ACOUSTIC ANALYSIS OF STUTTERERS' FLUENT SPEECH BEFORE AND AFTER THERAPY

Transcription

1 Journal of Speech and Hearing Research, METZ ET AL., Volume 26, , December 1983 ACOUSTIC ANALYSIS OF STUTTERERS' FLUENT SPEECH BEFORE AND AFTER THERAPY DALE EVAN METZ VINCENT J. SAMAR Natiomtl Technical Institute.[}Jr tl DeaJ, Rvchester Institute of Technology, Rochester. Ne York PAT RICHARD SACCO State Unicersit!/ College of Arts and Sciences, Geneseo, Ne York The speech of 14 stutterers as analyzed prior to and at the termination of a 5-eek stuttering therapy program to examine the relationship beteen nine selected acoustic variables mad stuttering frequency. Group analyses indicated that pre- to posttherapy changes in stuttering frequency ere accompanied by mean changes in five of the nine acoustic variables, a finding hich is consistent ith previous literature. Correlational analyses indicated that only silence in the voiced stop consonant inter~calic interval (IVI) as significantl3 correlated ith stuttering frequency prior to therapy (i.e., loer stnttering frequency values ere associated ith shorter durations of silence during the IVI). Furthermore, the degree of reduction in silence as positively correlated ith the magnitude of reduction in stuttering t)equency dne to therapy. These findings suggest that silence in the IVI may reflect the operational status of some mechanism hich may underlie disfluent speech. In the decade folloing Wingate's seminal articles (1969, 1970) on artificially induced fluency, considerable research effort as directed toard identifying underlying conditions related to fluency enhancement among stutterers. The emphasis of much of this research as on isolating acoustic speech parameters of stutterers' fluent speech that varied concomitantly ith stuttering frequency. For example, the relationships among changes in vocal sound pressure level (Adams & Hutchinson, 1974; Adams & Moore, 1972; Conture, 1974), vocal fundamental frequency, and voel duration and the groth of fluency associated ith stutterers' speaking in the presence of loud noise and speaking in accordance ith a rhythmic stimulus (Brayton & Conture, 1978) ere investigated. Attention as paid to the relationships among changes in voel duration, continuity of phonation, and vocal sound pressure level and the groth of fluency associated ith stutterers" singing (Colcord & Adams, 1979; Healey, Mallard, & Adams, 1976) and choral speaking (Adams & Ramig, 1980). Additionally, it is clear from both perceptual (Runyan & Adams, 1978) and acoustie studies (Metz, Onufrak, & Ogburn, 1979) that stuttering therapy can alter certain acoustic properties of the stutterer's fluent speech. These and other related studies have advanced our knoledge regarding hich acoustic variables associated ith the fluent speech of stutterers change concomitantly ith the groth of fluency. Hoever, tiaese studies are based on group data and consequently provide limited information about the nature of the relationships beteen acoustic variable changes and fluency enhancement. Furthermore, there is no a priori reason to assume that all systematic changes in the acoustic properties of stutterers' speech are necessarily related to fluency enhancement. Some of these alterations may reflect changes in the operation of the motor control processes that underlie fluency enhancement, hereas others may be systematic by-products of the particular fluencyenhancing condition. That is, a given fluency-enhancing condition (or therapy procedure) may have to types of effects on a stutterer's speech: (a) one that imposes changes in speech that directly enhance fluency, and hich are reflected in certain acoustic changes and hose occurrenee is probably common, to greater or lesser degrees, to all fluency-enhancing conditions; and (b) one that imposes changes in speech that do not directly enhance fluency, and hich are also reflected in certain acoustic changes but hose occurrence is probably an idiosyncratic result of ancillary aspects of the particular fluency-enhancing condition. By examining the relationship beteen stuttering frequency and various acoustic features of stutterers' speech across individuals (in addition to examining group data), e may obtain some clues regarding those variables that are most closely related to changes in stuttering frequency. Although correlation does not necessarily imply causation, acoustic variables closely related to stuttering frequency more likely reflect control processes underlying fluency than variables that are statistically independent of the stuttering frequency dimension ithin the population of stutterers. In the present study e explored such relationships in more detail. Specifically, e examined several acoustic variables measured from a single group of stutterers. These measurements included both traditional and previously unexplored acoustic properties of stutterers' fluent speech production. Furthermore, e examined the relationship beteen these acoustic variables and fluency ithin a group of mild-to-severe stutterers both prior to and after a concentrated program of stuttering therapy. 1983, American Speech-Language-Hearing Association /83/ /0

