Spectral Analysis of Digital Kymography in Normal Adult Vocal Fold Vibration *Wenli Chen, †Peak Woo, and ‡Thomas Murry, *yzNew York, New York Summary: The purpose of this study was to examine the spectrum of normal adult vocal fold vibration obtained through direct visualization technique using digital kymography (DKG). DKG extracts high-speed kymographic images of each vocal fold sampled at a single or multiple points along the vocal folds. Thus, direct and objective quantification of the cycle-to-cycle movements of the left and right vocal folds were obtained. In this study, DKG spectrum configuration in males and females were examined. Samples were obtained from seven subjects (three males and four females) with no history of voice disorders. Subjects were asked to produce tokens obtained from a standard clinical laryngeal evaluation producing tokens at modal, low, and high frequency and at modal frequency with increased loudness. Results demonstrated that the spectrum of normal adult phonation consisted of a large and robust H1 (fundamental frequency) and significant attenuation of power in the higher harmonics (H2, H3). The spectral peaks were quasiperiodic with no spectral smearing. Analysis of the spectral data revealed variations in the spectrum that were influenced by the frequency of phonation and gender. Subjects demonstrated a range of left-right asymmetry of the spectral peaks (2–26%). This study presents a preliminary database of direct spectral characteristics of normal adult vocal fold vibration over a range of frequencies and intensities. Furthermore, these measures provide data from which similar measures obtained from dysphonic pathologies can be compared. Key Words: Digital kymography–Kymography–Spectrum–Vocal fold vibration–Normative data–High-speed videoendoscopy. INTRODUCTION Spectral analysis has been used to objectively quantify vibratory movements of the vocal folds. Because vibratory movement of the vocal folds critically determines the quality of voice production, spectral analysis can further elucidate the dynamic vibratory behavior of the vocal folds in both normal and pathologic voices. Recent applications using spectral analysis to study human adult phonation have focused on indirect approaches, such as using acoustic measures.1–4 Although these indirect measures provide valuable evidence of the vibratory motion of normal and disordered vocal folds, there remains little data from direct spectral analysis of vocal fold vibration. High-speed videoendoscopy (HSV) has been used to examine vibratory movements of the vocal folds in normal phonation.5–8 HSV provides direct visualization of vocal fold movement in real-time and captures images at frame rates of 2000 to 10 000 frames per second. This improved capturing rate enables one to visualize the entire cycle-to-cycle vibratory motion of the vocal folds in normal and severely disordered phonation that cannot be tracked with stroboscopy.9 HSV can also be used to generate kymographic data, also known as digital kymography (DKG). Kymography quantifies cycle-to-cycle movement of each vocal fold sampled from single or multiple lines across the vocal folds over time.10,11 As a result,
kymography is ideally suited for visual judgment of left-right asymmetry12–14 and deriving the vocal fold vibratory spectrum in a wide variety of vibratory features in healthy and disordered voices.15 Although DKG spectrum has been reported in excised canine larynges,16 DKG studies in human have focused on their relationship to perceptual ratings12–14 and to time series measurements.17–19 To our knowledge, spectral analysis of normal vocal fold vibration from DKG has not been examined across a wide range of frequencies or intensities in human phonation. Previous approaches to objectively quantify direct vocal fold movement in human phonation focused on measurements of glottal area5–8,20,21 and laryngeal topogram.22 Spectral analysis of kymographic data provided additional information (eg, frequency, periodicity, peak power, symmetry) on the vibratory behavior of each vocal fold movement. Such a technique may provide clinically useful information for diagnostic clarity in cases, in which other endoscopic techniques may have limited usefulness. Therefore, the purpose of this study was to examine the spectrum of normal male and female phonation during standard clinical laryngeal evaluation tasks using DKG obtained from HSV. Furthermore, the study aims to establish a preliminary normative database for the vibratory parameters derived from DKG spectrum of the vocal folds in normal adult human phonation.
Accepted for publication October 21, 2013. A version of the article was presented at the 42 Annual Symposium Care of the Professional Voice; June 1, 2013; Philadelphia, PA. From the *Department of Biobehavioral Sciences, Teacher’s College, Columbia University, New York, New York; yDepartment of Otolaryngology - Head and Neck Surgery, Icahn School of Medicine, New York, New York; and the zDepartment of Otolaryngology-Head and Neck Surgery, Weill Cornell Medical College, New York, New York. Address correspondence and reprint requests to Wenli Chen, 145 East 48th Street, Apt 27G, New York, NY 10017. E-mail: [email protected] Journal of Voice, Vol. 28, No. 3, pp. 356-361 0892-1997/$36.00 Ó 2014 The Voice Foundation http://dx.doi.org/10.1016/j.jvoice.2013.10.015
METHODS Subject selection criteria and description Following a laryngeal examination by a laryngologist, seven adult subjects (four females and three males) with ages ranging between 24 and 65 years were entered into the study. All subjects had no reported history of smoking, neurologic disease, laryngeal surgery, voice disorders, hearing problems, and speech or language impairment. Examination of their vocal folds before obtaining the samples verified no lesions or significant asymmetry of vibration.
