom (DF), the P value, and eta-squared, which represents the proportion of the total variance explained by the effect in question. The effects marked with an asterisk are considered as statistically signiﬁcant. Meike Brockmann, et al Vowel Inﬂuence on Jitter and Shimmer 49

Jitter

Shimmer

0 0.2 0.4 0.6 0.8 1

SPL Fo Gender Vowel All interactions Subject Error

FIGURE 2. Relative inﬂuence of the investigated confounding factors on jitter and shimmer. The sources of variance for jitter and shimmer, as given by the eta-squared statistic. It is clear that in each case, the voice SPL, intersubject differences, and random (unexplained) measurement error are the biggest sources of variability. Eta-squared has the desirable property of always summing to 1.

Finally, Figure 4 is directly comparable with Figure 1, show- terms can be neglected. This is fortunate; for example, a large ing the results after correction for voice SPL. interaction between gender and vowel would mean that we would need to apply a separate correction for each combination DISCUSSION of gender and vowel. As discussed in the Introduction, the effects of vowel and gen- Moving to the main effects, it is clear even from Figure 1 that der on objective acoustic measurements have been described in voice SPL has a dramatic effect on both jitter (eta- contradictory terms in the literature. In our previous work, we squared ¼ 0.24) and particularly shimmer (eta-squared ¼ 0.62). showed that voice SPL differences between individuals and That is, almost two-thirds of all the variance in acoustic shim- between genders have an enormous effect on measurements mer measurements is simply because of variability in voice of jitter and shimmer.42 This effect is largely unrecognized SPL. Fundamental frequency plays a smaller although statisti- and uncontrolled in the literature. Also, fundamental frequency cally significant role (eta-squared ¼ 0.02). differences between vowels and between women and men The effect of correcting for voice SPL and F0 is shown in Fig- might contribute to this misunderstanding. Therefore, we spec- ures 3 and 4, particularly on the shimmer measurements where ulated that assessing the confounding effects of voice SPL and the overall variability is substantially reduced. We therefore F0 might clear up the influence of vowel and gender on acoustic conclude that voice SPL differences have to be considered the 42,46 measures. In this article, we have assessed the influence of predominant confounding factor in acoustic measurements. vowel and gender in 57 healthy women and men while correct- For jitter, the subject’s gender has a statistically significant ing for the confounding effects of voice SPL and fundamental effect but vowel did not. In principle, we therefore do not need frequency.3,4,26 to correct our jitter measurements for vowel differences. Never- theless, it appears that women have higher jitter (more irregular Which statistically significant effects are clinically voice signals) in some vowels that cannot be explained entirely relevant? by their softer phonation and higher fundamental frequency. As demonstrated in Table 3, all examined factors vowel, gender, Vowel had a significant effect on shimmer, and therefore, voice SPL and F0 and a number of interactions between them vowel effects must be considered as an independent influencing have measurable and statistically significant effects on either factor affecting shimmer measurements. Rather more surpris- jitter or shimmer, or both. In principle, all these factors could ingly, gender was not. After the large correction for voice be measured in the clinic, and our jitter and shimmer measure- SPL, we can find no evidence that there is any difference ments could be adjusted accordingly. Indeed, many clinical between men’s and women’s shimmer measurements. In con- measurement systems have separate normative values for trast to previous findings, women appear not to have naturally male and female. However, a complex correction with multiple higher shimmer when the confounding effects of voice SPL 35,36,38 factors clearly becomes unwieldy and unlikely to be used in are corrected for. We believe that the earlier results are practice. Therefore, we need pragmatic decisions about which explained by the fact that women tend to phonate systematically are the relevant factors, and so we should consider which effects softer in the same clinical voice task.46 are clinically important in the context of the other factors known to affect acoustic measurements. One pragmatic index of use in our analysis is eta-squared, How important are these effects in the context of which represents the proportion of the total variance that can between-subject effects and random variability? be explained by the factor. By convention, a value of 0.01 or In clinical diagnostics, we are most interested in between- lower is considered a small effect. By reference to Table 3 subject variability (ie, the difference between one subject and and Figure 2, we can therefore conclude that all the interaction the next, who has clinically a different condition). If we know 50 Journal of Voice, Vol. 25, No. 1, 2011

