JOURNAL

OF COMMUNICATION

DISORDERS

THE EFFECT OF SELECTED THE INTELLIGIBILITY BARBARA Doctoral

G. PARKHURST

11(1978), 24%256

249

PROSODIC ERRORS ON OF DEAF SPEECH AND HARRY LEVITT

Program in Speech and Hearing Sciences, The City University of New York Graduate School and University Center, 33 West 42nd Street, New York, New York 10036

A trained speech pathologist listened to recordings of 40 deaf children, each reading 15 sentences and recorded instances of four types of prosodic errors: adventitious sounds, excessive phoneme duration, pitch breaks, and pauses. The scores were correlated with previously obtained intelligibility scores. It was found that adventitious sounds, except those obviously operating as transitional sounds to ease the movement from one place of articulation to another, had the greatest negative effect on intelligibility, followed by very long duration and pitch breaks. Medium and long pauses, long duration, prolonged closure (i.e., long duration plosives), and adventitious transitional sounds had a lesser negative effect. Short pauses had a small positive effect on intelligibility. The method used to identify the errors was one that can hopefully be evolved into an evaluative instrument for practical use in schools for the deaf.

Introduction In order to develop more effective speech training procedures for deaf children, it is necessary to know how their speech deviates from that of normally hearing children and the effect on intelligibility of the various errors and abnormal speech patterns that are typical of deaf speech. Errors involving individual speech phonemes, i.e., segmental errors, have been studied in depth by a number of researchers (Hudgins and Numbers, 1942; Nober, 1967; Markides, 1970; Smith, 1972, 1975a; Monsen, 1977). These researchers have reported similar, but not identical, patterns to the segmental errors produced by deaf children. Although the relative effect of the various individual error types on intelligibility is not yet fully understood (Levitt, 1977), it is clear from the work of Smith (1972) and others that, in statistical terms, there is a high negative correlation between the frequency of segmental errors and intelligibility, i.e., the higher the incidence of segmental errors, the poorer the intelligibility of the speech, on the average. In addition to segmental errors, there are errors of prosody (e.g., errors of intonation, stress, and/or phrasing), as well as other deviant suprasegmental characteristics, such as abnormal voice quality, hyper- or hyponasality, inappropriate average pitch, and improper control of voicing, to cite a few (Nickerson, 1975; Levitt et al., 1974). Of these, errors of duration and timing have received the greatest attention, partly because these errors are perceptually prominent, and also because improved timing can be obtained with good training. The effect of improved timing on intelligibility has met with mixed results. John and CJElsevier North-Holland,

Inc.,

1978

0021-9924/78/001

l-0249$01.25

250

BARBARA G. PARKHURST and HARRY LEVITI

Howarth (1965) obtained significant improvement in intelligibility, whereas Boothroyd et al. (1974) did not obtain comparable improvements. In a related study, Osberger (1977) corrected timing errors by computer processing of recorded speech samples. The results showed marked individual differences. For certain children, the correction of timing errors improved intelligibility; for other children the correction of timing errors had little effect on intelligibility. Another major source of suprasegmental problems in the speech of the deaf is that of inappropriate pitch. Pitch problems can take several forms, ranging from average pitch too high (or too low), to improper control or lack of control of phonation. Of these error types, sudden breaks in voicing and other evidences of inadequate control of phonation appear to be more highly correlated with poor intelligibility than the more common problem of inappropriate average pitch levels (McGarr and Osberger, 1977, this issue; Levitt et al., 1974). The occurrence of adventitious sounds in the speech of the deaf was reported by Hudgins and Numbers (1942) and other early researchers. Smith (1975b) has shown that these sounds follow distinct patterns depending on the articulatory characteristics of the surrounding phonemes. Although the effect of adventitious sounds on intelligibility has yet to be clearly documented, it is to be expected that these sounds would have a deleterious effect in that they clearly break the flow of speech. Because of the importance of individual differences, particularly in the development of remedial speech training programs, it is essential that the speech teacher have access to reliable methods of evaluating the speech of the deaf. Research is currently in progress on developing improved instrumental procedures for assessing the speech of the deaf (Nickerson and Stevens, 1973; Levitt et al., 1974; Fourcin and Abberton, 197 1; Monsen, 1977). Although good progress on improved instrumentation is being made, it would be a mistake to rely entirely on instrumental methods. The considerable difficulties encountered in the field of automatic speech recognition attest to the remarkable qualities of the human ear in identifying speech sounds and, for the problem under discussion, in identifying errors of speech. It is anticipated that with time, much improved methods of speech assessment will be developed, which will combine instrumental techniques of proven reliability and accuracy with those auditory judgments for which more effective instrumental procedures have yet to be developed. Although subjective judgments typically exhibit a good-deal of variability, it is nevertheless possible to use such judgments in a systematic way to develop practical speech assessment methods. Stewart (1968) discusses the value of the analysis of deaf speech by ear and the need for developing practical procedures for systematic auditory analysis. The present study is part of a larger research project, the overall thrust of which is to develop a practical procedure whereby a speech teacher or speech pathologist can evaluate the speech of a deaf child in a school setting where there