Wenli Chen, et al
Spectrum of Normal Adult Vocal Fold Vibration
Phonatory tasks All subjects performed four phonatory tasks typically used in standard laryngovideostroboscopy. In the modal task, subjects produced /i/ at a comfortable pitch and loudness. This was each subject’s reference point for the other three tokens. In the low task, subjects produced /i/ at a low pitch and at a comfortable loudness. In the high task, subjects produced /i/ at a high pitch and at a comfortable loudness. Finally, in the modal loud task, subjects produced /i/ at their modal pitch with a loud voice. Using these four measures, the subjects established a broad frequency and intensity range. A second experimenter monitored the subjects’ fundamental frequency (F0) and intensity and provided verbal feedback to each subject to ensure that they met the target production established in the trials. Subjects were instructed to practice their individual targets in short, successive 2-second tokens. Throughout the tasks, the investigator provided modeling and feedback to assist the subject in maintaining target frequency and intensity. In short, the targets were chosen as common comfortable effort level phonation similar to a standard clinical laryngeal examination in which comfortable effort level, a high frequency, and low frequency relative to the comfortable effort level as well as a loud comfortable effort level were sampled. High-speed video recording HSV was acquired using KayPENTAX High-Speed Digital Imaging (HSDI) system (KayPENTAX Photronmotion; Pentax Medical, Montvale, NJ), which consisted of a 70-degree rigid endoscope (model 9100) coupled with a 300-W Xenon light source. The HSDI system acquired gray-scale images at a rate of 2000-frames per second with a spatial resolution of 256 3 120 pixels rotated to a vertical position for capture. Videolaryngoscopy was performed as in conventional videostroboscopy procedure. A condenser microphone was placed 6 in from the lips to monitor the intensity readings from an A-weighted sound level meter in the HSDI system. A contact microphone was held at the neck to monitor the fundamental frequency. Following the instructions and practice trials, the samples were obtained when the examiner observed a clear and full view of the larynx and frequency and intensity of phonation were verified to be consistent by the second investigator. A template outlining the vocal folds was mounted on the monitor in order for the examiner to fit the visual image of the vocal folds within the template. This provided a consistent size of the vocal folds across tokens for later analysis. Six continuous 2-second tokens phonations were captured. The three best token with a clear and full view of the larynx were saved onto the hard drive for analysis. The F0 and intensity of each phonation segment were manually recorded. All subjects tolerated the data collection procedure without any difficulties. Data analysis Kymography image processing. Kymograph analysis of the vibratory samples is critically dependent on accurate delineation of the edge of the vocal folds from HSV.23 Therefore, HSV samples were preprocessed using video editing
357
software (VirtualDub, V.1.9.11). Image rotation (bicubic resampling) was implemented such that the image was judged to be vertically aligned. An initial brightness and contrast adjustments were manually applied to images that were perceived to have a less distinct contrast between the vocal fold edges and the glottis. Images were not adjusted for size as they were obtained within the template described above. A 400- to 500frame video segment that demonstrated a full view of the vocal fold with minimal movement of the subject was extracted from the recorded HSV samples. Kay’s Image Processing Software (KIPS, model 9181; Pentax Medical, Montvale, NJ) was used to generate the kymogram by placing a transverse line across the glottis at the mid-membranous portion of the vocal folds, where vocal fold contact is greatest at this point.13 Edge detection was subsequently applied to identify and trace the vocal fold edges (Figure 1). If automatic tracing was imprecise, a new kymographic sample was generated and manual correction functions internal to the KIPS (eg, brightening, darkening, and erase function) were applied onto the new kymograph. This process was repeated until the automatic tracing accurately delineated the vocal fold edges. When the vocal fold edges were delineated, Kymograph Edge Analysis function was applied. The resulting values were Kymograph Edge Data (KED), which described the coordinate values of the left and right edges of the vocal fold presented across time (Figure 2). Subsequently, Fourier transform function was applied to the KED, resulting in a spectrum ranging from 0 to 1000 Hz for the left and right edges of the vocal fold. The kymographic image processing was repeated for all phonatory tasks (Figure 3). Spectral data analyses. The frequency and spectral peak power values of the F0 (also known as H1), second harmonic (H2), and third harmonic (H3) were obtained to examine the spectral shape of male and female phonation during the four laryngeal tasks. Twenty percent of the postprocessed 500-frame HSV tokens were randomly selected and reanalyzed by the same experimenter to evaluate error of measurement. Comparison of the H1, H2, and H3 values between the original and reanalyzed sample yielded a reliability of 93%. The following vocal fold vibratory parameters were examined: (1) spectral smearing; (2) spectrum shape; (3) spectral power configuration; (4) gender effects; and (5) spectral power symmetry. Spectral smearing was determined by examining the presence of the noise-band at a specific frequency along the spectrum. Spectrum shape was determined by summing the spectral peak values of H1, H2, and H3 individually across all
FIGURE 1. Kymogram of P3 during modal phonation task. The top image is a kymogram created with the sampling line selected at the mid-membranous portion of the vocal fold. The bottom image is the same kymogram following edge detection.
358
Journal of Voice, Vol. 28, No. 3, 2014
FIGURE 2. KED data of P3 during modal phonation task. Kymograph Edge Analysis function was selected, which resulted in KED. KED visually displays the movements of the left and right vocal folds across time.