FIGURE 3. Box and whisker graphs showing the median, interquartile range, 95% range, and outliers for (top) jitter and (bottom) shimmer. In the left panel are the original measurements. In the right panel are the measurements after correction for voice SPL and F0 using ANCOVA. the between-subject variability, then any effects that are small Our ﬁrst observation is that random variability is relatively by comparison are arguably unimportant. large and typically similar to the between-subject variability As we have already concluded, voice SPL is the single most (Figure 1). The random variability illustrates the fundamental important factor. For shimmer, the typical effect of voice SPL is uncertainty of objective acoustic measurements in voice clinics. greater even than the variability introduced by changing the This should be always considered when interpreting acoustic subject. The other factors F0, gender, and vowel are less impor- measurement results. tant. In particular, we must conclude that although it seems intuitively almost essential to control for gender in acoustic How can we make jitter and shimmer measurements, the evidence is much less clear. Certainly, our measurements more accurate? data suggest there is just as much reason to control for vowel The effects of voice SPL and F0 can be relatively easily as for gender. measured and controlled; Figures 3 and 4 show the effect of By taking multiple measurements from each subject, we also this correction. However, the practicalities of doing so in the estimated the measurement error because of random variability clinic would require the cooperation of the equipment manufac- or factors that were not controlled for in this experiment. This turer to measure the voice SPL and perform the correction. In gives an indication of the typical error on a single measurement. the meantime, because gender has such a clear effect on voice Meike Brockmann, et al Vowel Inﬂuence on Jitter and Shimmer 51

FIGURE 4. The relationship of (left) jitter and (right) shimmer with gender and vowel, after the effect of voice SPL has been modeled using ANCOVA, and subtracted from the original data. For comparison with Figure 1, the measurements are shown against their original voice SPL

(ie, the distribution on the x-axis is identical). However, the spread against the y-axis is considerably less and has no relationship with F0.

SPL, we would recommend that gender-specific norm values be Random variability by itself cannot (by definition) be com- used where possible. These normative values are currently built pletely controlled but can normally be reduced by taking the into many, but not all, acoustic analysis systems. However, mean of repeated measurements. The mean of N measurements given that voice SPL has such a dramatic effect on our measure- would reduce this term by a factor of ON. A previous work by ments, we also suggest that, as much as possible, women and Scherer56 suggests a number of 6 repetitions in stable voices men should phonate at a standard voice SPL in acoustic assess- and 15 repetitions in unstable phonations. ments. Based on the evidence from our previous work, we suggest a minimum standard voice SPL of 80 dB (at 10 cm 42 What improvements can we hope to see? distance). However, it has to be further investigated if it In a relatively simple change to our usual clinical protocol,1,5 introduces jitter or shimmer when adults or voice patients are we might standardize on a minimum target voice SPL of 80 dB asked to produce a target voice SPL or if patients in voice clinics (at 10 cm distance), on the vowel /a/ for all measurements, and are able to match this level satisfactorily. Based on the current 29 48 take the mean of six phonations. Table 3 and Figure 1 allow evidence from works by Orlikoff and Gelfer, we know some estimate of the improvement we can make by these simple that jitter and shimmer should not increase considerably measures. because of this change to the usual clinical voice task. Further- If we give our original measurement protocol a sensitivity of more, 80 dB was achievable even for patients with functional 54 55 100%, then Table 4 shows the sensitivity of the new protocol. dysphonia or polyps. For example, a sensitivity of 200% would correspond directly Likewise, using different normative values for each vowel to the ability to detect changes two times smaller than with would improve the reliability of perturbation measurements. the original protocol. Clearly, there are significant gains to be However, this still leaves the difficulty of interpreting clinical made from relatively simple changes to our clinical practice. voice assessments and research reports where different vowels Finally, in Table 4, we have also considered the potential are used. A much simpler solution for clinical purposes would effect of controlling SPL. Although we do not expect that perfect be always to use the same vowel in acoustic assessments. control of voice SPL is humanly possible, it seems plausible that According to the physical linkage hypothesis, vowel effects we might reduce the effects by half. This must be the target of might be produced by an interrelation between movements in future research and will be the key aim of our next study. the upper vocal tract and laryngeal movements.32,33 Based on this and from a pragmatic standpoint, we suggest the use of an open vowel that is easy to imitate (ie, articulation movements CONCLUSIONS should be easily seen and interpreted) irrespective of the native In this pragmatic study, we investigated the influence of gender language, linguistic competence, or individual health problems and vowel on jitter and shimmer in a typical clinical voice task (such as hearing disorders). To our knowledge, the vowel /a/ while correcting for the confounding effects of voice SPL and would fulfill these criteria best. According to the source-filter F0. Surprisingly, the effects of vocal intensity differences theory, a change in the nasal resonance because of language between individuals had a stronger impact on jitter and shimmer or dialect variations should not influence jitter and shimmer measurements than any other factor. Fundamental frequency measurements.29,31 had a relatively small influence. Also, vowel and gender effects 52 Journal of Voice, Vol. 25, No. 1, 2011