2 532 Journal of Speech and Hearing Research December 1983 Subjects METHOD The subjects ere 1"2 male and to female young adult stutterers hose stuttering frequency ranged from 2 to 54 stutterings per 100 ords read (~ of 15.4) as assessed during the pretherapy reading of the experimental passages. Each subject exhibited sound and syllable repetitions, sound prolongations, and/or broken ords. All subjects ere enrolled in a residential stuttering clinic at the State University College of Arts and Sciences, Geneseo, Ne York. Treatment Each subject received both group and individual stuttering therapy daily for 5 eeks. The procedure used generally folloed the treatment philosophy described by Van Riper (1973). The essential elements of the treatment ere (a) detailed analysis of all core and accessory stuttering behaviors; (b) training the stutterer to modify core stuttering behaviors through strategies like maintenance of air flo and continuous forard articulatory movements during sound, syllable, and ord transitions; and (c) stabilization of these therapeutic strategies. It is recognized that different therapy procedures might produce different changes in the stutterers' post-therapy fluent speech. The Van Riper procedure as chosen, hoever, because it deals directly ith the stutterers' speech symptomatology and is a procedure commonly used to modify stuttering. Stimulus Materials On both the first and last day of the clinic, each subject read a 1,000-ord prose passage designed by Brayton and Conture (1978). Sixteen stop consonant-voel-stop consonant (CVC) target ords ere located in the middle section of the passage. Additionally, each subjeet read a list of 24 monosyllabic (CV) English ords. This ord list consisted of 12 ord-initial voiced and 12 ord-initial voiceless consonants joined ith a voel. Each target ord as embedded in the carrier phrase Say again and individually typed on 3 x 5-in. (7.6 x 12.7 cm) cards. The prose passage and randomly ordered sentence cards ere presented to the subjects in counterbalanced order. These stimulus materials ere selected because they provide specific eontrolled reference to acoustic variables that frequently have been examined (and related to the groth of fluency) under conditions of fluency enhancement. These acoustic variables include voel duration, voice onset time, and selected intervocalic interval acoustic parameters. Procedure Before the experimental readings, each subject as in- structed to read all the test materials at a comfortable rate and loudness level. The subject as then seated in a sound-attenuating room ith one of the experimenters, ho presented the test materials. All readings ere recorded on a Nagra, Model IV-S tape recorder ith a constant mouth-to-microphone distance of 15 cm. Any target ord that as produced disfluently (by agreement of the first and third authors) during the preor posttherapy readings as eliminated from the final data pool. If a prose passage target ord as produced dis fluently, it simply as not analyzed. If a sentence target ord as produced disfluently during the experimental task, the card as placed at the end of the card deck and presented a second time. Disfluent productions during the second reading ere not analyzed. This procedure ensured that only fluently produced ords ere included in the data analysis. Approximately 10% of the total 560 pretherapy target ords (range = 0%-13% per subject) and 7% of the total 560 posttherapy target ords (range = 0%-9% per subject) ere eliminated from subsequent analyses. Intra- and Interjudge Reliabilit~d of Stuttering Frequency Judgments One of the experimenters and an independent judge (a graduate student in speech pathology) conducted separate reliability tests for part-ord repetitions, sound prolongations, and broken ords to assess intra- and interjudge agreements regarding instances of stuttering during the experimental readings. The experimenter's intrajudge interclass agreement coefficients (Sander, 1961) ere 1.0,.98, and.75 for pretherapy and 1.0, 1.0, and 1.0 for posttherapy instances of part-ord repetitions, sound prolongations, and broken ords, respectively. The overall average agreement coefficient as.97. The interjudge-interelass agreement coefficients ere.99,.99, and.50 for pretherapy and 1.0, 1.0, and.67 for