subjects and tasks. Spectral power configuration is defined as the proportion of the fundamental power (H1) relative to the spectral power of the harmonics (summation of H1, H2, and H3). Gender effects were determined by comparing the sum of the spectral peak power of the harmonics (summation of H1, H2, and H3) between the male and female subjects. Additionally, spectral power configuration across the phonatory tasks was also compared between genders. Spectral power asymmetry examined the differences between the left versus right spectral peak power divided by the sum of the left and right spectral peak power, expressed as a percentage. RESULTS The mean frequency and intensity of phonation were as follows: modal task (female x ¼ 216.3 Hz, male x ¼ 119.7 Hz), low task (female x ¼ 177.5 Hz, male x ¼ 106.7 Hz), high task (female x ¼ 320.3 Hz, male x ¼ 203.1 Hz), and modal loud task (female x ¼ 241.5 Hz and 7.25 dB increase, male x ¼ 126 Hz and 7 dB increase). Frequency and intensity for each subject are shown in Table 1. Spectral smearing No spectral smearing was observed in all subjects. DKG spectrum shape In healthy adult voice, DKG spectrum was characterized by a large and robust peak (H1; x ¼ 7.1, standard error [SE] ¼ 1.6), followed by a medium-to-small second spectral peak (H2; x ¼ 2.5, SE ¼ 1.5), followed by a third small peak (H3; x ¼ 0.48, SE ¼ 0.13; Figure 4). This spectral shape was present in all tasks. A repeated measures analysis of variance (ANOVA) demonstrated significant differences in H1, H2, and H3 (P < 0.001). Pairwise comparison with Bonferroni correction demonstrated a significant difference for amplitude in every level (P < 0.001). Thus, the spectral shape of the spectrum in normal adult phonation remained relatively consistent regardless of the phonation task. Changes in spectral configuration with variations of frequency and intensity Spectrum configuration was determined by examining the relative contribution of the spectral peak power of H1 divided by
the spectral power of the harmonics (summation of H1, H2, and H3). High frequency tokens (x ¼ 0.74, SE ¼ 0.079) demonstrated a larger ratio of H1 relative to spectral power of the harmonics ratio than modal frequency tokens (x ¼ 0.68, SE ¼ 0.084). Modal frequency tokens demonstrated a higher ratio of H1 relative to spectral power of the harmonics ratio than low frequency tokens (x ¼ 0.62, SE ¼ 0.089; Figure 5). A repeated measures ANOVA found an increasing trend but not statistical significance (P ¼ 0.07) of phonation frequency to the ratio of H1 to spectral power of the harmonics. Modal frequency loud tokens (x ¼ 0.65, SE ¼ 0.077) demonstrated a slightly lower H1 relative to spectral power of the harmonics than modal frequency tokens (x ¼ 0.68, SE ¼ 0.084). However, the effect of intensity on the ratio of H1 to overall spectral power was not significant. Gender Gender differences in spectral power of the harmonics were examined. Adult males (x ¼ 19.67, SE ¼ 5.07) demonstrated greater spectral power of the harmonics than female (x ¼ 9.29, SE ¼ 3.46). A repeated measures ANOVA showed that a significant difference (P < 0.002) in spectral peak power was found between males and females. Spectral configurations were compared between the adult male and females across phonatory tasks. Adult male and female vocal fold showed similar patterns of spectral configuration across phonation tasks. A repeated measures ANOVA demonstrated that gender did not have significant interaction effect across tasks. Thus, although male vocal folds showed significantly greater total spectral power than females, the changes in spectral power during different phonation tasks were similar at the mid-membranous region of the vocal folds. Asymmetry Asymmetry of vocal fold vibration was examined by percentage of spectral power symmetry. Spectral power symmetry was determined by the difference between the left versus the right spectral peak power divided by the summation of the left and right spectral peak power, expressed as a percentage. All subjects demonstrated spectral peak asymmetry (x ¼ 13.42%, ranged from 2% to 26%). A repeated measures
Wenli Chen, et al
359
Spectrum of Normal Adult Vocal Fold Vibration
FIGURE 3. DKG spectrum for P3 in modal, low, high, and modal loud tasks. An example of DKG spectrum for P3 across the four phonatory tasks. Spectral configuration changes are seen with changes in frequency (modal, low, high) and intensity (modal loud). ANOVA showed that symmetry of left and right spectral peaks did not significantly differ among tasks.
Although spectral configuration in normal adults presented a tilted shape overall, the specific configuration of the spectral peaks was influenced in part by the frequency of the phonation. Gauffin and Sundberg25 reported that acoustic spectral energy of the first harmonic is associated with the degrees of vocal fold excursion, whereas the energies of the higher harmonics can be associated with the discontinuity that occurs with vocal fold impact. During high frequency phonation, the vocal folds close quickly and stay closed for a shorter portion of the glottal cycle.21,26,27 The DKG spectral configuration changes were consistent with acoustical findings. With higher frequency of phonation, we found an increase in the power of the fundamental (H1) relative to the spectral power of the harmonics (summation of H1, H2, and H3). Reduction of spectral power in the higher harmonics may be due to the shorter discontinuities (eg, closed phase) during highfrequency phonation. Spectral configuration changes were inconsistent with the loud phonation tokens. During loud phonation, the vocal folds close quickly and stay closed for a longer portion of the glottal cycle.21,26,27 This should increase the spectral power in the higher harmonics, which decreases the fundamental power
DISCUSSION The purpose of this study was to determine the spectral characteristics of a broad range of voice samples obtained from DKG. The normal adult spectrum obtained from the present study consisted of robust harmonic peaks that were quasiperiodic. The spectrum had no spectral smearing and presented with a tilted shape. This consisted of a large spectral peak at H1 (F0) and significant attenuation of power in higher harmonics (H2 and H3). The tilted spectral shape was present across the variations of phonation examined. Similar results have been extensively reported in spectral studies using acoustical measures.1 Acoustic spectrum of normal adult phonation consisted of a tilted shape with no spectral smearing.1,24 The acoustical spectral slope for normal voices consists of approximately 12 dB/octave, 6 dB/octave for pressed voices, and 18 dB/ octave for breathy voices.24 Both acoustic and direct vocal fold vibratory-based spectrum in normal adult voice demonstrated a tilted configuration and an absence of spectral smearing.
TABLE 1. Frequency and Intensity of Phonation for Each Subject Modal
Low
High
Modal Loud
Subject
Gender
Hz
dB
Hz
dB
Hz
dB
Hz
dB
P1 P2 P3 P4 P5 P6 P7
Male Male Male Female Female Female Female
132 117 110 218 234 195 218
71 75 70 74 75 76 74
109 109 102 195 187 164 164
71 74 70 70 74 74 72
203 219 187 320 328 343 289
75 75 78 72 76 78 70
128 117 133 249 257 234 226
77 86 75 84 85 81 78
360