TABLE 4. Increase in Sensitivity According to Changes in the Measurement Protocol Perturbation N ¼ 1 Without Any Mean of N ¼ 6 Controlling Mean of N ¼ 6 Controlling Parameter Control (%) for Vowel (%) for Vowel and SPL (%) Jitter 100 135 194 Shimmer 100 111 202 Notes: The relative sensitivities of three different acoustic measurement protocols. For convenience, the first protocol has a sensitivity of 100%. The data for the third protocol are speculative, based on a reduction in the variability of the voice SPL variability by half. were considerably smaller but statistically significant and clini- 16. Gelzinis A, Verikas A, Bacauskiene M. Automated speech analysis applied cally important. Therefore, current guidelines are not sufficient to laryngeal disease categorization. Comput Methods Programs Biomed. to control for these effects in clinical practice. 2008;91:36–47. 17. Ortega J, Cassinello N, Dorcatto D, et al. Computerized acoustic voice anal- A voice assessment protocol that always uses the same vowel ysis and subjective scaled evaluation of the voice can avoid the need for lar- at a predefined voice intensity would clearly enhance measure- yngoscopy after thyroid surgery. Surgery. 2009;145:265–271. ment reliability. Based on the current evidence, our preliminary 18. Wang J, Jo C. Vocal folds disorder detection using pattern recognition suggestion is to use the vowel /a/ and phonations at a minimum methods. In: Annual International Conference of the IEEE Engineering of 80 dB. Furthermore, gender-specific thresholds applying in Medicine and Biology Society, August 22–26, 2007. 19. Hanson DG, Jiang JJ, chen J, et al. Acoustic measurement of change in these guidelines should be established. However, the feasibility voice quality with treatment for chronic posterior laryngitis. Ann Otol Rhi- and efficiency of these measures have to be further verified and nol Laryngol. 1997;106:279–285. tested. This will be the key aim of our next study. 20. Stojadinovic A, Shaha AR, Orlikoff RF, et al. Prospective functional voice assessment in patients undergoing thyroid surgery. Ann Surg. 2002;236: 554–560. 21. Ludlow C, Bassich CJ, Connor NP, et al. The validity of using jitter and shimmer to detect laryngeal pathology. In: Baer T, Sasaki C, Harris K, REFERENCES eds. Laryngeal Function in Phonation and Respiration: Vocal Fold Physi- 1. Dejonckere PH, Crevier-Buchmann L, Marie JP, et al. A basic protocol for ology Series. Boston, MA: College Hill; 1987:492–508. functional assessment of voice pathology, especially for investigating the 22. Zyski B, Bull GL, McDonald WE, et al. Perturbation analysis of normal and efficacy of (phonosurgical) treatments and evaluating new assessment tech- pathological larynges. Folia Phoniatr. 1984;36:190–198. niques. Eur Arch Otorhinolaryngol. 2001;258:77–82. 23. Rabinov CR, Kreimann J, et al. Comparing reliability of perceptual Ratings 2. Titze IR. Workshop on Acoustic Voice Analysis: Summary Statement.Iowa of roughness and acoustic measures of jitter. Journal of Speech and Hearing City, IA: National Center for Voice and Speech; 1994. Research 38:26–32. 3. Amplitude perturbation (shimmer). In: Baken R, Orlikoff RF, eds. Clinical 24. Carding PN, et al. The reliability and sensitivity to change of acoustic Measurements of Speech and Voice. Albany, NY: Thomson Delmar Learn- measures of voice quality. Clin Otolaryngol Allied Sci. 2004;29:538–544. ing; 2000:130–137. 