posttherapy instances of part-ord repetitions, sound prolongations, and broken ords, respectively. The overall average coefficient as.86. Data Analysis The subjects' fluently produced readings ere digitized at 20 khz ith a Data General Eclipse computer and stored on magnetic tape. Expanded aveform representations of the readings ere visually displayed on a Tektronix CRT terminal. Target ords and critical boundary locations ere identified ith user-controlled cursors and interactive digital-to-analog acoustic playback. Thus, visual identification of critical boundaries as facilitated by the experimenter's ability to hear events ithin the boundary'. Specifically, avefonn representations of the target ords from the prose passage ere visually displayed. The interconsonantal voels ere sectioned ith cursors, and voel durations ere measured using the Eclipse aveform editor. Similarly, aveform repre-

3 METZ ET AL.: Stutterers" Fluent Speech 533 sentations of the test sentences ere visually displayed, and to separate analyses ere performed. First, the voice onset time (VOT) interval of the target ord's initial stop consonant as sectioned by cursors and measured. The VOT interval as defined as beginning at the point hen aperiodic energy associated ith consonant release as observed and ending hen the first regular quasi-periodic avefonn (associated ith the voel in the CV ord) as observed folloing consonant release. Second, the interval beteen the ord Sa~d and the voel onset of the target ord (intervocalic interval) as sectioned by cursors. This interval as defined as beginning at the last discernible quasi-periodic aveform associated ith the/e/voel in Sa!! and ending ith the first discernible quasi-periodic avefonn folloing the target ord's stop consonant. This interval then as analyzed ith a specially developed signal processing routine (Bancroft, 1979) that determined the absolute time the interval as void of energy, contained frication, or contained voicing energy. This analysis provided an index of the acoustic nature of intervocalic interval events associated ith voel to stop consonant closure and release. Nine variables resulted from these analyses: (a) voiced stop consonant VOT's; (b) voiceless stop consonant VOT's; (c) voel duration; and (d) the absolute times of frication, voicing, and silence associated ith the intervocalic intervals of both voiced and voiceless stop consonants. Intra- and Interjudge Boundary Location Reliability Sixty-five VOT intervals, associated intervoealic intervals, and 45 voel aveforms (approximately 10% of the total data corpus) ere randomly selected from the preand posttherapy conditions and remeasured by the senior author. Pearson correlation coefficients beteen the original and the remeasured interval and aveform durations ere all.99 or greater. Mean millisecond differences beteen the original and remeasured values ere small: (a) VOT, ~ =.1 ms, SD =.6 ms; (b) intervocalic interval, ~ =.4 ms, SD =.6 ms; and (c) voel duration, ~- =.2 ms, SD =.6 ms. An independent judge (a trained research assistant) measured these same intervals and aveforms. Interjudge correlations ere.99 for the VOT intervals,.98 for the intervocalic intervals, and.98 for the voel durations. Mean millisecond differences ere small: (a) VOT, ~- =.4 ms, SD = 3.8 ms; (b) intervocalic interval, =.5 ms, SD = 6.0 ms; and (3) voel duration, ~- = 1.7 ms, SD = 8.3 ms. RESULTS Group Mean Pre-Posttherapy Changes Table 1 displays the pre- and posttherapy group means and standard deviations for the stuttering frequency measure and the nine acoustic variables. Since stuttering frequency as computed as the proportion of stutterings per 100 ords read, this variable as transformed to arcsin values prior to statistical analysis. Because of the large number of dependent variables, e employed Bonferroni's procedure to adjust the alpha level for rejection of the null hypothesis experimentise. With 10 variables the alpha level is.05/10 =.005. The t ratios for the prepost difference scores on each of the variables listed in Table 1 exceeding the.005 alpha level are indicated by an asterisk. Table 1 shos a significant and dramatic decrease in stuttering frequency beteen the pre- and posttest. Furthermore, both voiceless and voiced VOT duration and voel duration significantly increased folloing therapy. Of the intervocalic interval parameters, frication duration significantly increased during production of voiceless stop consonant intervocalic intervals and voicing duration significantly increased during production of voiced stop consonant intervocalic intervals. Relationships Among Acoustic Variables and Stuttering Frequency Post minus pretherapy "change" scores ere computed for the transformed stuttering