FIGURE 4. Spectral shape of normal adult vocal fold vibration. H1, H2, and H3 were tabulated across subjects to determine the spectral shape of normal adult phonation. Healthy adult voices exhibited a tilted spectral shape with greater power in the H1, followed by attenuation of higher harmonics. relative to the higher harmonics. Our results were inconsistent and may be explained by the concomitant and variable elevation of pitch during loud phonation trials in some of the subjects (Table 1). Spectral power of normal phonation in the males studied was significantly larger than female phonation. This can be explained by the structural and physiological differences between the male and female vocal folds. Male vocal folds are generally larger and thicker than female vocal folds; they also vibrate with greater lateral excursions and larger amplitude than in females.21 Interestingly, we found that male and female spectra demonstrated similar spectral patterns across the range of samples studied. This may suggest that the vibratory behaviors at the mid-membranous portion are similar between normal males and females. Spectral power asymmetry was observed across tasks. Minor irregularities in normal phonation have been previously identified using HSV28,29 and kymography.12–14,29 Our results brought further evidence to support the presence of amplitude asymmetry in normal adult voices.28,29 Several limitations exist in this study. Our study consisted of seven subjects. More subjects would lend greater confidence to the present results. Although a capturing template was used, movements during video recording may have occurred; however, we only accepted samples that were within the template to maintain equal size of the images. Another limitation of this study is the range of frequency analysis of the software (0–1000 Hz). As a result, higher harmonics (eg, H4 or higher) were not obtained in all subjects. Thus, the full range of spectral peak power was not completely captured in our analysis. Future applications for spectral analysis of DKG should focus on identifying normative spectral peak values for the entire range of phonation as well as for various registers of phonation. Based on the present study, frequency, intensity, and gender appear to be variables that should be systematically examined with a larger cohort of subjects. Future study should extend the present study to compare DKG spectrum in normal and path-
Journal of Voice, Vol. 28, No. 3, 2014
FIGURE 5. Changes in spectral configuration with variation in the frequency. Individual data points across low, modal, and high and modal frequency loud tokens. Spectral configuration demonstrated an increase in the ratio of H1 (F0) to the sum of the harmonic energies (H1, H2, H3) related to the increase in F0. ologic voices, as several vibratory qualities of the vocal fold can be directly obtained and objectively quantified. Although acoustical spectra obtain vibratory movements of the vocal fold indirectly (eg, inverse filtering), DKG spectra captures directly vocal fold vibration. Future studies should also include comparisons between acoustic and image-based vibratory spectra to examine normal and abnormal vocal fold vibration. CONCLUSION The present study demonstrated that DKG is a direct and objective quantitative analysis of normal vocal fold vibratory behavior across a range of frequencies and intensities for males and females. Clinical appreciation of this procedure may provide use for diagnostic information when other methods of acoustic analysis or video assessment may lack specificity. This study provided additional information about the dynamic properties of the left and the right vocal folds during four phonatory conditions. In this study, seven normal adult subjects produced modal, high, low, and modal loud phonation. The spectrum of normal adult vocal fold vibration presented a tilted configuration with no spectral smearing. Spectral configuration was influenced in part by frequency and gender. Spectral power asymmetry was present in normal adult phonation. In summary, the present results offer a preliminary database of direct spectral characteristics of normal adult vocal fold vibration over a range of frequencies and intensities. Furthermore, these measures provide data from which similar measures obtained from dysphonic pathologies can be compared. REFERENCES 1. Hixon TJ, Weismer G, Hoit JD. Preclinical Speech Science: Anatomy, Physiology, Acoustics, Perception. 1st ed. San Diego, CA: Plural Publishing; 2008:357–386. 2. Cannito MP, Buder EH, Chorna LB. Spectral amplitude measures of adductor spasmodic dysphonic speech. J Voice. 2005;19:391–410. 3. Mehta DD, Zeitels SM, Burns JA, Friedman AD, Deliyski DD, Hillman RE. High-speed videoendoscopic analysis of relationships between cepstralbased acoustic measures and voice production mechanisms in patients
Wenli Chen, et al
4.
5.
6.
7.
8.
9. 10.
11. 12.
13. 14.
15.
Spectrum of Normal Adult Vocal Fold Vibration
undergoing phonomicrosurgery. Ann Otol Rhinol Laryngol. 2012;121: 341–347. Lowell SY, Colton RH, Kelley RT, Hahn YC. Spectral- and cepstral-based measures during continuous speech: capacity to distinguish dysphonia and consistency within a speaker. J Voice. 2011;25:223–232. Ahmad K, Yan Y, Bless D. Vocal-fold vibratory characteristics in normal female speakers from high-speed digital imaging. J Voice. 2012;26: 239–253. Yan Y, Damrose E, Bless D. Functional analysis of voice using simultaneous high-speed imaging and acoustic recordings. J Voice. 2007;21: 604–616. Yan Y, Ahmad K, Kunduk M, Bless D. Analysis of vocal-fold vibrations from high-speed laryngeal images using Hilbert Transform based methodology. J Voice. 2005;19:161–175. Yamauchi A, Imagawa H, Yokonishi H, et al. Evaluation of vocal fold vibration with an assessment form for high-speed digital imaging: comparative study between healthy young and elderly subjects. J Voice. 2012;26: 742–750. Herteg ard S, Larsson H, Wittenberg T. High-speed imaging: applications and development. Logoped Phoniatr Vocol. 2003;28:133–139. Wittenberg T, Tigges M, Mergell P, Eysholdt U. Functional imaging of vocal fold vibration: digital multislice high-speed kymography. J Voice. 2000;14:422–442. Svec JG, Schutte HK. Videokymography: high-speed line scanning of vocal fold vibration. J Voice. 1996;10:201–205. Bonilha H, Deliyski D, Whiteside J, Gerlach T. Vocal fold phase asymmetries in patients with voice disorders: a study across visualization techniques. Am J Speech Lang Pathol. 2012;21:3–15. Bonilha H, Deliyski D. Period and glottal width irregularities in vocally normal speakers. J Voice. 2008;22:699–708. Bonilha H, Deliyski D, Gerlach T. Phase asymmetries in normophonic speakers: visual judgments and objective findings. Am J Speech Lang Pathol. 2008;17:367–376. Svec JG, Sram F, Schutte HK. Videokymography in voice disorders: what to look for? Ann Otol Rhinol Laryngol. 2007;116:172–180.