25. Behrmann A, et al. Microphone and electroglottographic data from 4. Baken R, Orlikoff RF. Amplitude perturbation (shimmer). In: Clinical dysphonic patients: type 1, 2 and 3 signals. J Voice. 1998;12:249–260. Measurements of Speech and Voice. New York, NY: Thomson Delmar 26. Delyiski D, et al. Regression tree approach to studying factors influencing Learning; 2000:130–137. acoustic voice analysis. Folia Phoniatr Logop. 2006;58:274–288. 5. Titze IR. Workshop on Acoustic Voice Analysis: Summary Statement.Iowa 27. Heiberger V, Horii Y. Jitter and shimmer in sustained phonation. In: Lass N, City, IA: National Center for Voice and Speech; 1995. ed. Speech and Language: Advances in Basic Research and Practice.New 6. Hodge F, Colton R, Kelley R. Vocal intensity characteristics in normal and York, NY: Academic Press; 1982:299–332. elderly speakers. J Voice. 2001;15:503–511. 28. KiliçMA, O¨ gu¨t F, Dursun G, et al. Effects of vowels on voice perturbation 7. Ma E, Yiu E. Multiparametric evaluation of dysphonic severity. J Voice. measures. J Voice. 2004;18:318–324. 2006;20:380–390. 29. Orlikoff R. Vocalstability and vocal tract configuration: an acoustic electro- 8. Eksteen E, Rieger J, Nesbitt M, et al. Comparison of voice characteristics glottographic investigation. J Voice. 1995;9:173–181. following three different methods of treatment for laryngeal cancer. J Oto- 30. Milenkovic P. Least mean square measures of voice perturbation. J Speech laryngol. 2003;32:250–253. Hear Res. 1987;30:529–538. 9. Sjo¨gren EV, Rossum van MA, Langeveld TP, et al. Voice outcome in T1a 31. Kent RD. Vocal tract acoustics. J Voice. 1993;7:97–117. midcord glottic carcinoma: laser surgery vs radiotherapy. Arch Otolaryngol 32. Honda K. Relationship between pitch control and vowel articulation. In: Head Neck Surg. 2008;134:965–972. Bless DM, Abbs JH, eds. Vocal Fold Physiology: Contemporary Research 10. Orlikoff RF, Baken RJ. The effect of the heartbeat on vocal fundamental and Clinical Issues. San Diego, CA: College-Hill Press; 1983:286–297. frequency perturbation. J Speech Hear Res. 1989;32:576–582. 33. Lin E, et al. Effects of head extension and tongue protrusion on voice 11. Titze IR. A model for neurologic sources of aperiodicity in vocal fold vibra- perturbation measures. J Voice. 2000;14:8–16. tion. J Speech Hear Res. 1991;34:460–472. 34. Horii Y. Vocal shimmer in sustained phonation. J Speech Hear Res. 12. Schoentgen J. Qualitative evaluation of the discriminating performance of 1980;23:202–209. acoustic features in detecting laryngeal pathology. J Speech Hear Res. 35. Sorensen D, Horii Y. Frequency and amplitude perturbation in the voices of 1982;13:489–502. female speakers. J Commun Disord. 1983;16:57–61. 13. Liebermann P. Some acoustic measures of the fundamental periodicity of 36. Deem J, Manning WH, Knack JV, et al. The automatic extraction of pitch normal and pathologic larynxes. J Acoust Soc Am. 1963;35:344–353. perturbation using microcomputers: some methodological considerations. 14. Murry T, Doherty ET. Selected acoustic characteristics of pathologic and J Speech Hear Res. 1989;32:689–697. normal speakers. J Speech Hear Res. 1980;23:361–369. 37. Fitch JL. Consistency of fundamental frequency and perturbation in 15. Klingholz F, Martin F. Quantitative spectral evaluation of shimmer and repeated phonations of sustained vowels, reading, and connected speech. jitter. J Speech Hear Res. 1985;28:169–174. J Speech Hear Disord. 1990;55:360–363. Meike Brockmann, et al Vowel Influence on Jitter and Shimmer 53