frequency variables and the nine acoustic variables. Correlations beteen the pretherapy and post minus pretherapy stuttering frequency variables and the acoustic pretherapy and post minus pretherapy variables ere then computed. Table 2 shos that only silence in the voiced stop consonant intervocalic interval as significantly correlated ith stuttering frequency prior to therapy. The relationship as such that loer stuttering frequency as associated ith less silence during that interval. Similarly, the change in the duration of silence in the voiced stop consonant intervocalic interval significantly correlated ith the change in stuttering frequency folloing therapy. The relationship as such that greater reduction of stuttering frequency after therapy as associated ith greater reduction of silence in that interval. Figure 1 displays the silence/stuttering frequency scatter data underlying the significant pretherapy and post minus pretherapy change correlations reported in Table 2. It is clear from the scattergrams that the relatively eak linear con-elation coefficients for the silence variable resulted from the presence of an extreme outlier (subject 3) rather than from a uniform elliptical bivariate scatter about the regression line. This suggests that the underlying relationships beteen silence and stuttering frequency represented in the graphs may be stronger over the population thar~ are indicated by the present small sample data. Inspection of the scatter data for all other correlations in Table 2 revealed no clear cases of outliers hich might have compromised the strength of the observed relationships. Additionally, stepise linear, multiple-regression analyses ere conducted separately for the pretherapy and post minus pretherapy change data. In each analysis the stuttering frequency variable as the dependent variable and the acoustic variables ere the independent

4 534 Journal of Speech and Hearing Research December 1983 TABLE 1. Group means, standard deviations (in parentheses), pre-posttherapy difference scores and t values for percent stuttering frequency and the nine acoustic variables (values given in milliseconds). All tests of significance ere based on Bonferronfs alpha adjustment procedure. The t test for the stuttering frequency variable as based on arcsin transformed scores. Mean difference Variables Pretherapy Posttherapy score t (dr = 13) Stuttering frequency (%) (14.2) (0.6) * VOT (voiceless stop consonants) ,1 (10.7) (20.1) * Voicing (voiceless stop consonant intervoealic intervals) (30.7) (23,1) Frication (voiceless stop consonant intervocalic intervals) (12,9) (33,1) " Silence (voiceless stop consonant intervocalic intervals) (46.4) (52,1) ,6 VOT (voiced stop consonants) (6.7) (11.8) * Voicing (voiced stop consonant intervocalic intervals) (13.7) (30.2) * Frication (voiced stop consonant intervocalic intervals) (10.6) (7.8) Silence (voiced stop consonant intervocalic intervals) (77.0) (14.3) 62,1 3.0 Voel duration 136, (28,2) (20.9) " *p <.005. TABLE 2. Simple correlations beteen the nine acoustic and the stuttering frequency variables for the pretherapy and post minus pretherapy measures. Variable VOT (voiceless stop consonants) Voicing (voiceless stop consonant intervocalic intervals) Frication (voiceless stop consonant intervoealie intervals) Silence (voiceless stop consonant intervocalic intervals) VOT (voiced stop consonants) Voicing (voiced stop consonant intervocalic intervals) Frication (voiced stop consonant intervocalie intervals) Silence (voiced stop consonant intervoealie intervals) Voel duration * p <.05. Pretherapy r * 0.32 Post minus pretherapy difference scores r * 0.33

5 METZ ET AL.: Stutterers' Fluent Speech 535 variables. The results from each analysis indicated that after the variable representing silence in the voiced stop consonant intervoealie interval entered the equation to predict significantly the stuttering frequency variables (pretherapy: F = 4.88, df = 1, 19., p <.05; change: F = 4.87, df = 1, 19,, p <.05), the partial correlations beteen the stuttering frequency residuals and the other acoustic variables became negligible and nonsignificant. tl.l 3 > z_ co < >- (3 z O 1~ 0.83 u. z,..r g t.d LLI 3 z n.- < ~z < "1-0 >.- 0 Z tic u_ ~ rr " , SILENCE DURATION IN MILLISECONDS SILENCE DURATION CHANGE IN MILLISECONDS FIOUI1E 1. Scatter data (upper graph) of the relationship beteen stuttering frequency and silence in the voiced stop consonant intervocalic interval prior to therapy. In the upper graph, positively increasing ordinate and abscissa values reflect increasing magnitudes of stuttering frequency and silence duration. Also shon (loer graph) is the relationship beteen the change in stuttering frequency and the change in silence folloing therapy. In the loer graph, negatively increasing ordinate and abscissa values reflect increasing magnitudes of change in stuttering frequency and silence duration. Note the extreme data outlier (subject 3) in both cases. 