EFFECT OF PROSODIC

ERRORS

ON INTELLIGIBILITY

251

is no access to sophisticated equipment for the acoustic or articulatory analysis of speech. The specific aim of this study was to examine the relationship between those distortions in the prosody of deaf speech that can be judged by ear and the overall intelligibility of such speech. It is hoped that the information so obtained would be of value in determining those errors or distortions of deaf speech that are correlated with a breakdown in intelligibility, and consequently, which speech problems to work with as part of a remedial speech-training program. Method A trained speech pathologist listened to recordings of 40 deaf children, each reading 15 sentences. These recordings were taken from the corpus of material obtained by Smith (1972) and, thus, measures of intelligibility and broad phonetic transcriptions of each recording were available. The children had all been attending the same school for the deaf for at least 3 years, were of normal intelligence, and had no apparent physical or psychological anomalies, other than deafness. All were congenitally deaf, but had some measurable audiometric response at least to 750 Hz in the better ear. The children were equally divided into two groups, one ranging in age from 8 to 10 years, the other from 13 to 15 years. The speech pathologist listened to the recordings with a broad phonetic transcript of a standard production of the sentences in front of her. She made judgments as to the occurrence of the four categories of prosodic error found in Table 1. These prosodic distortions were chosen for analysis because of their relative importance as reported in the literature and because they were of a form that could be detected by ear. Each of the distortions listed in Table 1 are defined below. Adventitious

Sounds

Adventitious-repeat was defined as an interjected sound or group of sounds that appeared to the listener to be a repetition of a preceding phoneme or group of phonemes or else seemed to be a false start on an utterance that was next repeated in a more complete form. Adventitious-transition was defined as an interjected sound that was produced during the articulatory movement in going from one phoneme to the next. Adventitious-other refers to interjected sounds not easily classified into either of the above two categories. Clearly, more categories of adventitious sounds could be devised and have indeed been used in the past (Smith, 1972, 1975b). However, the present study was concerned with evolving a system to identify features that could be evaluated by a speech pathologist or speech teacher routinely testing many children and, therefore, broad categories were used.

BARBARA

252

G. PARKHURST

TABLE List of prosodic distortions analyzed ----_---__---~_-Adventitious sounds (interjected sounds) 1, Adventitious-repeat 2. Adventitious-transitional 3 Adventitious-other Excessive duration 1. Duration-long 2. Duration-very long 3. Duration-plosive, prolonged

closure

and HARRY

LEVITT

1 Multiple linear-regression

-0.0151 -0.0042 -0.0122

-0.0037 -0.0111 -0.0037

Pitch breaks

--0.0092

Pause5 I Pause-short 2, Pause-medium 3. Pause-long

+0.0033 -0.0043 -0.0037

Excessive

slope

Duration

was defined as a sound that was perceived as Excessive phoneme duration either (1) longer or (2) much longer than normal. was defined as the inappropriate proExcessive Duration-plosive closure longation of the closure period prior to the release in plosive sounds. A sep.arate category has been used here since plosives are not lengthened in the same manner as vowels and continuant consonants. Pitch Breaks A pitch break was defined as a sudden and abrupt change in pitch within a phoneme, within a word, or from one word to the next. Pauses Pause-short was defined simply as a pause in which the listener could count to one (or tap a pencil once) between the end of one phoneme and the beginning of another. was defined as a pause in which the listener could count to Pause-medium two. was defined as a pausein which the listener could count to three or Pause-long more. Level recordings on a sample of 30 sentences indicated that in physical terms these pause times were roughly: short, 200-450 msec; medium, 450-1200 msec; and long, over 1 second.

EFFECT OF PROSODIC ERRORS ON INTELLIGIBILITY

253

Results A multiple linear regression (Nie et al., 1975) was performed relating the intelligibility to the prosodic distortions that were judged to occur in each child’s speech. The measure of intelligibility was that previously obtained by Smith (1972), in which listeners not familiar with the speech of the deaf listened to the recordings and wrote down what they believed was said. The proportion of words correctly identified was used as the measure of intelligibility. The slope of each variable in the multiple linear regression is shown on the right-hand side of Table 1. The slopes, with one exception, are all negative, indicating that an increase in the frequency of the prosodic distortion corresponds to a decrease in intelligibility. The unit of slope is percent intelligibility/ frequency of occurrence of prosodic distortion. The prosodic distortions that showed the greatest negative regression with intelligibility were, in order of relative magnitude, adventitious sounds--repeat (-0.0151), adventitious sounds-other (-0.0122), excessive duration-very long (-0.011 l), and pitch breaks (-0.0092). The numbers in parentheses are the slopes. The slopes of these variables were 2 to 4 times larger than those of the remaining variables, all of which had slopes on the order of -0.004. The one prosodic distortion that was found to have a positive correlation with intelligibility was that of short pauses. The slope of this variable in the multiple regression analysis was small (0.0033), but positive, indicating that short pauses may be helpful in improving intelligibility. It should be noted that the words in the sample were mostly short monosyllabic words and that the pauses typically occurred at word boundaries. Discussion The judged prosodic distortions showing significant negative regression with intelligibility have two important properties. First, with one exception, they correlated negatively with intelligibility, which suggests that correction of these distortions would result in improved intelligibility. Second, the judgments of these distortions were sufficiently repeatable to form a systematic pattern. It is conceivable that one or more of the variables found to have a low correlation with intelligibility may, in fact, have a large effect on intelligibility, but could not be judged reliably. For this reason, the variables chosen for consideration in this study required no more than gross subjective judgments. It is an open question whether prosodic distortions requiring subtle perceptual judgments would have more of an effect on intelligibility than the gross effects considered here. The effect of more subtle prosodic distortions will be studied at a later stage after the effects of the perceptually most prominent variables have been considered.

BARBARA

254

G. PARKHURST

and HARRY

LEVITT

0

0 0 0 0

0°0 0 0

(SlNlOd a3mia3tld

39W1N33H3d) I-a3iw3sao -wnals3tl

I

EFFECT OF PROSODIC

ERRORS

ON INTELLIGIBILITY

255

The results of the analysis show that three factors correlated highly with reduction in intelligibility. These are unexpected adventitious sounds, i.e., adventitious sounds other than those linked to normal articulatory movements, ver$ long prolongation of individual sounds, and unexpected changes in pitch. In contrast, those breaks in the flow of speech produced by excessive or prolonged pauses, including prolonged closure in plosives, appears to have a secondary effect (in a correlational sense) on intelligibility and short pauses may have actually aided intelligibility. One interpretation of this result is that although excessive and/or prolonged pauses are perceptually very prominent, they nevertheless provide the listener with additional time to process the distorted speech that has been heard. In the study by Osberger (1977), it was found that the elimination of inappropriate pauses sometimes reduced, rather than improved, intelligibility. Prosodic distortions represent only a portion of the variables influencing intelligibility. It is already well established that segmental errors play an important role in reducing intelligibility. To what extent do prosodic and segmental errors interact in their negative effect on intelligibility? Figure 1 shows the difference between the observed intelligibility scores and that predicted from the multiple linear regression of the judged prosodic distortions as a function of the segmental error rate. The latter is expressed as a percentage of the intended phonemes judged to have one or more major segmental errors (i.e., omission, substitution, severe distortion). The plot shows a marked correlation between the residual and the relative frequency of segmental errors. Those children with relatively few segmental errors were found to have more intelligible speech than predicted from the prosodic distortions alone. Similarly, most, but not all, of the children having a high rate of segmental errors had intelligibility scores lower than would be predicted from prosodic distortions alone. Of special interest is a small group of children, represented by the points lying roughly on a horizontal line on the right-hand side of the diagram. For these children, the prediction of intelligibility based on prosodic distortions was virtually independent of the segmental error rate. An examination of the children in this group indicated that the majority of the children were in the younger age group and they all exhibited severe prosodic distortions in their speech. So much so, that their speech was virtually unintelligible. For these children, prosodic, rather than segmental distortions, appeared to be the dominant problem. References Boothroyd, A., Nickerson, R., Stevens, K. Temporal patterns in the speech of the deaf-A study in remedial training. Research Report S.A.R. # 15, Clarence Barron Research Dept., Clarke School for the Deaf, Northampton, Mass., 1974. Fourcin, A. J., Abberton, E. First applications of a new laryngograph. Med. Illus., 1971. 21 (3), 172-182.

256

BARBARA

G. PARKHURST

and HARRY LEVlTr

Hudgins, C. V., Numbers, G. C. An investigation of the intelligibility of the speech of the deaf. Gener. Psychol. Monogr., 1942, 25, 289-392. John, J. E., Howarth, J. N. The effect of time distortion on the intelligibility of deaf children’s speech. Lang. Speech, 1965, 8, 2127-134. Levitt, H. Conference on Sensory Aids for the Hearing Impaired. Washington, D.C., May 1977. Levitt, H., Smith, C. R., Stromberg, H. Acoustic, articulatory and perceptual characteristics of the speech of deaf children. Proc. Speech Commun. Seminar, Stockholm, Sweden, 1974. Markides, A. The speech of deaf and partially hearing children with specific reference to factors affecting intelligibility. Br. J. Dis. Commun., 1970, 5, 126140. McGarr, N. S., Osberger, M. I. Pitch deviancy and intelligibility of deaf speech (this issue). McGinnis, M. Aphasic children. Washiqgton, D.C.: A. G. Bell Association for the Deaf, 1963. Monsen, R. B. Measuring how well deaf children speak. Presented at the Annual Convention of the American Speech and Hearing Association, Chicago, Ill., Nov. 2, 1977. Myklebust, M. Auditory disorders in children. New York: Grune & Stratton, 1954. Nickerson, R. S. Characteristics of the speech of deaf persons. Volta Rev., 1975, 77, 342-362. Nickerson, R. S., Stevens, K. N. Teaching speech to the deaf: can a computer help? IEEE Trans. Audio Elecrroacousr., 1973, AU-21 (5), 445-455. Nie, N. H., Hull, C. H., Jenkins, J. G., Steinbrenner, K., Bent, D. H. Sratisrical package for the social sciences, 2nd ed. New York: McGraw-Hill, 1975. Nober, E. H. Articulation of the deaf. Excepr. Child., 1967, 33, 61 l-621, Osberger, M. J. Personal communication, 1977. Smith, C. R. Residual hearing and speech production in deaf children. Doctoral dissertation, City University of New York, 1972. Smith, C. R. Residual hearing and speech production in deaf children. J. Speech Hear. Res. 1975a. 18, 795811. Smith, C. R. Interjected sounds in deaf children’s speech. .I. Commun. Dis., 1975b, 123-128. Stewart, R. B. By ear alone. Am. Ann. DeaJ 1968, 113, 147-155.

The effect of selected prosodic errors on the intelligibility of deaf speech.

JOURNAL OF COMMUNICATION DISORDERS THE EFFECT OF SELECTED THE INTELLIGIBILITY BARBARA Doctoral G. PARKHURST 11(1978), 24%256 249 PROSODIC ERROR...
479KB Sizes 0 Downloads 0 Views