38. Dwire A, McCauley R. Repeated measures of vocal fundamental frequency 48. Gelfer MP. Fundamental frequency, intensity, and vowel selection: effects perturbation obtained using the Visi-Pitch. J Voice. 1995;9:156–162. on measures of phonatory stability. J Speech Hear Res. 1995;38: 39. Jafari M, et al. Time-shift, trial and gender effects on vocal perturbation 1189–1198. measurements. J Voice. 1993;7:326–336. 49. Hirano M. Clinical examination of voice. In: Arnold G, Winckel F, 40. Sussmann JE, Sapienza C. Articulatory, developmental, and gender effects on Wyke BD, eds, Disorders of Human Communication, Vol. 5. Wien/New measures of fundamental frequency and jitter. J Voice. 1994;8:145–156. York: Springer-Verlag; 1981. 41. Orlikoff RF, Baken R. Consideration of the relationship between the funda- 50. Winholtz W, Titze I. Conversion of a head-mounted microphone signal into mental frequency of phonation and vocal jitter. Folia Phoniatr. 1990;42: calibrated SPL units. J Voice. 1997;11:417–421. 31–40. 51. Wagener F, Ku¨hnel V, Kollmeier B. Entwicklung und Evaluation eines 42. Brockmann M, Storck C, Carding PN, et al. Voice loudness and gender Satztestes in deutscher Sprache I: Design des Oldenburger Satztestes effects on jitter and shimmer in healthy adults. J Speech Lang Hear Res. (German) ‘‘Development and evaluation of a sentence based hearing test 2008;51:1–9. in German language I: Design of the Oldenburger Satztest.’’. Z Audiol. 43. Orlikoff RF, Kahane JC. Inﬂuence of mean sound pressure level on jitter 1999;38:4–15. and shimmer measures. J Voice. 1991;5:113–119. 52. Mazzoni D, Brubeck M, Crook J, et al. Audacity. Hamburg, Germany: Free 44. Pabon J. Objective acoustic voice-quality parameters in the computer Software Foundation; 2005. phonetogram. J voice. 1991;5:203–216. 53. Boersma P, Weenink D. PRAAT. Amsterdam, The Netherlands: University 45. Brown W, Rothman H, Sapienza C. Perceptual and acoustic study of of Amsterdam; 2006. professionally trained versus untrained voices. J Voice. 2000;14:301–309. 54. Gramming P. Relationship between changes in voice pitch and loudness. 46. Brown WJ, Morris R, Murry T. Comfortable effort level revisited. J Voice. J Voice. 1988;2:118–126. 1996;10:299–305. 55. Pabon JP, Plomp R. Automatic phonetogram recording supplemented with 47. Nittrouer S, McGowan RS, Milenkovic PH, et al. Acoustic measurements acoustical voice-quality parameters. J Speech Hear Res. 1988;31:710–722. of men’s and women’s voices. A study of context effects and covariation. 56. Scherer RC. Required number of tokens to determine representative voice J Speech Hear Res. 1990;33:761–775. perturbation values. J Speech Hear Res. 1995;38:1260–1269. 本文献由“学霸图书馆-文献云下载”收集自网络，仅供学习交流使用。

学霸图书馆（www.xuebalib.com）是一个“整合众多图书馆数据库资源，

提供一站式文献检索和下载服务”的24 小时在线不限IP 图书馆。图书馆致力于便利、促进学习与科研，提供最强文献下载服务。

图书馆导航：

图书馆首页文献云下载图书馆入口外文数据库大全疑难文献辅助工具