5 These data suggest that any trends toard significant correlations beteen the stuttering frequency variables and the other acoustic variables appearing in Table 2 ere due to shared variance beteen silence in the voiced stop consonant intervocalic interval and the other acoustic variables. Furthermore, e did not find any nonlinear trends hen e inspected all pairise scattergrams beteen the stuttering frequency variables and the acoustic variables corresponding to the correlations presented in Table 2. Although the correlational analyses indicate that silence in the stop consonant intervocalic interval is a significant correlate of stuttering frequency, this acoustic variable did not display a significant change in the group data using the stringent Bonferroni criterion, despite that numerically large pretherapy-to-posttherapy difference. Inspection of Table 1 reveals that this failure to achieve significance probably as due to the extremely large pretherapy variability. DISCUSSION The results of this study are consistent ith previous research indicating that stuttering therapy can alter the acoustic properties of stutterers' fluent speech concomitantly ith reducing stuttering frequency (Metz et al., 1979). Furthermore, these results extend previous research findings by indicating that frication during voiceless stop consonant intervoealic intervals and voicing during voiced stop consonant intervoealic intervals changed concomitantly ith stuttering frequency as a result of therapy. The principle finding of this research is the association of stuttering frequency and silence in the voiced stop consonant intervocalic interval (IVI). Silence as found to be significantly positively correlated ith stuttering frequency prior to therapy. Furthermore, the degree of reduction in silence as positively related to the magnitude of reduction in stuttering frequency due to therapy. After the voiced stop consonant silence variable as taken into account, the other acoustic variables did not significantly contribute to the prediction of pretherapy or post minus pretherapy changes in stuttering frequency. Finally, the large therapy-induced reduction in the variability of stuttering frequency evident in Table i as paralleled by a large reduction in variability in the voiced stop consonant silence variable. These data suggest that silence in the intervocalic interval may reflect the operational status of some underlying mechanism responsible for disfluent speech. We cannot unequivocally assert a necessary connection beteen silence in the intervocalic interval and fluency based on correlational data. Hoever, the consistent covariation of silence ith fluency state and therapy related fluency changes support the hypothesis of a necessary connection more strongly than the simple group data changes do for the other acoustic variables. Our results for the silence variable accord ell ith previously reported data on pause latency differences

6 536 Journal of Speech and Hearing Research December 1983 beteen stutterers and nonstutterers. Love and Jeffress (1971) found that the fluent speech of stutterers shoed, on the average, a significantly greater number of prolonged silent pauses compared ith the speech of nonstutterers. The ork of Love and Jeffress and the present data support previous suggestions (Adams & Runyan, 1981; Klich & May, 1982; Zimmermann, 1980; Zimmermann, Smith, & Hanley, 1981) that the perceptually fluent speech of stutterers is not fully "normal." Rather, at least one perceptually subliminal acoustic event (silence in the intervocalic interval) varies as a function of disfluency. We speculate that increased silence reflects an underlying disruption in the regulation of the normal timing of articulatory events involved in voiced stop consonant production hich can be expected to vary both ithin and across individuals. If so, one ould expect that the likelihood of disfluent event occurring during attempted productions of voiced ordinitial stop consonants, ithin samples from individual stutterers, ould be a positive function of the duration of silent events in the temporal vicinity of the initial consonant. This hypothesis can be tested by continuous monitoring of silence durations in continuous speech and examining the relationship beteen the duration of given silent events and the probability of a subsequent disfluency ithin individual subjects. In summary, the findings of Love and Jeffress suggest that stutterers have abnormally prolonged silent pauses in their