361
16. Zhang Y, Krausert CR, Kelly MP, Jiang JJ. Typing vocal fold vibratory patterns in excised larynx experiments via digital kymography. Ann Otol Rhinol Laryngol. 2009;188:598–605. 17. Mehta DD, Zanartu M, Quatieri TF, Deliyski DD, Hillman RE. Investigating acoustic correlates of human vocal fold vibratory phase asymmetry through modeling and laryngeal high-speed videoendoscopy. J Acoust Soc Am. 2011;130:3999–4009. 18. Mehta DD, Deliyski DD, Quatieri TF, Hillman RE. Automated measurement of vocal fold vibratory asymmetry from high-speed videoendoscopy recordings. J Speech Lang Hear Res. 2011;54:47–54. 19. Freeman E, Woo P, Saxman JH, Murry T. A comparison of sung and spoken phonation onset gestures using high-speed digital imagining. J Voice. 2012; 26:226–238. 20. Lohscheller J, Eysholdt U. Phonovibrogram visualization of entire vocal fold dynamics. Laryngoscope. 2008;118:753–758. 21. Woo P. Quantification of videostrobolaryngoscopic findings: measurements of the normal glottal cycle. Laryngoscope. 1996;106:1–27. 22. Granqvist S, Lindestad PA. A method of applying Fourier analysis to highspeed laryngoscopy. J Acoust Soc Am. 2001;110:3193–3197. 23. Patel R, Dixon A, Richmond A, Donohue K. Pediatric high speed digital imaging of vocal fold vibration: a pilot normative study of glottal closure and phase closure. Int J Pediatr Otorhinolaryngol. 2012;76: 954–959. 24. Titze IR. Principles of Voice Production. Iowa City, IA: National Center for Voice and Speech; 2000. 25. Gauffin J, Sundberg J. Spectral correlates of glottal voice source waveform characteristics. J Speech Hear Res. 1989;32:556–565. 26. Timcke R, von Leden H, Moore P. Laryngeal vibrations: measurements of the glottic wave, part one. AMA Arch Otolaryngol. 1958;68:1–19. 27. Timcke R, von Leden H, Moore P. Laryngeal vibrations: measurements of the glottic wave, part two. AMA Arch Otolaryngol. 1959;69:438–444. 28. Shaw HS, Deliyski DD. Mucosal wave: a normophonic study across visualization techniques. J Voice. 2008;22:23–33. 29. Kendall K. High-speed laryngeal imaging compared with videostroboscopy in healthy subjects. Arch Otolaryngol Head Neck Surg. 2009;135:274–281.
kymography is ideally suited for visual judgment of left-right asymmetry12–14 and deriving the vocal fold vibratory spectrum in a wide variety of vibratory features in healthy and disordered voices.15 Although DKG spectrum has been reported in excised canine larynges,16 DKG studies in human have focused on their relationship to perceptual ratings12–14 and to time series measurements.17–19 To our knowledge, spectral analysis of normal vocal fold vibration from DKG has not been examined across a wide range of frequencies or intensities in human phonation. Previous approaches to objectively quantify direct vocal fold movement in human phonation focused on measurements of glottal area5–8,20,21 and laryngeal topogram.22 Spectral analysis of kymographic data provided additional information (eg, frequency, periodicity, peak power, symmetry) on the vibratory behavior of each vocal fold movement. Such a technique may provide clinically useful information for diagnostic clarity in cases, in which other endoscopic techniques may have limited usefulness. Therefore, the purpose of this study was to examine the spectrum of normal male and female phonation during standard clinical laryngeal evaluation tasks using DKG obtained from HSV. Furthermore, the study aims to establish a preliminary normative database for the vibratory parameters derived from DKG spectrum of the vocal folds in normal adult human phonation.
Accepted for publication October 21, 2013. A version of the article was presented at the 42 Annual Symposium Care of the Professional Voice; June 1, 2013; Philadelphia, PA. From the *Department of Biobehavioral Sciences, Teacher’s College, Columbia University, New York, New York; yDepartment of Otolaryngology - Head and Neck Surgery, Icahn School of Medicine, New York, New York; and the zDepartment of Otolaryngology-Head and Neck Surgery, Weill Cornell Medical College, New York, New York. Address correspondence and reprint requests to Wenli Chen, 145 East 48th Street, Apt 27G, New York, NY 10017. E-mail: [email protected] Journal of Voice, Vol. 28, No. 3, pp. 356-361 0892-1997/$36.00 Ó 2014 The Voice Foundation http://dx.doi.org/10.1016/j.jvoice.2013.10.015
METHODS Subject selection criteria and description Following a laryngeal examination by a laryngologist, seven adult subjects (four females and three males) with ages ranging between 24 and 65 years were entered into the study. All subjects had no reported history of smoking, neurologic disease, laryngeal surgery, voice disorders, hearing problems, and speech or language impairment. Examination of their vocal folds before obtaining the samples verified no lesions or significant asymmetry of vibration.
Wenli Chen, et al
Spectrum of Normal Adult Vocal Fold Vibration
Phonatory tasks All subjects performed four phonatory tasks typically used in standard laryngovideostroboscopy. In the modal task, subjects produced /i/ at a comfortable pitch and loudness. This was each subject’s reference point for the other three tokens. In the low task, subjects produced /i/ at a low pitch and at a comfortable loudness. In the high task, subjects produced /i/ at a high pitch and at a comfortable loudness. Finally, in the modal loud task, subjects produced /i/ at their modal pitch with a loud voice. Using these four measures, the subjects established a broad frequency and intensity range. A second experimenter monitored the subjects’ fundamental frequency (F0) and intensity and provided verbal feedback to each subject to ensure that they met the target production established in the trials. Subjects were instructed to practice their individual targets in short, successive 2-second tokens. Throughout the tasks, the investigator provided modeling and feedback to assist the subject in maintaining target frequency and intensity. In short, the targets were chosen as common comfortable effort level phonation similar to a standard clinical laryngeal examination in which comfortable effort level, a high frequency, and low frequency relative to the comfortable effort level as well as a loud comfortable effort level were sampled. High-speed video recording HSV was acquired using KayPENTAX High-Speed Digital Imaging (HSDI) system (KayPENTAX Photronmotion; Pentax Medical, Montvale, NJ), which consisted of a 70-degree rigid endoscope (model 9100) coupled with a 300-W Xenon light source. The HSDI system acquired gray-scale images at a rate of 2000-frames per second with a spatial resolution of 256 3 120 pixels rotated to a vertical position for capture. Videolaryngoscopy was performed as in conventional videostroboscopy procedure. A condenser microphone was placed 6 in from the lips to monitor the intensity readings from an A-weighted sound level meter in the HSDI system. A contact microphone was held at the neck to monitor the fundamental frequency. Following the instructions and practice trials, the samples were obtained when the examiner observed a clear and full view of the larynx and frequency and intensity of phonation were verified to be consistent by the second investigator. A template outlining the vocal folds was mounted on the monitor in order for the examiner to fit the visual image of the vocal folds within the template. This provided a consistent size of the vocal folds across tokens for later analysis. Six continuous 2-second tokens phonations were captured. The three best token with a clear and full view of the larynx were saved onto the hard drive for analysis. The F0 and intensity of each phonation segment were manually recorded. All subjects tolerated the data collection procedure without any difficulties. Data analysis Kymography image processing. Kymograph analysis of the vibratory samples is critically dependent on accurate delineation of the edge of the vocal folds from HSV.23 Therefore, HSV samples were preprocessed using video editing
357
software (VirtualDub, V.1.9.11). Image rotation (bicubic resampling) was implemented such that the image was judged to be vertically aligned. An initial brightness and contrast adjustments were manually applied to images that were perceived to have a less distinct contrast between the vocal fold edges and the glottis. Images were not adjusted for size as they were obtained within the template described above. A 400- to 500frame video segment that demonstrated a full view of the vocal fold with minimal movement of the subject was extracted from the recorded HSV samples. Kay’s Image Processing Software (KIPS, model 9181; Pentax Medical, Montvale, NJ) was used to generate the kymogram by placing a transverse line across the glottis at the mid-membranous portion of the vocal folds, where vocal fold contact is greatest at this point.13 Edge detection was subsequently applied to identify and trace the vocal fold edges (Figure 1). If automatic tracing was imprecise, a new kymographic sample was generated and manual correction functions internal to the KIPS (eg, brightening, darkening, and erase function) were applied onto the new kymograph. This process was repeated until the automatic tracing accurately delineated the vocal fold edges. When the vocal fold edges were delineated, Kymograph Edge Analysis function was applied. The resulting values were Kymograph Edge Data (KED), which described the coordinate values of the left and right edges of the vocal fold presented across time (Figure 2). Subsequently, Fourier transform function was applied to the KED, resulting in a spectrum ranging from 0 to 1000 Hz for the left and right edges of the vocal fold. The kymographic image processing was repeated for all phonatory tasks (Figure 3). Spectral data analyses. The frequency and spectral peak power values of the F0 (also known as H1), second harmonic (H2), and third harmonic (H3) were obtained to examine the spectral shape of male and female phonation during the four laryngeal tasks. Twenty percent of the postprocessed 500-frame HSV tokens were randomly selected and reanalyzed by the same experimenter to evaluate error of measurement. Comparison of the H1, H2, and H3 values between the original and reanalyzed sample yielded a reliability of 93%. The following vocal fold vibratory parameters were examined: (1) spectral smearing; (2) spectrum shape; (3) spectral power configuration; (4) gender effects; and (5) spectral power symmetry. Spectral smearing was determined by examining the presence of the noise-band at a specific frequency along the spectrum. Spectrum shape was determined by summing the spectral peak values of H1, H2, and H3 individually across all
FIGURE 1. Kymogram of P3 during modal phonation task. The top image is a kymogram created with the sampling line selected at the mid-membranous portion of the vocal fold. The bottom image is the same kymogram following edge detection.
358
Journal of Voice, Vol. 28, No. 3, 2014
FIGURE 2. KED data of P3 during modal phonation task. Kymograph Edge Analysis function was selected, which resulted in KED. KED visually displays the movements of the left and right vocal folds across time.
subjects and tasks. Spectral power configuration is defined as the proportion of the fundamental power (H1) relative to the spectral power of the harmonics (summation of H1, H2, and H3). Gender effects were determined by comparing the sum of the spectral peak power of the harmonics (summation of H1, H2, and H3) between the male and female subjects. Additionally, spectral power configuration across the phonatory tasks was also compared between genders. Spectral power asymmetry examined the differences between the left versus right spectral peak power divided by the sum of the left and right spectral peak power, expressed as a percentage. RESULTS The mean frequency and intensity of phonation were as follows: modal task (female x ¼ 216.3 Hz, male x ¼ 119.7 Hz), low task (female x ¼ 177.5 Hz, male x ¼ 106.7 Hz), high task (female x ¼ 320.3 Hz, male x ¼ 203.1 Hz), and modal loud task (female x ¼ 241.5 Hz and 7.25 dB increase, male x ¼ 126 Hz and 7 dB increase). Frequency and intensity for each subject are shown in Table 1. Spectral smearing No spectral smearing was observed in all subjects. DKG spectrum shape In healthy adult voice, DKG spectrum was characterized by a large and robust peak (H1; x ¼ 7.1, standard error [SE] ¼ 1.6), followed by a medium-to-small second spectral peak (H2; x ¼ 2.5, SE ¼ 1.5), followed by a third small peak (H3; x ¼ 0.48, SE ¼ 0.13; Figure 4). This spectral shape was present in all tasks. A repeated measures analysis of variance (ANOVA) demonstrated significant differences in H1, H2, and H3 (P < 0.001). Pairwise comparison with Bonferroni correction demonstrated a significant difference for amplitude in every level (P < 0.001). Thus, the spectral shape of the spectrum in normal adult phonation remained relatively consistent regardless of the phonation task. Changes in spectral configuration with variations of frequency and intensity Spectrum configuration was determined by examining the relative contribution of the spectral peak power of H1 divided by
the spectral power of the harmonics (summation of H1, H2, and H3). High frequency tokens (x ¼ 0.74, SE ¼ 0.079) demonstrated a larger ratio of H1 relative to spectral power of the harmonics ratio than modal frequency tokens (x ¼ 0.68, SE ¼ 0.084). Modal frequency tokens demonstrated a higher ratio of H1 relative to spectral power of the harmonics ratio than low frequency tokens (x ¼ 0.62, SE ¼ 0.089; Figure 5). A repeated measures ANOVA found an increasing trend but not statistical significance (P ¼ 0.07) of phonation frequency to the ratio of H1 to spectral power of the harmonics. Modal frequency loud tokens (x ¼ 0.65, SE ¼ 0.077) demonstrated a slightly lower H1 relative to spectral power of the harmonics than modal frequency tokens (x ¼ 0.68, SE ¼ 0.084). However, the effect of intensity on the ratio of H1 to overall spectral power was not significant. Gender Gender differences in spectral power of the harmonics were examined. Adult males (x ¼ 19.67, SE ¼ 5.07) demonstrated greater spectral power of the harmonics than female (x ¼ 9.29, SE ¼ 3.46). A repeated measures ANOVA showed that a significant difference (P < 0.002) in spectral peak power was found between males and females. Spectral configurations were compared between the adult male and females across phonatory tasks. Adult male and female vocal fold showed similar patterns of spectral configuration across phonation tasks. A repeated measures ANOVA demonstrated that gender did not have significant interaction effect across tasks. Thus, although male vocal folds showed significantly greater total spectral power than females, the changes in spectral power during different phonation tasks were similar at the mid-membranous region of the vocal folds. Asymmetry Asymmetry of vocal fold vibration was examined by percentage of spectral power symmetry. Spectral power symmetry was determined by the difference between the left versus the right spectral peak power divided by the summation of the left and right spectral peak power, expressed as a percentage. All subjects demonstrated spectral peak asymmetry (x ¼ 13.42%, ranged from 2% to 26%). A repeated measures
Wenli Chen, et al
359
Spectrum of Normal Adult Vocal Fold Vibration
FIGURE 3. DKG spectrum for P3 in modal, low, high, and modal loud tasks. An example of DKG spectrum for P3 across the four phonatory tasks. Spectral configuration changes are seen with changes in frequency (modal, low, high) and intensity (modal loud). ANOVA showed that symmetry of left and right spectral peaks did not significantly differ among tasks.
Although spectral configuration in normal adults presented a tilted shape overall, the specific configuration of the spectral peaks was influenced in part by the frequency of the phonation. Gauffin and Sundberg25 reported that acoustic spectral energy of the first harmonic is associated with the degrees of vocal fold excursion, whereas the energies of the higher harmonics can be associated with the discontinuity that occurs with vocal fold impact. During high frequency phonation, the vocal folds close quickly and stay closed for a shorter portion of the glottal cycle.21,26,27 The DKG spectral configuration changes were consistent with acoustical findings. With higher frequency of phonation, we found an increase in the power of the fundamental (H1) relative to the spectral power of the harmonics (summation of H1, H2, and H3). Reduction of spectral power in the higher harmonics may be due to the shorter discontinuities (eg, closed phase) during highfrequency phonation. Spectral configuration changes were inconsistent with the loud phonation tokens. During loud phonation, the vocal folds close quickly and stay closed for a longer portion of the glottal cycle.21,26,27 This should increase the spectral power in the higher harmonics, which decreases the fundamental power
DISCUSSION The purpose of this study was to determine the spectral characteristics of a broad range of voice samples obtained from DKG. The normal adult spectrum obtained from the present study consisted of robust harmonic peaks that were quasiperiodic. The spectrum had no spectral smearing and presented with a tilted shape. This consisted of a large spectral peak at H1 (F0) and significant attenuation of power in higher harmonics (H2 and H3). The tilted spectral shape was present across the variations of phonation examined. Similar results have been extensively reported in spectral studies using acoustical measures.1 Acoustic spectrum of normal adult phonation consisted of a tilted shape with no spectral smearing.1,24 The acoustical spectral slope for normal voices consists of approximately 12 dB/octave, 6 dB/octave for pressed voices, and 18 dB/ octave for breathy voices.24 Both acoustic and direct vocal fold vibratory-based spectrum in normal adult voice demonstrated a tilted configuration and an absence of spectral smearing.
TABLE 1. Frequency and Intensity of Phonation for Each Subject Modal
Low
High
Modal Loud
Subject
Gender
Hz
dB
Hz
dB
Hz
dB
Hz
dB
P1 P2 P3 P4 P5 P6 P7
Male Male Male Female Female Female Female
132 117 110 218 234 195 218
71 75 70 74 75 76 74
109 109 102 195 187 164 164
71 74 70 70 74 74 72
203 219 187 320 328 343 289
75 75 78 72 76 78 70
128 117 133 249 257 234 226
77 86 75 84 85 81 78
360
FIGURE 4. Spectral shape of normal adult vocal fold vibration. H1, H2, and H3 were tabulated across subjects to determine the spectral shape of normal adult phonation. Healthy adult voices exhibited a tilted spectral shape with greater power in the H1, followed by attenuation of higher harmonics. relative to the higher harmonics. Our results were inconsistent and may be explained by the concomitant and variable elevation of pitch during loud phonation trials in some of the subjects (Table 1). Spectral power of normal phonation in the males studied was significantly larger than female phonation. This can be explained by the structural and physiological differences between the male and female vocal folds. Male vocal folds are generally larger and thicker than female vocal folds; they also vibrate with greater lateral excursions and larger amplitude than in females.21 Interestingly, we found that male and female spectra demonstrated similar spectral patterns across the range of samples studied. This may suggest that the vibratory behaviors at the mid-membranous portion are similar between normal males and females. Spectral power asymmetry was observed across tasks. Minor irregularities in normal phonation have been previously identified using HSV28,29 and kymography.12–14,29 Our results brought further evidence to support the presence of amplitude asymmetry in normal adult voices.28,29 Several limitations exist in this study. Our study consisted of seven subjects. More subjects would lend greater confidence to the present results. Although a capturing template was used, movements during video recording may have occurred; however, we only accepted samples that were within the template to maintain equal size of the images. Another limitation of this study is the range of frequency analysis of the software (0–1000 Hz). As a result, higher harmonics (eg, H4 or higher) were not obtained in all subjects. Thus, the full range of spectral peak power was not completely captured in our analysis. Future applications for spectral analysis of DKG should focus on identifying normative spectral peak values for the entire range of phonation as well as for various registers of phonation. Based on the present study, frequency, intensity, and gender appear to be variables that should be systematically examined with a larger cohort of subjects. Future study should extend the present study to compare DKG spectrum in normal and path-
Journal of Voice, Vol. 28, No. 3, 2014
FIGURE 5. Changes in spectral configuration with variation in the frequency. Individual data points across low, modal, and high and modal frequency loud tokens. Spectral configuration demonstrated an increase in the ratio of H1 (F0) to the sum of the harmonic energies (H1, H2, H3) related to the increase in F0. ologic voices, as several vibratory qualities of the vocal fold can be directly obtained and objectively quantified. Although acoustical spectra obtain vibratory movements of the vocal fold indirectly (eg, inverse filtering), DKG spectra captures directly vocal fold vibration. Future studies should also include comparisons between acoustic and image-based vibratory spectra to examine normal and abnormal vocal fold vibration. CONCLUSION The present study demonstrated that DKG is a direct and objective quantitative analysis of normal vocal fold vibratory behavior across a range of frequencies and intensities for males and females. Clinical appreciation of this procedure may provide use for diagnostic information when other methods of acoustic analysis or video assessment may lack specificity. This study provided additional information about the dynamic properties of the left and the right vocal folds during four phonatory conditions. In this study, seven normal adult subjects produced modal, high, low, and modal loud phonation. The spectrum of normal adult vocal fold vibration presented a tilted configuration with no spectral smearing. Spectral configuration was influenced in part by frequency and gender. Spectral power asymmetry was present in normal adult phonation. In summary, the present results offer a preliminary database of direct spectral characteristics of normal adult vocal fold vibration over a range of frequencies and intensities. Furthermore, these measures provide data from which similar measures obtained from dysphonic pathologies can be compared. REFERENCES 1. Hixon TJ, Weismer G, Hoit JD. Preclinical Speech Science: Anatomy, Physiology, Acoustics, Perception. 1st ed. San Diego, CA: Plural Publishing; 2008:357–386. 2. Cannito MP, Buder EH, Chorna LB. Spectral amplitude measures of adductor spasmodic dysphonic speech. J Voice. 2005;19:391–410. 3. Mehta DD, Zeitels SM, Burns JA, Friedman AD, Deliyski DD, Hillman RE. High-speed videoendoscopic analysis of relationships between cepstralbased acoustic measures and voice production mechanisms in patients
Wenli Chen, et al
4.
5.
6.
7.
8.
9. 10.
11. 12.
13. 14.
15.
Spectrum of Normal Adult Vocal Fold Vibration
undergoing phonomicrosurgery. Ann Otol Rhinol Laryngol. 2012;121: 341–347. Lowell SY, Colton RH, Kelley RT, Hahn YC. Spectral- and cepstral-based measures during continuous speech: capacity to distinguish dysphonia and consistency within a speaker. J Voice. 2011;25:223–232. Ahmad K, Yan Y, Bless D. Vocal-fold vibratory characteristics in normal female speakers from high-speed digital imaging. J Voice. 2012;26: 239–253. Yan Y, Damrose E, Bless D. Functional analysis of voice using simultaneous high-speed imaging and acoustic recordings. J Voice. 2007;21: 604–616. Yan Y, Ahmad K, Kunduk M, Bless D. Analysis of vocal-fold vibrations from high-speed laryngeal images using Hilbert Transform based methodology. J Voice. 2005;19:161–175. Yamauchi A, Imagawa H, Yokonishi H, et al. Evaluation of vocal fold vibration with an assessment form for high-speed digital imaging: comparative study between healthy young and elderly subjects. J Voice. 2012;26: 742–750. Herteg ard S, Larsson H, Wittenberg T. High-speed imaging: applications and development. Logoped Phoniatr Vocol. 2003;28:133–139. Wittenberg T, Tigges M, Mergell P, Eysholdt U. Functional imaging of vocal fold vibration: digital multislice high-speed kymography. J Voice. 2000;14:422–442. Svec JG, Schutte HK. Videokymography: high-speed line scanning of vocal fold vibration. J Voice. 1996;10:201–205. Bonilha H, Deliyski D, Whiteside J, Gerlach T. Vocal fold phase asymmetries in patients with voice disorders: a study across visualization techniques. Am J Speech Lang Pathol. 2012;21:3–15. Bonilha H, Deliyski D. Period and glottal width irregularities in vocally normal speakers. J Voice. 2008;22:699–708. Bonilha H, Deliyski D, Gerlach T. Phase asymmetries in normophonic speakers: visual judgments and objective findings. Am J Speech Lang Pathol. 2008;17:367–376. Svec JG, Sram F, Schutte HK. Videokymography in voice disorders: what to look for? Ann Otol Rhinol Laryngol. 2007;116:172–180.
361
16. Zhang Y, Krausert CR, Kelly MP, Jiang JJ. Typing vocal fold vibratory patterns in excised larynx experiments via digital kymography. Ann Otol Rhinol Laryngol. 2009;188:598–605. 17. Mehta DD, Zanartu M, Quatieri TF, Deliyski DD, Hillman RE. Investigating acoustic correlates of human vocal fold vibratory phase asymmetry through modeling and laryngeal high-speed videoendoscopy. J Acoust Soc Am. 2011;130:3999–4009. 18. Mehta DD, Deliyski DD, Quatieri TF, Hillman RE. Automated measurement of vocal fold vibratory asymmetry from high-speed videoendoscopy recordings. J Speech Lang Hear Res. 2011;54:47–54. 19. Freeman E, Woo P, Saxman JH, Murry T. A comparison of sung and spoken phonation onset gestures using high-speed digital imagining. J Voice. 2012; 26:226–238. 20. Lohscheller J, Eysholdt U. Phonovibrogram visualization of entire vocal fold dynamics. Laryngoscope. 2008;118:753–758. 21. Woo P. Quantification of videostrobolaryngoscopic findings: measurements of the normal glottal cycle. Laryngoscope. 1996;106:1–27. 22. Granqvist S, Lindestad PA. A method of applying Fourier analysis to highspeed laryngoscopy. J Acoust Soc Am. 2001;110:3193–3197. 23. Patel R, Dixon A, Richmond A, Donohue K. Pediatric high speed digital imaging of vocal fold vibration: a pilot normative study of glottal closure and phase closure. Int J Pediatr Otorhinolaryngol. 2012;76: 954–959. 24. Titze IR. Principles of Voice Production. Iowa City, IA: National Center for Voice and Speech; 2000. 25. Gauffin J, Sundberg J. Spectral correlates of glottal voice source waveform characteristics. J Speech Hear Res. 1989;32:556–565. 26. Timcke R, von Leden H, Moore P. Laryngeal vibrations: measurements of the glottic wave, part one. AMA Arch Otolaryngol. 1958;68:1–19. 27. Timcke R, von Leden H, Moore P. Laryngeal vibrations: measurements of the glottic wave, part two. AMA Arch Otolaryngol. 1959;69:438–444. 28. Shaw HS, Deliyski DD. Mucosal wave: a normophonic study across visualization techniques. J Voice. 2008;22:23–33. 29. Kendall K. High-speed laryngeal imaging compared with videostroboscopy in healthy subjects. Arch Otolaryngol Head Neck Surg. 2009;135:274–281.