fluent speech. The present findings suggest that decreases in the duration of silence in the stutterers' fluent speech relates to decreases in their disfluency. Collectively, these findings suggest that improvements in fluency are associated ith manipulations hich reduce or regularize the duration of silent hesitations in stutterers' speech to approximate normal limits. We speculate that the observed fluency related changes in the silence parameter reflect a normalization by speech therapy of underlying articulatory timing relationships and that they thus fundamentally reflect changes in the operation of mechanisms of fluent speech production. ACKNOWLEDGMENTS We are grateful to Dr. Nicholas Schiavetti for his critical revie of a previous version of this manuscript. Portions of this manuscript ere presented at the Annual Convention of the American Speech-Language-Hearing Association, Los Angeles, This research as sponsored by the Summer Residential Stuttering Clinic at the State University College of Arts and Science at Geneseo, Ne York. Computer facilities for data analysis ere generously provided by the National Technical Institute for the Deaf at Rochester Institute of Technology,, Rochester, Ne York. This manuscript is dedicated to the memory of Dr. Harold Starbuck, the founder of the Summer Residential Stuttering Clinic at SUNY at Geneseo, Ne York. Hal as a fine teacher, colleague, and friend. We deeply regret his passing. REFERENCES ADAMS, M. R., & HUTCHINSON, J. The effects of three levels of auditory masking on selected vocal characteristics and the frequency of disfluency of adult stutterers. Journal of Speech and Hearing Research, 1974, 17, ADAMS, M. R., MOORE, W. H. The effects of auditory masking on the anxiety level, frequency of disflueney, and selected vocal characteristics of stutterers. Journal of Speech and Hearing Research, 1972, 15, ADAMS, M. R., & RAMIG, P. Vocal characteristics of normal speakers and stutterers during choral reading. Journal of Speech and Hearing Research, 1980, 23, ADAMS, M. R., & RUNYAN, C. Stuttering and fluency: Exclusive events or points on a continuum. Journal of FIuenc~ Disorders, 1981, 6, BANCROFT, J. C. A periodicity measure for voiced speech.journal of the Acoustical Society of America, 1979, 65(Suppl. 1), 67. BRAYTON, E., & CONTURE, E. G. Effects of noise and rhythmic stimulation on the speech of stutterers.journal of Speech and Hearing Research, 1978, 21, COLCORD, a. D., d ADAMS, M. R. Voicing duration and vocal SPL changes associated ith stuttering reduction during singing. Journal of Speech and Hearing Research, 1979, 22, CONTURE, E. G. Some effects of noise on the speaking behavior of stutterers. Journal of Speech and Hearing Research, 1974, 17, HEALEY, C., MALLARD, R., & ADAMS, M. R. Factors contributing to the reduction of stuttering during singing. Journal of Speech and Research, 1976, 19, KLICH, R. J., & MAY, G. M. Spectrographic study of voels in stutterers' fluent speech. Journal of Speech and Hearing Research, 1982, 25, LOVE, L. R., & JEFFRESS, L. A. Identification of brief pauses in the fluent speech of stutterers and nonstutterers. Journal of Speech and Hearing Research, 1971, 14, METZ, D. E., ONUFRAK, J., & OGBURN, R. An acoustical analysis of stutterer's speech prior to and at the termination of therapy.journal of Fluency Disorders, 1979, 4, RUNYAN, C., & ADAMS, M. R. Perceptual study of"successfully therapeutized" stutterers.journal of Fluency Disorders, 1978, 3, SANDER, E. K. Reliability of the Ioa Speech Disfluency Test. Journal of Speech and Hearing Disorders, Monograph Supplement, 1961, 7, VAN RIPER, C. The treatment of stuttering. Engleood Cliffs, NJ. Prentice-Hall, WINGATE, M. E. Sound and pattern in "artificial" fluency.journal of Speech and Hearing Research, 1969, 12, WINGATE, M. E. Effect on stuttering of changes in audition. Journal of Speech and Hearing Research, 1970, 13, ZIMMERMANN, G. Artieulatory dynamics of "fluent" utterances of stutterers and nonstutterers.journal of Speech and Hearing Research, 1980, 23, ZIMMERMANN, G., SMITH, A., & HANLEY, J. Stuttering: In need of a unifying conceptual frameork. Journal of Speech and Hearing Research, 1981, 24, Received February 17, 1982 Accepted March 23, 1983 Requests for reprints should be sent to Dale Evan Metz, Department of Communication Research, National Technical Institute for the Deaf, Rochester Institute of Technology, 1 Lomb Memorial Drive, Rochester, NY

7 Acoustic Analysis of Stutterers' Fluent Speech before and after Therapy Dale Evan Metz, Vincent J. Samar, and Pat Richard Sacco J Speech Hear Res 1983;26; This information is current as of April 24, 2011 This article, along ith updated information and services, is located on the World Wide Web at: