THE UNIVERSITY OF MICHIGAN COLLEGE OF LITERATURE, SCIENCE, AND THE ARTS Department of Speech and Speech Clinic Final Report September 1, 1959 August 31, 1965 PHONETIC CHARACTERISTICS OF DYSARTHRIA Ronald S, T ikofsky ORA.i P..roject 06066 sponsored by: DEPARTMENT OF HEALTH, EDUCATION, AND WELFARE U. S PUtBLIC HEALTH SERVICE NATIONAL INSTITUTES OF HEALTH GRANT NO NWB-02221-05 BETHESDA, MARYLAND administered through, OFFICE OF RESEARCH ADMINISTRATION ANN ARBOR August 1965

Acknowledgements The principal investigator wishes to take this opportunity to publicly acknowledge the significant contributions made to the pursuit of this research by persons directly and indirectly concerned with the program. Perhaps the most significant contribution to the program was made by Dr. Ilse Lehiste, who served as co-principal investigator from the inception of the project through 1962 when she left the University of Michigan. Although not formally associated with the project after 1962, she continued her work on the acoustic-phonetic analyses and produced a significant monograph on dysarthria. Thanks must also go to Professor Gordon E. Peterson, who graciously granted us the use of the facilities of the Communications Sciences Laboratory, which included space, equipment and staff, as well as his own time for many hours of consultation. One member of the CSL staff, Dr. John Hanne, deserves special mention for his work on writing the computor program to analyze listener responses. High praise and appreciation go to the research assistants who carried out the day-to-day work on the project. Particular thanks go to Mr. Theodore Glattke and Mrs. iii

Rita Tikofsky whose tireless efforst went well beyond the call of duty. Important contributions to data collection and analyses were made by Mr. Gale Reynolds and Misses Judith Gallatin and Nancy Reed. Special thanks go to the follow for their assistance in obtaining dysarthric subjects for this research: Dr. R.N. DeJong, Chairman, Dept. of Neurology, The University of Michigan Medical Center Dr. D.E, Morely, The University of Michigan Speech Clinic Dr. R.F. Hejna, Chief, Speech Pathology-Audiology, Veterans Administration Hospital, Ann Arbor, Michigan Dr. P.S, Perry, Audiology and Speech Clinic, Veterans Administration Hospital, Denver, Colorado Miss Josephine Simonson, Speech Pathologist, Henry Ford Hospital, Detroit, Michigan Dr. Thomas Sphipp, Research Speech Pathologist, Veterans Administration Hospital, San Francisco, Calif. Mrs. Milfred McKeown, Chief, Aphasia Unit, Veterans Administration Hospital, Long Beach, California. Mrs. Aleen Agranowitz, Lukewood Speech Clinic, Lakewood, California Charles Johnson, Language Retaining Therapist, Veterans Administration Center, Dayton, Ohio The problems of handling many of the small administrative and scheduling problems and the typing of manuscripts fell to the secretaries who worked with us. The contributions of Miss Caroline Schonschack, who typed the manuscript for Dr. Lehistes manuscript, and Miss Gloria Bowles, who prepared the manuscript for this report under severe time pressure, deserves particular notice. iv

Table of Contents Page Introduction 1 Methods of Procedure 4 Results 30 Summary of Results 66 List of Publications and Papers Presented at Meetings 70 Appendix Io Instructions to Listeners in Intelligibility Study I Al IIo Flow Diagram for Computer Program to Analyze Listener Responses to CNC Words and Initial and Final Confusion Matrices for Dysarthrics A5 III. Preliminary Report on the Study of the Acoustic Characteristics of Dysarthric Speech All IV, An Acoustic Description of Dysarthric Speech A21 V. Phonetic Characteristics of Dysarthria A39 VI. A Study of the Intelligibility of Dysarthric Speech A57 VII. Intelligibility of Dysarthric Speech A71 VIII. Listener Confusions of Dysarthric Speech A87 V

Introduction Much of the literature on dysarthria and dysarthric speech is characterized by the absence of empirical data. Coriclusions.by.a wide range of investigators concerning the relation of the dysarthrics' speech to locus and type of lesion producing the symptoms have been based on gross clinical evaluations of the speech signals. The neurological literature has been equally imprecise with respect to several important parameters which are necessary in order that accurate localization can be achieved. Analysis of the speech has been equally gross, with few operational definitions of either the speech pattern ssortthet-term us d. to- indicate.fthe.'degree of..speech tmpairment, The basic objectives of the research being presented in this report were to determine the possibility of providing better descriptions of dysarthric speech. Two points of view were surveyed. On the one hand, we sought to describe the acoustic-phonetic characteristics of this speech impairment by employing analytic procedures common to experimental phonetics and speech science. Our second basic purpose was to attempt to devise methods by which the degree of speech -1

impairment in dysarthria could be quantitatively and objectively evaluated. It was our hope that data obtained from such approaches might correlate Vwith each other, and with clinical information relating the neurologial basis of the speech problems as well as the assessment of dysarthria as determined by speech therapists. While this was not the major goal of the research program, it was always prominent in our evaluation and description of the experiments and their subsequent interpretation. Another, and perhaps less overt objective, was to demonstrate that the techniques of experimental phonetics and intelligibility testing: could make an important contribution to the clinicisan and open the way for others to apply the procedures employed in research in other types of speech impairments. Some of...the particulat expectations and objectves of the reseatchliprogram~ -such as;ipectrographic analysis of sentence utterances, analysis of supersegmental features, measures of phonation and respiration during speech and a scale to assess intelligibility, were not achieved. But the major purposes were accomplished. The results obtained from the research that was carried out did fulfill our basic goals. -2

The five major objectives of the research program were: 1. To provide a detailed description of the acousticphonetic characteristics of the speech 2. To compare spectrographic data obtained from dysarthrics to data from normal speakers; 3. To compare the acoustic characteristics of the steady state of vowels contained in wbrds recorded by dysarthric and normal speakers; 4. To compare the intelligibility of dysarthric and normal speech by means of listener responses to words and sentences recorded by both groups of speakers; 5. To examine the power of the words used to discriminate among dysarthrics as a means of estimating the range of reduction in intelligibility. -3

-1 Methods of Procedure As was described above, two approaches to the systematic study of dysarthric speech were employed in this research. One approach was acoustic-phonetic and the other the assessment of intelligibility. However, materials and speakers were identical in the first series of experiments. This provided an appropriate means by which to examine the relations between acousticphonetic and intelligibility measures derived from the same sources. A second grop of dysarthric speakers was used to yield data on revised measures of intelligibility. The "new" materials for assessing intelligibility were derived from the original set of data. Results of the studies will be described in the following section. This section describes the materials, subjects, equipment and general approach used to obtain the data discussed in the results section. Material Recorded by Speakers: In the initial series speakers recorded 160 words and 40 brief sentences. were organized into three lists: 50 monosyllabic CNC Syllable Nucleus-Consonant) words; 60 Cluster (to be below) words and 50 Spondee words. -4 of studies Tests words (Consonantdescribed

The CNC words used contained an initial consonant phoneme, one of the 15 stressed syllable nuclei of American English, and a final consonant phoneme. Words of this type have been shown to be useful in the study of speech intelligibility. The 50 words used for the acousticphonetic analysis and the initial intelligibility studies were taken from the lists of CNC words compiled by G.E. Peterson and I. Lehiste. Table 1 presents the 50 CNC words. The 60 cluster words (Table 2) were compiled specifically for this study. These monosyllabic words have initial and/or final consonant clusters. Although not all possible consonantal clusters of American English are contained in the list, the most frequent ones are represented. Consonantal clusters were used to permit an evaluation of the dysarthrics' ability to make rapid articulatory and phonatory shifts. The 50 Spondee words (Table 3 ) were selected randomly from lists I and II of Auditory Test No. 9 of the Psycho2 Acoustic Laboratpry, Harvaird.University. Spondees provided a sample of dysarthric speech with polysyllabic words. "Linguistic considerations in the study of speech intelligibility" by I. Lehiste and G.E. Peterson, J. acoust. Soc. Amer. 31: 280-286 (1959); "Revised CNC lists for auditory tests'" b G.E. Peterson and I. Lehiste, J. Speech Hearing Disorders 27: 62-70 (1962 ) CoV. Hudgins et al.: "The development of recorded auditory tests for measuring hearing loss for speech". Laryngoscope 57: 57-89 (1947 )

Table 1. CNC Words kid rig noise gull shine dirge rat four dab dip tone room lake fade leave hut gap keg keen yam well mate work pod with bush house toll faith sheep toss bar pike more vote piece what size rouge j oke Of perch sung head man soon bell jail numb chin life -6

Table 2. Cluster Words wasp bronze twist left forth sleeps planned speaks range field cats hums tests bolt built grove schools spice sticks pride text sketch arch showered shrug fraud charge gloss train bangs thread champ barb queens scrub bounce pinch shank else pact strife dwarf crown part purse tongue clears drop smiled dark harp switched chant sprig flowers fused blush sledge bind cute -7

Table 3. Spondee Words bagpipe footstool scarecrow housework pinball mushroom yardstick duckpond inkwell doorstep whitewash toothbrush airplane daybreak hotdog horseshoe cookbook shipwreck outside washboard cupcake backbone sundown oatmeal hedgehog northwest schoolboy hardware pancake coughdrop blackboard watchword woodchuck headlight wildcat woodwork nutmeg padlock eggplant earthquake workshop lifeboat birthday eardrum playground drawbridge buckwheat starlight armchair platform

Tables 4,5, 6 present the tabulations of the occurrence of some of the important phonetic characteristics of the test words. Table 4 gives the distribution of syllable nuclei among the lists. Table 5 gives the frequency of occurrence of each consonant in the initial position and as a member of an initial cluster, and similar information for consonants in the final.margin is given in Table 6. Forty sentences were constructed to sample dysarthric speech beyond the single word utterance. The sentences recorded by the speakers is given in Table 7. All of the acoustic-phonetic analysis was carried out on the three word lists described above. In addition, recordings of these words were employed to evaluate the intelligibility of dysarthric speech. The forty sentences were also used for intelligibility analysis. Results of the initial intelligibility studies were employed to reduce the test list from 160 to 50 words. This new li-st (.Table 8 ) was derived from an item analysis of listener responses to the original 160 words. The revised list contains 6 CNC's, 24 Clusters and'20 Spondees. A new sample of dysarthrics recorded these words for a followup study. of intelligibility. Subjects. Ten dysarthrics (male ) obtained through the cooperation of the Speech Clinic, Department of Neurology, -9

Table 4. Occurrence of Syllable Nuclei in the Test Words Syllable Nucleus CNC words Cluster wor s Spondee First component Words Second component Total i I ei C a ou U u u r au aI 0I 4 5 5 4 5 4 3 3 4 1 3 3 1 4 1 7 2 7 7 5 9 4 2 3 1 4 5 2 5 6 6 6 6 1 4 2 3 2 3 2 2 6 7 2 3 4 7 3 1 3 3 2 4 1 10 18 15 23 20 18 22 20 10 6 11 10 9 16 2 50 60 50 50 210 -10 -

Table 5 Occurrence of Consonant Phonemes in the Initial Margin Word- In first In second In third initial position in position in position in Total initial cluster initial cluster initial cluster i in Phoneme initial Utter- Post- Utter- Post- Utter- Post- Utter- Post- margin ance junc- ance- Junc- ance- Junc- ance Juncinitial tural initial tural initial tural initial tural b, d P H g p t k J C v V f s f h m n 1 r w W 13 7 2 11 7 7 2 7 3 2 3 2 3 2 2 4 3 2 2 25 15 4 17 12 21 2 3 2 5 4 1 6 4 5 11 4 4 5 5 10 2 2 10 2 1 1 2 1 2 3 2 7 1 3 15 1 3 1 6 1 10 23 8 12 7 4 20 25 22 3 4 10 10 10 4 2 5 1 3 2 - ------- --- - - - - - -

Table 6 Occurrence of Consonant Phonemes in the Final. Margin Word- In first In second In third final position in position in position in Total Phoneme final cluster final cluster final cluster in Utter- Pre- Utter- Pre- Utter- Pre- Utter- Pre- final ance junc- ance junc- ance junc- ance Juncfinal tural final tural final tulral final tural H ro I b d g p P t k iJ v f 3 a 0 z s m n 1 r # 2 9 7 8 11 15 3 2 2 2 1 4 1 1 2 6 5 11 2 9 5 4 4 2 2 6 6 1 1 2 1 7 1 1 4 3 11 2 2 2 3 1 1 3 23 9 14 1 30 28 6 6 2 6 1 6 1 6 10 1 1 1 3 23 9 14 30 28 6 6 2 6 1 6 1 6 10 20 9 25 5 17 29 7 2 3 1 3 1 5 3 1 4 2 11 2 6 14 1 7 7 1 1 1 1 1 1 5

Table 7. Sentence List 1. The wagon moved on well oiled wheels. 2. The fleas hopped on both the cat and the chair. 3. Never buy a blind pig in a bag. 4. March the soldiers past the next hill. 5. A cup of sugar makes sweet fudge. 6. Place a rose bush near the porch steps. 7. George gave his sister a lot of coin. 8. Both lost their lives in the raging storm. 9. We talked of the' side show in the circus. 10. Use a pencil to write the rough draft. 11. He ran half way to the hardware store. 12. Eight-cops visit the new cook. 13. A cute baby is not shy or corss. 14. The clock struck to mark the third period. 15. College girls are full of zip- arid verve. 16. A small creek cut across the field. 17. Boys thrive on rough games and candy. 18. Cars and buses stalled in snow drifts. 19. The set of china hit the floor with a crash. 20. May is a grand season for hikes on the road. -13

Table 7. Sentence Lis'c (-co-nt. 21. He told the queen to mend her harsh manner. 22. The slush lay deep along the street. 23. Seldom have I seen such a dope. 24. A wisp of cloud hung in the blue air. 25. Shave the whiskers and show your own face. 26. Type one is right, the doctor said. 27. A pound of sugar costs more than eggs. 28. The fin was sharp and cut the clear water. 29. The play seems dull and quite stupid. 30. Sail the boat to stop it from sinking. 31. The term ended in late June that year. 32. The team fell into a mean humor then. 33. The hoary rock was huge and rose high. 34. A tusk is used to make costly gifts. 35. Ten pins were set in order. 36. The bill was paid every third week. 37. Oak is strong and also gives shade. 38. Cats and dogs each hate the otherO 39~ The park stretched from the pond to the ancient hous:e. 40. Rooms to let go begging this spring. -14

Table 8. Revised 50 item word list. blush champ sledge grove bind sleeps barb showered chant bounce train shank sketch j oke spice thread bush duckpond with crown playground drawbridge twist shrug cookbook fused scrub tongue earthquake charge more cute scarecrow dwarf daybreak inkwell shipwreck buckwheat sheep woodwork starlight platform mushroom schoolboy washboard eggplant wildcat northwest lifeboat man -15

both of the University of Michigan and the Ann Arbor Veteran's Administration Hospital, served as speakers. Each dysarthric was given a neurologic examination by staff members of the Department of Neurology. Clinical evaluation of each subjects speech based on a recording of conversation as well as the test materials was carried out by a member of the Speech Clinic staff. He provided a qualitative judgement of the severity of the dysarthric impairment. Recordings made by each of these speakers provided the material for acoustic-phonetic analysis and initial intelligibility studies. A second group of 20 dysarthric speakers recorded the revised 50 word list;for the second intelligibility study. These speakers were recorded in several. centers and hospitals around the country. The essential clinical and biographical data for all dysarthric speakers is given in Table 9. Data from both groups of dysarthrics were employed in the analysis of the revised word list. Ten normal speakers were selected as a control group for the original sample of dysarthrics. An attempt was made to select normals that approximated the dysarthric group in age, education and occupational status. Table 10 presents the biographical data for these speakers. In the first series of intelligibility studies, 30 different listeners responded to each talker. Ten different listeners responded to each talker's recording of the 40 sentences. A total of 600 listeners participated in the ex-16

Table 9. Biographical and clinical data for dysarthric subjects. Time between Clinician's Etiology Related Patient Age Sex Place of Birth onset and judgment of speech finding recording degree of dys- as judged by arthric im- the clinician pairment where recording made (2) O.B. 57 M Wilburton, Okla. 7 yrs. Moderate CVA Paraphrasia, motor aphasia (2) F.B. 65 M San Francisco, Calif. 9 yrs. Moderate CVA Residuals of motor aphasia H (2) R.B. 54 M Morhall, Maryland (2) J.B. 43 M McAllister, Okla. 6 mos. 1 yr. Mild Mild Thrombosis None Throm- Amnesic aphasia, bosis Formulation and spelling loss, acalculia (1) F.B., Jr. 42 M Michigan 4 yrs. Moderate Posttraumatic brain and cerebellar deficit None *(1) J.F. B. 65 M OklahomaMichigan 2 yrs. Very Progressive severe bulbar palsy Loss of labial and lingual function *Used in spectrographic analysis only. **These are findings in addition to dysarthria and when listened to by experimenter judged not to interfere with participation in experiment.

Table 9. (cont.) Biographical and clinical data for dysarthric subjects. Patient Age Sex Place of Birth Time between onset and recording Clinician's judgment of degree of dysarthric impairment Etiology Related speech finding as judged by the clinician where recording made - -- (2) F.C. 41 M Mt. Harris, Colo. 6 mos. Moderate Cerebellar Atrophy Ataxia I 0oD! (1) O.D. 68 M Michigan 3 yrs. Severe CVA embolism None (2) R.G. (1) G.H. 29 M Denver, Colorado 35 M Battle Creek, Michigan 11 yrs. 2 yrs. Mild Severe Multiple None Sclerosis known PostEncephalitic None (2) O.H. (1) D.H. 42 M South Dakota 38 M Yonkers, New York 13 yrs. 3 yrs. Moderate Moderately severe CVA Closed head injury None None known (2) R.H. 24 M Baker, Oregon 5 mos. Very mild Auto accident Ataxia (2) R.Hen. 64 M Arkansas 4 yrs. Severe CVA Motor aphasia

Table 9. (cont.) Biographical and clinical data for dysarthric subjects. Patient Age Sex Place of Birth Time between onset and recording Clinician's judgment of degree of dysarthric impairment Etiology Related speech finding as judged by the clinician where recording made (1) L.K. 20 M North Dakota 4 yrs. Moderately severe Pbsttraumatic encephalopathy None known (2) C.K. 21 M Detroit, Michigan 16 yrs. (?) Mild Huntington's chorea o Mild choreatic speech (2) E.La. (2) E. Ly 50 M El Paso, Texas 27 M Toledo, Ohio 4 yrs. 4 yrs. Mild Mild Trauma Traumarontal lesion Amnesic Aphasia None (2) H.M. (2) J.M. 44 M Denver, Colorado 25 M Longbeach, Calif. 4 yrs. 8 yrs. Mild Severe Encephalitis Trauma Ataxia None (2) D.O. 30 M Mass. 9 yrs. Severe TraumaFrontal None (2) F.P. 46 M Montana 24 yrs. Very mild MS None known (2) J.P. 56 M Los Angeles, Calif. 3 yrs. Mild CVA Amnesic aphasia Oral formulation loss

Table 9. (cont.) Biographical and clinical data for dysarthric subjects. I O! Patient Age Sex Place of Birth Time between Clinician's Etiology Related onset and re- judgment of speech finding cording degree of dys- as judged by arthric im- the clinician pairment where recording made (1) B.S. 38 M Michigan 33 yrs. Moderate Frederick's None known ataxia (1) C.S. 25 M Michigan 5 yrs. Moderate Posttrau- Moderate dysmatic upper phonia brain stem damage (2) D.S. 40 M Arkansas 5 yrs. Moderate M.S. Mid-brain None known (i) F.S. 47 M Wyandotte, 26 yrs. Moderately Unilateral None known Michigan severe Parkinson's disease (2) H.T. 53 M Missouri 3 yrs. Severe Embolus to None basi.lar artery system (2) G.W. 16 F Michigan 1 yr. Moderate-Mild Cerebellar lesion Ataxia (1) A.Z. 52 M Indiana 4 yrs. Severe CVA None known 1) Used in Intelligibility S-udy I. (2) Used in Intelligibility Study II.

Talble 10.. Biographical data for the normal subjects. Subject Number Age Education Occupation IN 28 B.A. Teacher-coach 2N 20 College: 2 yrs. College Student 3N 48 Ph.D. College Teacher 4N 36 High School Mail Carrier 5N 51 High School Barber 6N 26 M.A. Graduate Student 7N 59 High School Group Leader 8N 50 Ph.D. College Teacher 9N 36 9th Grade Farmer 1ON 63 High School Farmer -21

periments. The listeners were students in speech courses at The University of Michigan. Three hundred additional listeners served in the second intelligibility study. Analyses of the reliability of response on a word by word basis suggested that 15 listeners per talker be used in the study of intelligibility based on the revised word list. Recording Procedures. For the first series of experiments, (i.e. acoustic-phonetic analyses and initial intelligibility studies ) all recordings were made with an Ampex Tape Recorder, Model 350, and Altec Microphone, Model 660. Whenever possible, the recordings were made in the soundproof recording room of the Communication Sciences Laboratory. In two instances the recordings were made in a sound treated room provided by the Department of Neurology at the University of Michigan Medical Center. Speakers repeated the test words after the experimenter, and read the sentences from a typewritten sheet. The recordings of the second group of dysarthrics were not made under similar conditions. The recording equipment employed was that available at the various centers from which speakers were recruited. In addition, there was some variation in the amount of background noise present in the recordings. Some potential speakers were not used because of background noise or poor recording quality. Tape speed in all cases was 7 1/2 ips. -22

Acoustic-Phonetic Analysis. Recordings made by all speakers were subjected to spectrographic analysis. The Bell Telephone Laboratoriest Model D Sound Spectrogranh at the Communication Sciences Laboratory of the University of Michigan was employed for the analysis. In addition to the broad-band spectrograms made of all the words recorded, narrow-band spectrograms and continuous amplitude displays were made of a selected set of utterances. Approximately 3000 spectrograms were examined for this aspect of the research program. Two analyses of the spectrographic data were carried out. The first, executed by Dr. Ilse Lehiste,3 was concerned with providing a detailed acoustic-phonetic analysis of dysarthric speech. To achieve this end the following measures were made on each spectrogram: 1. The center frequency of the first three formants at the target or steady state of the syllable nucleus. 2. The duration of the syllable nucleus. 3. Formant positions at the point of transition from the initial consonant to the syllable nucleus and at the point of transition from the syllable nucleus to the final consonant. 4. The duration of the initial and final consonants. 5. The formant positions within the resonant consonants. 6. The frequency of energy concentrations in fricative consonants and in the releases of plosives. Co-investigator while at The University of Michigan. Now at Ohio State University. -23

Complete details of the procedures and standards for measurement and interpretation are given in the monograph entitled, "Some Acoustic Characteristics of Dysarthric Speech," 4 by Dr. Lehiste. The second analysis was concerned only with comparing and studying the vowel production of the dysarthrics and their controls. Many of the measures on the vowels for the analysis carried out by Lehiste were used for this study. However, the spectrograms for vowel production'f all the normal controls were examined. This analysis did not require all of the measures described above. Only the center frequency of the first two formants at the target or steady state of the syllable nucleus were employed. A gross statistical analysis of the formant frequencies was carried out. The results of this analysis are given in the results section of this report. Most of the work for the vowel study was carried out by Rita Tikofsky, who also assisted Dr. Lehiste in her spectrographic analyses. 4 The monograph was completed as part of the research carried out for this project and is submitted along with the final progress report. The complete reference is: Lehiste, Ilse, Some Acoustic Characteristics of Dysarthric Speech, Bibliotheca Phonetica, Fasc.2, (1965). -24

Intelligibility analysis. Four studies of dysarthric intelligibility were carried out in the course of the research program. While the information varied in each study, the approach to data collection and analysis did not vary markedly from experiment to experiment. For the first intelligibility experiment sets of listening tapes containing each of the three word lists was prepared from the original tapes recorded by each speaker. Three listening tapes were prepared'for each listener. Instructions'-i-to the.-listeners were recorded on these tapes. Each tape had one of three word list orders: CNC-Clusters-Spondees; Clusters-Spondees-CNC; and Spondees-CNC-Clusters. Using this pattern no list appeared in the same position on any two tapes for a given speaker. The listening tapes were played to listeners through a single TDH39 earphone at a comfortable intensity level. The opposite ear was covered with a dummy earphone. Generally, five listeners were present at each group listening session. Ten different listeners responded to each listening tape for each speaker. Listeners were instructed'to: write -the word they thought they heard through the earphone on a record sheet. If they did not understand the utterance, they were asked to draw a line alongside the appropriate -25

number on the answer sheet. The instructions given to the listeners on the tape recording are given in Appendix I. All listening sessions were held in quiet rooms with minimal amounts of outside or ambient noise. Data obtained from the experiment described above was used for two of the four intelligibility studies. For the first evaluation of intelligibility only gross measures were made. A speakerts intelligibility was defined as the number of words correctly identified by the listeners. Four separate intelligibility scores were derived: (a) total intelligibility, the total number of correct responses made by all listeners for a given speaker for the three word lists; (b) total order intelligibility, the total number of correct responses made by all the listeners for a given speaker for each order of word list presentation; (c ) list intelligibility, the total number of correct responses made by all listeners for a given speaker for each word list, without regard to the position of the list on the listening tapes; and (d) list-by-order intelligibility, the total number of correct responses made by all of the listeners for a given speaker on a word list occurring in a particular order on the listening tape. These scores were subjected to statistical analyses. Results of the analyses are described in the next section -26

of this report and in the papers presented at professional meetings. These papers are found in the appendices of this report. A computer program was developed for the second part of the first intelligibility program. This aspect of the research on intelligibility was concerned with the errors made by listeners in response to the CNC words. The number as well as type of consonanantal errors made for initial and final consonants for each speaker was tabulated. In addition, these counts were summed across all speakers. A copy of the final confusion matrices for dysarthric and normal speakers along with the flow diagram and key punch code appear as Appendix II. Results of this analysis will be described later. An analysis of listener responses to sentences read by dysarthrics and normals was carried out. Speakers in this experiment were the same as those used to obtain the single word intelligibility data described above. Ten different listeners responded to each tape under the same conditions described for the initial single word intelligibility study. Listeners were told to listen to the complete sentence (each sentence was presented twice in rapid succession ) and then to write what they thought they heard. Instructions indicated that the stimuli -27

would be complete sentences, not just word-phrases. A computer program to score and tabulate the errors for each speaker and all speakers in the two groups, dysarthrics and normals, was written. The computer output included the word substitutions, total word errors and the rank order of the errors. A sentence intelligibility score for each speaker was computedo This score represented the percent correct responses and was obtained by dividing the number of errors made by the total number of possible errors. The total number of possible errors equaled the number of words contained in all sentences (312) multiplied by the number of listeners per speaker. (10)@ Thus, there were a possible 3120 responses/ speaker. Intelligibility curves for each speaker as well as the relative intelligibility of the stimulus words for the two groups of speakers were studied. These scoring and analytic procedures served as the source for interpreting the intelligibility of dysarthrics using short sequences of connnected speech. The first three investigations of measures of dysarthric intelligibility used the same population of normal and dysarthric speakers. In part, they were concerned with comparing the dysarthric and normal. It was evident, as will be seen in more detail in the Results section, that the two groups in no way overlapped on the measures obtained. 28 =28=

However, the results of the first word intelligibility study suggested that further work in this area on the dysarthric was warranted. As a first step in the preparation of a follow-up study of dysarthric intelligibility, an item analysis was corried on the 160 words which made up the stimulus material used in the initial studies. Our goal was to reduce the 160 word list to the 50 words which might best be used in the follow-up experiments. A standard and available computer program was used on the IBM 7090 computer to carry out the analysis. The resultant 50 —word list contained the most reliable items (minimum reliability of.550 with a range of.550 to.805 ); the highest variance items (yielding maximum spread ); and the items nearest the mean difficulty range of the item sample. Difficulty level ranged from.407 to.841 over nine levels of difficulty in increments of.050 steps. This new list contained 6 CNC, 20 spondee and 24 cluster words. The revised list was recorded by a new sample of dysarthric speakers described in the section on subjects above. Listening tapes of these dysarthrics were presented to listeners in the same manner as in the first word intelligibility study. Fifteen different listeners for each talker served as respondents in this experiment. -29

Listener responses were examined to determine the range of agreement among listeners as well as the range of difference between dysarthrics. The number of words correctly identified was also used as the measure of intelligibility for this study. Results Acoustic-Phonetic Studies: The first results of the spectrographic analysis of dysarthric speech were based on a study of two dysarthric speakers. A preliminary report of the findings was made by Dr. Lehiste, then co-principal investigator. The report appears as Appendix III. Her analysis of the spectrograms suggested that dysarthrics show at least one, but more often several, of the difficulties noted below: 1. Lack of coordination between phonation and articulation. 2. Difficulty in performing controlled movements. 3. Difficulty in kinesthetic monitoring of point and manner of articulation. This preliminary analysis was extended to a description of all of the dysarthrics and a comparison with a normal speaker. Results of the extended analysis were presented at the meetings of the Acoustical Society of America and the American Speech and Hearing Association in 1961. (Appendices IV and V.) An attempt was made to relate -30

the spectrographic findings to intelligibility as determined by listener responses. It was found that in some instances it was possible to predict listener errors from the spectrographic data. While this was not universally true for all speakers, it occurred quite frequently for several speakers. There was on the other hand no clear-cut relation between etiology, time since onset and the nature of the type of difficulty in speeahbproduction as deduced from the spectrograms. This in part may be accounted for by the small sample of speakers as well as the lack of precision in determining the locus of the lesion producing the dysarthric symptoms. The major area of difference in the spectrograms of dysarthrics and normals appeared in production of consonants. Some differences were also noted in the analysis of the vowels. Initial comparison of the vowel diagrams of a dysarthriciand a normal suggested some compression of the diagram and shift in average formant frequencies. A separate analysis of the vowels was undertaken and includes all normal speakers as well as dysarthrics. The results of this analysis will be discussed later. A detailed description of the speech of the ten dysarthric speakers who participated in the first segment of the research program was carried out by Dr. Lehiste. In addition to the individual descriptions, Dr. Lehiste summarized the dysarthric features observed in the ten -31

speakers. Her findings have been published as an entire issue of Bibliotheca Phonetica. A comparison of vowel production between dysarthrics and normals was carried out. It appeared that, despite marked differences in consonant production and intelligibility between the two groups of speakers, accuracy of vowel production by dysarthrics as determined by spectrographic analysis was very much the same as for normals. To assess this apparent similarity, a detailed study of the F1 and F2 characteristics of the 11 vowels employed in the test lists was carried out. For this study, only nine of the ten dysarthrics was analyzed. One speaker was considered as anarthric and was almost competely unintelligible. Thus, he and his control were not included in this or the intelligibility studies. F1 and F2 measurements were made for all of the vowels produced at the steady state segment of the utterance. Measurements were within an accuracy range of #25 cps. Means and standard deviations for F and F2 were computed for each vowel for each of the two groups of talkers and lIbid. -32

for each speaker. Table II presents the mean F and F2 frequencies for each of the 11 vowels across all speakers in the two groups. It also shows the standard deviations and range of means and standard deviations for the vowels For the most part, the mean F1 values for dysarthrics and normals are quite similar. However, the range of mean values for a given vowel show greater variability for the two groups of speakers. Normal speakers tend to show a greater dispersion than do dysarthrics. This is also reflected in the size of the standard deviations and range of standard deviations within vowels. There is a tendency for the mean F2 values for the two populations to be further apart than there is for the F1 values. However, there seems to be no particular pattern to the observed differences. The means and standard deviations reported in Table 11 were used to illustrate graphically the general configuration of the vowel diagrams for the two populations of speakers. Figure 1 is the vowel diagram based on the mean F1 and F2 measurements for each vowel for all speakerso Figures 2 and 3 permit a graphic comparison to be made between groups for each of the formants. These diagrams illustrate a remarkable degree of agreement in terms of their general shape and positioning. Such a -33

Table 11. Range of Means and Standard Deviations for F and F2 frequencies for 11 vowels for normal and dysarthric speaiers. Range of Means Range S.D. S.D. Range of Means Range S.D. Number Means F1 Units F1 Units Means F2 F2 S.D Units occurences F1 Units F2 vowel in **,, F2 sample N D N D N D N D N D N D N D N D Vowels i 47 30 366 142 343 44 16.7 264 9.6 200 2133 277 74.4 1908 121 71.2 10 I I e 89 123 167 100 106 84 178 148!p 403 434 409 413 628 603 148 528 619 596 665 633 517 537 140 100 168 117 73 96 65 105 33.5 29.5 42.0 33.9 96.4 10.2 52.7 35.4 44.2 21.7 40.5 69.9 75.0 44.8 325 350 361 308 1736 275 95.3 1635 191 97.4 1982 315 88.4 1800 155 44.5 503 1691 336 1525 522 1623 533 1511 197 124.0 200 82.5 273 117.0 193 130.0 18 15 20 23 a a 92 86 156 94 226 136 105 103 -16 100 64 125 233 289 170 150 353 228 1183 104 65.0 1175 160 85.8 1200 1163 96 36.9 176 40.8 18 22 20 953 109 94.4 968 113 71.8 0 u u 127 78 59 50 92 90 426 465 427 438 397 388 108 95 66 22 108 91 35.2 26.0 21.2 17.1 26.9 15.9 100 832 133 891 245 184 431 242 1063 975 996 925 96 37.6 169 41.7 147 78.7 66 65.1 198 178.3 214 90.4 10 6 11 1 *The mear:s represent the mean value Upper figure represent N, lower D. based on mean for all speakers in each group.

35 2 4000 2000 1000 500 -- 200 I I, ~I I I I I Normals-(N = 9) Dysorthrics —-(N= 9) 300 400 F1 600 800 1000 ji A e e I 0 ae * U x E. 0 v u 0 I I, I Figure 1. Mean frequency of vowels for 11 vowels for F1 versus F2. for all speakers

Normals (N — (N9) Dysarthrics - - - - - (N=9) 900 850 800 750 700 650 L. 600 550 500 450 400 350 300 W, i e I I e ~ a a 3 0 U u VOWELS Figure 2. Mean frequency on 11 vowels for F1. of vowels for all speakers Normals (N=9) Dysorthrics - --— (N=9) 2200 2100 2000 1900 1800 1700 1600 LL 1500 1400 1300 1200 1100 1000 900 800 i e I e ~ aQ 3 0 U u VOWELS Figure 3. Mean frequency of vowels for all speakers on 11 vowels for F2.

finding was not expected in view of the marked difference in intelligibility and other features of speech production between dysarthrics and normals. This apparent absence of difference may be accounted for by the fact that the formants were measured during the steady state of vowel production, and the dysarthric impairment shows up most often in the transitions. Secondly, use of the mean values mask differences in the variability present in the two populations. The differences and similarities in means and standard deviations are illustrated in figures 4 and 5. These figures show, not only the mean formant frequencies for the vowels, but also the range of variability around the mean for F1 and F2 for each vowel. The vertical lines originating at the meanvalues are values for one standard deviation above and below the mean. The range of difference within and between the groups for each of the formants now stand out. There is obviously considerable overlap, but on the whole the standard deviations for dysarthrics are somewhat smaller than those observed for normals. Figures 6 and 7 are plots of the individual mean F1 and F2 values for normals and dysarthrics, respectively. While the general configuration for the two groups was:generally"the same and for the most part is in agreement -37

900 875 850 825 800 775 750 725 700 675 650 625 LL 600 575 550 525 500 475 450 425 400 375 350 325 300 ( D ) Dysarthrics (N ) Normal T Dysarthrics - Means i T Normals - Means T I / S.D. I SD. TT I0 1= IT I TT ii \,A 00 I _ TT 11 T T1 o / T T o 1 ~L T o T ITx N D /i/ N D /I/ N D /eI/ N D /e/ N D /IC/ N D /a/ N D /0/ N D /a/ N D /0/ N D /U/ N D /u/ Figure 4. Means and standard deviations of vowels for all speakers on 11 vowels for F1.

Dysorthrlcs - MeansT (D) Dysorthrics _ ISD (N) Normal 2250 2200 2150 2100 2050 2000 1950 1900 1850 1800 1750 1700 c 1650 1600 1550 1500 1450 1400 1350 1300 1250 1200 1150 1100 1050 1000 950 900 850 800 Normals - Means T _ / SD. -T -I T 1IT I T I T- T I II II 0_ xI IT Ii IT IT T I T Ti 0 I N D /i/ N D /I/ ND N D N D N D ND N D ND N D /U/ N D /u/ /a/ /3/ /a/ /0/ Figure 5. Means and standard deviations of vowels for all speakers on 11 vowels for F2.

2 4000 3000 2000 1500 1000 800 600 500 -F-' 200 -250 1 300 350 400 450 500 550 600 t *. 0'-.%9~ -1700 -800 _ N /_( ie1', _x) e I._/ ~ X i 0 f',,,,, 3 z -1900 U 0~~, U A' \ / /"'-" \I/ "~ — 1000 a a 1500 Figure 6. Frequency of F2 versus Fl for vowels spoken by 9 normal speakers.

41 < F2 4000 3000 2000 1500 1000 800 I. I I I T 600 500 I - 200 -250 -300 — 350 400 450 500 550 600 F, 1:o: o 0: e. e. ee -1700 -1 800: o ae ~\DJ 2i i I,' I e (x) I " 0 o -- 900, - I\,o, a -- 1000 1500 Figure 7. Frequency of F2 versus spoken by 9 dysarthric speakers.. F1 for vowels

with similar data reported by Peterson and Barney. Our data show greater overlap between /u/ and /u / as well as /o / and /a/ than Peterson and Barney report. The degree of dispersion of mean formant frequencies when individual speakers are considered is greater for dysarthrics than for normals. Much of this difference is accounted for by values produced by one dysarthric subject, In summary, the acoustic-phonetic analysis of dysarthric speech reveals important clues relating to the articulatory and phonatory difficulties experienced by these speakers. It is our feeling that knowledge gained from spectrographic analysis can be profitably employed in developing techniques for speech therapy with these patients. Further changes in speech production following therapy could be expected to be reflected in changes in the spectrographic patterns. However, it does not seem likely that spectrographic analysis is a good predictor of listener errors of intelligibility. Peterson, G.E., H.L. Barney. Control methods used in a study of vowels. J. Acoust.Soc. Amer. 24, 175-184 (1952). -42

Intelligibility studies. Intelligibility Study I was confined to a comparison of the socres made by listeners to normal and dysarthric speakers. In addition, an attempt was made to determine if intelligibility varied with list (CNC, Cluster words or Spondees) or was influenced by the order in which the lists were presented to the listeners. Results of this initial analysis were presented at the Acoustical Society of America in 1961 (Appendix VI.) A second paper on intelligibility was presented to the American Speech and Hearing Association in 1963 (Appendix VII) and was published in the Journal of Speech and Hearing 7 Research. In summary, the initial word intelligibility analyses showed that dysarthrics differ from normal speakers on the intelligibility measures used in this study. The ranking of the dysarthric subjects was stable across lists and orders of presentation. These findings indicate that a speaker's rank on one list can be predicted from his rank on any of the other lists. Statistically significant differences were obtained among dysarthrics. There does not appear to be a significant list or order of presentation effect. However, some dysarthrics have more difficulty in producing intelligible speech on one list than on another. This does not seem 7Tikofsky, R.S. and RP. Tikofsky. Intelligibility measures of dysarthric speech. J. Speech Hearing Res. 4, 325-333 (1964.) Copies of this paper are submitted along with this report. -43

to be directly related to degree of impairment or etiology. In some cases there appears to be a slight practice effect. An analysis of the type and frequency of consonantal confusions made by listeners to the 50 CNC words recorded by the dysarthric and normal speakers was carried out. Results indicate that there is essentially no difference in the proportion of confusions for initial and final consonants. The least intelligible consonants for dysarthrics were those involving rapid shifts of articulatory position such as /d3 /. The consonant least confused by listeners was /s/ in the initial position and /r/; /1/ and /m/ in the final position. No clearly distinctive patterns of consonantal confusions appeared which could be said to characterize the population of dysarthrics as a whole. Rank order of difficulty for each consonant based on proportion of listener confusions varied greatly from dysarthric to dysarthric. Significantly fewer confusions are made for normal speakers than for dysarthrics. A paper reporting this study was read at the American Speech and Hearing Association in 1964 and a revision has been submitted to Folia Phoniatrica for publication. (Appendix VIII.) -44

In re-examining listener responses to the dysarthric speakers it became evident that many of the words which made up the three lists contributed little to the estimate of the estimate of the speaker s intelligibility. It appeared that a more efficient, or at least as efficient, estimate of dysarthric intelligibility could be obtained from a subset of the words used in Intelligibility Study I. We felt that, if possible, a new list should contain the 50 "best" words of the total available sample of 160 words. An item analysis was carried out on each of the 160 words. The analysis was carried out on the IBM 7090 Computer at The University of Michigan Computing Center. The results of the analysis for all the words in each list, lists combined and the 50 item list, are given in Table 12. The higher the difficulty level (Mean Difficulty in the table) the easier the item was to identify. The set of 50 words finally selected for the new list had the following characteristics~ 1. Nine difficulty levels from.400 to.800 (.050 interval width) 2. Average difficulty of.619 3o Average variance of.222 4. Average standard deviation of.472 5. Average reliability of.676 -45

Table 12. Summary Table for Item Analysis on Word Data for Intelligibility Study I. l, I CNC Clusters Spondee All lists 50 item List List List combined list Mean Difficulty.578.510.752.607.619 Difficulty Range.096 to.230 to.567 to.096 to.407 to..944.815.978.978.852 Mean Variance.212.231.178.208.222 Variance Range.053 to.151 to.022 to.022 to.134 to.250.250.246.250.250 Mean Standard D..458.480.415.453.472 Standard D. Range.230 to.389 to.148 to.148 to.366 to.500.500.496.500 Mean Reliability.427.527.445.458.676 Reliability Range -.362 to.037 to.010 to -.302 to.550 to.748.821.737.805.805

6. Difficulty range of.407 to.852 7. Variance range of.134 to.250 8. Reliability range of.550 to.805 9. Minimum reliability cut-off of.550 Table 13 presents the revised 50 word list, the difficulty levels, difficulty variance, reliability for each word. Also included in the table is the list from which the words were taken. Figure 8 is a plot of the mean number correct scores for listeners to the nine dysarthrics used in Intelligibility Study I. Subjects are numbered according to rank. Figure 9 shows the distribution of Standard Deviation units for the same speakers. The S.D. for the sample (i.e. across all speakers) was 11.76. These figures indicate that it was possible to distinguish between dysarthrics on the basis of the mean number of correct responses made by the listenerso It appeared from these preliminary results that the revised list might yield a clinically useful tool for quantitatively determining some aspects of the degree of dysarthric impairment. Since 30 listeners responded to each speaker, we decided to test the degree of agreement among the listeners for the revised listo Each group of 30 listeners was random-,47

Table 13. Results of the item analysis for the revised 50 item word list. Item Difficulty Variance Reliability List Difficulty level.400 [Interval.375-.425] Mean -.411 Difficulty level.450 [Interval.425-.475] Mean -.461 blush champ sledge grove bind sleeps barb showered chant bounce train shank.407.415.448.467.467.496.522.496.496.500.504.519.241.243.247.249.249.250.250.250.250.250.250.250.642.606 C1 C1.630.607.581 73'2.676.640.640.610.606.558 Cl Cl Cl Cl Cl Cl Cl Cl Cl Cl -Co I Difficulty level.550 [Interval.525 -.575] Mean -.555 sketch joke spice thread bush duckpond with crown.563.544.563.541.533.567.563.567.246.248.246.248.249.246.246.246.756.743.725.689.647.587.579.550 C 1 CNC C1 C1 CNC Sp CNC C1 Difficulty level.600 [Interval.575-.625] Mean -.640 playground drawbridge twist shrug fused.619.611.578.604.607.236.238.244.239.239.805.757.751.701.725 sp Sp C1 C1 C1

Table 13. (Cont.) Results of the item analysis for the revised 50 word list. Item Difficulty Variance Reliability List Difficulty level.650 [Interval.625-.675] Mean -.657 scrub tongue earthquake charge more cute scarecrow dwarf.663.670.667.633.659.652.666.644.223.221.222.232.225.227.222.229.778.738.720.695.694.693.669.624 C1 C1 Sp C1 CNC C1 Sp C1 Difficulty level.700 [Interval.675-.725] Mean -.703 I -p\O I daybreak inkwell shipwreck buckwheat sheep woodwork starlight platform mushroom schoolboy.711.700.719.693.704.719.704 678.685.715.205.210.202.213.208.202.208.218.216.204.759.710.706.697.691.679.675.675.660.622 Sp Sp Sp Sp CNC Sp Sp Sp Sp Sp Difficulty level. 750 [Interval.725-.775] Mean -.768 Difficulty level.800 [Interval.775-.825] Mean -.813 washboard eggplant wildcat northwest lifeboat man cookbook.770.767.767.800.811.800.841.177.179.179.160.153.160.134.760.730.655.720.680.558.675 Sp Sp Sp Sp Sp CNC Sp

50 30 - n-) w 0 U D: z z w I o I I I I I 1 2 3 4 5 6 7 8 9 Figure 8. responses SUBJECT Distribution of mean number correct listener to each of 9 dysarthric speakers. 12 - I I 10 9 8 _ 7 z 6 en 5 4 3 2 1 I I I I I I 1 2 3 4 5 6 7 8 9 SUBJECTS Figure 9. Distribution of Standard Deviation units for 9 dysarthric speakers.

ly assigned to one of two groups. A mean number correct score based on the responses to the 50 words were obtained for each group of 15 listeners for each speaker. Speakers ranked on the basis of mean number of correct responses. Thus, tow sets of ranks were obtained, one set for each grouping of 15 listeners. A Spearman Rank.Correlation was computed to determine the degree of agreement between the two sets of ranks. The obtained rs was.99, which for an N=9 is significant beyond the.01 level. This indicates the intelligibility scores obtained for the 9 speakers was quite reliableo Before beginning the Intelligibility Study II, which used the revised word list, we determined the minimum number of listeners necessary to obtain the most reliable estimates of intelligibility. We found that 15 listeners per speaker would yield the best estimates of reliability. A new sample of dysarthrics was obtained and recorded as described in the Methods of Procedure Section. This study set out to achieve two goals. One was to determine whether the new list would produce a range of intelligibility scores which colld ultimately be developed into a scale for estimating the degree of reduction of intelligibility of dysarthric speakers. The second goal was to see if the rank order of difficulty of the words -51

determined by listener responses would agree with the ranking of difficulty derived from the item analysis. Table 14 gives the total number of correct responses, mean number of correct responses and the standard deviations obtained for the 20 dysarthrics who served as speakers in Intelligibility Study II. With 15 listeners per speaker, a total 750 responses for each speaker was obtained. It is clear from this table that, although there is a wide range of scores in terms of mean numbers of correct responses across speakers, the variability among listeners to a given speaker is quite low. Thus, it appears that this revised list can be employed to discriminate or estimate degree of dysarthric impairment based on listener responses. The distribution of mean number of correct responses for the total population of dysarthric speakers (i.e. including the first 10 dysarthrics) according to rank is plotted in figure 10o Examination of this figure suggests that it should be possible to develop a scaling procedure based on these scores for the quantitative assessment of the reduction of intelligibility in dysarthric speakers. As a preliminary to the development of such a scale, the range of correct responses for each speaker was tabulated. Table 15 presents the results of this tabulation. -52

Table 14. Raw scores, means and standard deviations for dysarthrics in Intelligibility Study II. (15 listeners per speaker.) Number of Mean Correct Response Number S.D. Scores Correct S1 409 27.2 2.7 2 647 43.1 2.1 3 628 41.8 2.8 II 242 16.1 3.2 6 32 2.1 1.1 8 676 45.0 2.5 9 575 38.3 3.8 12 136 9.0 2.1 13 704 46.9 1.7 15 534 35.6 2.9 16 592 39.4 1.9 17 634 42.2 2.7 18 532 35.4 3.9 20 319 21.2 3.6 21 396 26.4 3.0 22 404 26.9 3.6 23 703 46.8 1.3 27 430 28.6 3.5 28 13.87.72 29 528 35.2 2.1 -53 -

54 50 - 48 46 44 42 40 38 36 34 o 32 30 0 cQ 28 m M 26 z 24 z W 22 20 18 - 1614 - 12 - 0 - 8 6 4 2 13 23 10 8 14 5 2 25 19 17 3 16 9 15 1829 242627 1 22 21 7 20 4 12 30 11 6 28 SUBJECT NUMBERS Figure 10. Distribution of mean number of correct listener responses to 30 dysarthric speakers.

I I n~ Table 15. Degree and range of agreement among listeners responding to dysarthric speakers for Intelligibility Study II.* Dvsarthric speakers Number of correct responses 15 9 6 3 S8 13 27 23 12 ~l8 ~9 1i5 S29 6 S3 8 13 S27 23 12 46-50 1 5 13 12 41-45 3 6 1 13 10 2 3 36-40 4 4 6 7 31-35 6 4 7 8 1 2 26-30 2 1 1 10 21-25 3 16-20 11-15 6 6-10 8 o-5 15 1 *Cell entries represent the number of listeners who gave correct responses in the ranges indicated. Fifteen listeners responded to each speaker.

Table 15. (cont.) Degree and range of agreement among listeners responding to dysarthric speakers for Intelligibility Study II. Number of Dysarthric speakers correct responses S16 S11 S14 S19 S24 S25 S26 S30 4 2 5 10 12 8 36-40 10 1 3 5 4 \1 0\ OI 31-35 5 1 5 26-30 9 1 1 21-25 16-20 11-15 6-10 0-5 3 12 11

Table 15. (Cont.) Degree and range of agreement among listeners responaing to dysarthric speakers for Intelligibility Study II. Number of correct responses -- Dysarthric speakers S28 S2 S4 S21 S17 S22 S1 S20 S5 S7 S10 I - ~ 46-50 41 -45 36 -40!I ---- I 31-35 26-30 21-25 16-20 11-15 6-10 -1 2 11 2 5 1 1 2 8 1 8 2 6 7 4 10 4 3 12 2 5 7 3 5 5 6 2 9 4 1 6 4 15 0-5

In most cases, there is very little spread in the range of correct responses for any given speaker. On the basis of these results, we are now exploring several different possible scaling procedures which may be applied to these data Each of the 50 words were ranked in order of their difficulty as determined by listener responses. This ranking was compared to that derived for the same population of words based on the item analysis. These rankings are given in Table 16 along with the difference in rank without regard to algebraic sign. A Spearman Rank Correlations was computed to estimate the degree of association between the two sets of ranks. A rs of.54 was obtained. In order to test the significance of this value for an N of 50, it was converted to a t score. This resulted in a value of t —4.42 which is significant beyond the 01 level. Thus, the range of difficulty as well as the difficulty of each word as determined by the original item analysis was confirmed by the new set of listener responses obtained in Intelligibility Study II. An analysis of listener responses to dysarthrics reading of sentences was also carried out. The purpose of this experiment was (a ) to try to determine if sentence material was useful technique for the assessment -58

Table 16. Rank or order words in revised word list based on number of correct responses and index of difficulty. WORD RI RD d d man 1 3.5 2.5 6.25 more 2 22. 20.0 400.00 northwest 3 3.5.5.25 inkwell 4.5 14. 9.5 10.25 earthquake 4.5 19 14.5 210.25 schoolboy 6 10 4.0 16.00 buckwheat 7.5 15 7.5 56.25 drawbridge 7.5 27 19.5 380.25 scarecrow 9 20 11.0 121. 00 platform 10 17 7.0 49.00 mushroom 11 16 5.0 25.00 playground 12 26 14.0 196.00 woodwork 13 8.5 4.5 20.25 thread 14 37 23.0 529.00 sketch 15 34 29.0 841.00 washboard 16.5 5 11.5 132.25 eggplant 16.5 6.5 10 100.00 charge 8 25 749.00 lifeboat 19 2 17 289.00 daybreak 20 11 9 81.00 wildcat 21 6.5 14.5 210.25 cookbook 23 1 22 484.00 tongue 23 18 5 25.00 sledge 23 48 25 625.00 cute 25 23 2 4.00 starlight 26 12.5 13.5 182.25 sleeps 27 44 17 289.00 train 28 41 13 169.00 spice 29 34 15 225.00 twist 30 30 0 0 bounce 31 42 11 121.00 bush 32 38 6 36.00 sheep 33.5 12.5 21 441.00 -59

Table 16. Rank of order words in revised word list based on number of correct responses and index of difficulty (cont.) WORD RD d joke shrug duckpond shipwreck with shank chant champ bind blush scrub showered fused dwarf barb crown grove 33.5 35 36 37 38 39.5 39.5 41 42 43 44 45 46 47 48 49 50 36 29 31.5 8.5 34 40 44 49 46.5 50 21 44 28 24 39 31.5 46.5 2.5 6 4.5 28.5 4..5 3.5 8. j13.5 7 23.0 1 18 23 9 17.5 3.5 6.25 36.00 20,o50 882.25 16.00.25 12.25 64.00 12.25 49.00 529.00 1.00 324.00 529.00 81.00 306.25 12.25 *RI = rank based RD rank based analysis, on number of correct identifications. on level of difficulty as determined by item d = RI - RD without algebraic sign. Ed2 = 9214.50 -60

of intelligibility with dysarthrics and (b ) to determine whether there was a high correlation between intelligibility as determined through sentence data and data obtained from listener responses to single words. Perhaps the greatest difficulty encountered in this part of the research was the development of an appropriate scoring system. The scoring system adopted was based on a count of the number of words correctly identified by the listeners. This permitted a percent correct score to be computed for each speaker and was designated as the Sentence Intelligibility Score (SIS). There was a total of 312 individual words contained in the 40 sentences read by the speakers, and 10 listeners to each speaker. Thus, there were 3120 responses made to each speaker. The SIS for each speaker was obtained by dividing the number of errors by the total number of possible responses. Normal speakers attained a mean SIS of 97.4~ (average intelligibility for all speakers ) as compared to 68% for dysarthrics. Tables 17 and 18 present the total number errors made in response to each talker and his SIS. While no dysarthric speaker obtained a SIS as great or greater than a normal, the best dysarthric speaker came within.58% of the poorest normal. However, the normal -61

Table 17. Number of errors and percent intelligibility scores for sentences based on ten listeners for each normal speaker. Subject Number of Errors Intelligibility Scores in % correct IN 25 99.19 2N 81 97.40 3N 19 99.39 4N 44 98.58 5N 33 98.94 6N 67 97.85 7N 132 96.02 8N 39 98.75 9N 193 93.94 Table 18. Number of errors and percent intelligibility scores for sentences based on ten listeners for each dysarthric speaker. Q,, I,' -% %4 A T' L - - 4- - *- I - -- -, - - -- f -o.,-;I I I = 0 UA kJJ CL U C lumoer oII errors Inteil. ~core in o7 correct - - 1D 2D 3D 4D 5D 6D 7D 8D 9D 207 680 326 1060 820 763 1082 1693 2438 93.36 78.70 89.55 66.03 73.71 75.50 65.30 45.73 21.85 -62

speaker was not the control for the dysarthric. On the whole, the dysarthrics were more intelligible using sentence estimates of intelligibility than for single unconnected words. The overall differences in intelligibility between normals and dysarthrics reported in Sentence Intelligibility Study I still hold. There is much less spread of scores among normals as compared to dysarthrics. The range of SISt's for dysarthrics was 71 percentage points (from a low SIS of approximately 22% to a high of approximately 93%) whereas the range of SIS's for normals was about 6 percentage points (from a low SIS of approximately 94% to a high greater than 99%). Thus, the technique employed to evaluate intelligibility with sentences does provide an estimate of the degree of dysarthric impairment with contextual materials which can discriminate among dysarthric speakers. The nine dysarthrics were ranked on the basis of their SISIs and these ranks were compared to their rankings for intelligibilityascores-obtained using single words. The Spearman Rank Correlation was employed to test the degree of association between the two sets of ranks. An rs of.92 was obtained, which for an N=9 is significant beyond the.01 level. -63 -

It may be concluded from this result that both variables, i.e. SIS and Word Intelligibility Score (WIS) are highly associated in this population. If this finding was shown to hold for other samples, then one might be able to use WIS as a good predictor of overall intelligibility. An analysis of the types of listener errors observed for the two populations of speakers was carried out. This represents a subjective rather than quantitative approach to the data' and was performed to determine what general patters could be observed. Several characteristic'patterns in listener responses were common to both groups of speakers: 1. The intelligibility of the various parts of speech.(i.e. nouns, verbs, adjectives, etc.) were generally equivalent. 2. Position of a word in the sentence had little or no effect on its intelligibility score (i.e. number of correct responses to the word.) 3. Length of word in a sentence as determined by number of syllables had little or no effect on its intelligibility. Some differences in the general listener response patterns between the two groups of speakers were observed. When responding to dysarthrics, listeners tended to omit sentence segments rather than to respond with "non-words" or to otherwise ignore the semantic rules of English. This was the the case in response -64 -

to normal talkers. Here the listeners responded with phonemic substitutions which resulted in "non-words." The substitutions which occurred followed patterns that could have been predicted with other experimental data on perceptual confusions under varying noise conditions. No consistent phonemic substitution pattern was apparent from the listener responses to the dysarthrics. Respondents to normal speakers tended not to omit words as was the case for the respondents to the dysarthrics. The results of the intelligibility studies suggest rather strongly that it should be possible to develop quantitative measures of some aspects of dysarthric speech which will be reliable and serve as a basis for future research as well as day-to-day clincial evaluations of these patients. -65

Summary of Results Results of the acoustic-phonetic analyses of dysarthric speech offer a fuller appreciation of the articulatory and phonatory difficulties which are characteristic of this impairment. They serve as a base from which spectrographic data can be translated to a prediction of what probably occurred during the act of speech production. The descriptions, when summarized across the dysarthrics in the sample, delineates the major features that differentiate these speakers from their normal speaking counterparts. Many of the characteristics have never been adequately described. Application of these descriptions to the design of speech therapy would be the next step in research with dysarthrics, since changes in speech production with therapy should be reflected in spectrograms. This was illustrated in the one case where data was collected during and at the conclusion of speech therapy. The acoustic-phonetic analysis demonstrates remarkably unique patterns of impairment in each dysarthric subject. These idiosyncratic patterns, even when reduced to features common to most dysarthrics, -66 -

suggest that there is no clearly definable relation between locus and extent of lesion and the speech impairment it produces. Much more needs to be known regarding the neurology of articulation and phonation as motor-sensory acts in the normal before inferences can be made about the dysarthric. Data obtained from spectrograms are only suggestive of errors listeners will make in response to dysarthric speech. Apparently, the average listener is able to ignore and, in fact, probably does not perceive, some of the more subtle acoustic alterations whichl characterize dysarthric speech. There are, however, many instances in which it is possible to predict with precision the confusions listeners will make. Further study of these occurrences appears to be warranted. The findings presented in this report indicate that objectivity in the assessment of dysarthric intelligibility can be achieved. There is a high degree of association between intelligibility scores obtained from responses to words and to sentences. Further, the degree of association between estimates of impairment given by clinicians (i.e. moderate, mild, severe) evaluating the speakers and the scores based on listener responses is not very great. Nor does there -67

appear any direct relation between neurological factors and reduction of intelligibility as measured in this research. Differences between dysarthrics and normals on all intelligibility measures was so great that no statistical test of the difference was needed. Both the initial and revised tests of intelligibility employed yield reliable scores which can be used to evaluate degree of dysarthric impairment. Results obtained from these experiments suggest that it should be possible to develop scales for evaluating the intelligibility of dysarthric speech. In addition, improvement in intelligibility following speech therapy should be reflected in intelligibility scores. It is our contention that the revised word list used in Intelligibility Study II can be developed into a useful evaluative tool for the speech clinician. Perceptual errors made by listeners further reveal the idiosyncratic nature of dysarthria. However, there does appear to be some relation between amount of error and complexity of the articulatory act necessary for accurate production. The type of confusion made by listeners can often be predicted from the sound spectrograms. -68

The results of our investigations lead us to conclude that the techniques of acoustic-phonetics, when combined with studies of speech intelligibility, are valuable tools for research on dysarthria and for speech pathology in general. Application of these methods can yield a greater degree of objectivity in the description of dysarthrics and provide a basis for a quantitative assessment which would be of both theoretic and clinical value. However, much research is still necessary before these goals can be achieved. -69

List of Publications and Papers Presented at Meetings Papers Presented at Meetings Tikofsky, RoS., I. Lehiste and R.P. Tikofsky. Phonetic Characteristics of Dysarthria. Amer. Speech and Hearing Assoc. Chicago, 1961. I. Lehiste, R.S. Tikofsky, and R.P, Tikofsky. An Acoustic Description of Dysarthric Speech. Acoustical Soc. of Amer. Cincinnati, 196-1. R.S. Tikofsky, I. Lehiste, R.P. Tikofsky. A Study of the Intelligibility of Dysarthric Speech. Acoustical Soc. of Amer. Cincinnati, 1961. R.P. Tikofsky, T.J. Glattke, and G.L. Reynolds. Intelligibility of Dysarthric Speech. Amer. Speech and Hearing Assoc. Chicago 1963 T.J. Glattke, R.P. Tikofsky. Listener Confusions of Dysarthric Speech. American Speech and Hearing Association. Chicago, 1963. R.P. Tikofsky. Assessing the Intelligibility of Dysarthric Speech. Invited Address. Houston Speech and Hearing Center, Houston, 1965. Publications (including abstracts of papers given at meetings) Tikofsky, R.S., Lehiste, I. and Tikofsky, R.P., Phonetic characteristics of dysarthria. ASHA, 3, 368 (1961 ) abs. Lehiste, I., Tikofsky, R.S. and Tikofsky, R.P., An acoustic description of dysarthric speech. J. Acous. Soc. Amer., 33, 1677 (1961 ) abs. Tikofsky, R.S., Lehiste, I. and Tikofsky, R.P,, A study of the intelligibility of dysarthric speech. J. Acous. Soc. Amer., 33, 1677 (1961) abs. Tikofsky, R.P., Glattke, T.J., and Reynolds, G.L., The intelligibility of dysarthric speech. ASHA, 5, 806 (1963 ) abs. -70

Tikofsky, R.S, and Tikofsky, R.P., Intelligibility measures of dysarthric speech. J. Speech Hearing Res., 7 325-333 (1964) - Lehiste, I., Some acoustic characteristics of dysarthric speech. Bibliotheca Phonetica, Fasc. 2, 1-124 (1965) Tikofsky, R.S., Glattke, T.J. and Tikofsky, R.P., Listener confusions in response to dysarthric speech. Submitted to Folia Phoniatrica for publication. -71

Appendix I Instructions to Listeners in Intelligibility Study I

Appendix I The instructions to the listeners were: You are about to hear three lists-of words. Please write down the word you think you hear on the sheets provided. If you do not understand a word, draw a line in the space alongside the appropriate number. Ready? List number 1. Additional instructions were played for the listeners before the second and third tests were presented: List number. Remember, if you do not understand a word, draw a line in the space opposite the appropriate number. Ready? List number -A3

Appendix II Flow Diagram for Computer Program To Analyze Listener Responses to CNC Words and ITitial and.Final Confusion Matrices for Dysarthrics

-A7READ I STO CARD TN LN SN READ REST OF LISTNER LN INITAL CON. N CORRECT IN W() J RECORD I ri d CORRECT IN W(J) RECORD WORD CORRECT ( TO HERE N \Y RECORD VOWEL CORRECT? PRINT I C M(TN) T N=3 PRINT TICM TICM I TICM 2 TICM3 TFCM TFCM I TFCM2 TFCM3 ICM I -_ ICM2 ICM3 ICMT FCM I FCM2 FCM3 FCMT (STOP)

TOTAL INITIAL CONFUSION MATRIX FOR ALL TAPES DR. AND MRS. RONALD TIKOFSKY ARE PROUD TO PRESENIT THE FOLLOWING RESULTS P B T D K G M N F V TH S Z SH R L CH J* H W WH Y BLANKOTHER P 799 104 8 0 4 2 I 0 15 0 0 0 0 0 10 1 C 0 26 1 5 0 84 20 B 134 566 3 7 2 2 4 0 8 5 0 0 0 0 0 0 1 0 0 4 0 0 69 5 T 5 5 606 40 24 14 0 0 1 0 0 0 0 0 0 4 1 0 29 0 0 0 50 31 0 1 8 66 547 0 10 0 0 3 4 24 0 0 0 0 2 29 8 0 0 0 0 83 25 K 5 0 33 2644 55 0 0 0 0 0 1 0 0 2 1 1 4 2 0 0 6 49 11 G 0 1 6 28 141 315 0 1 0 0 0 0 0 0 0 0 0 4 0 0 0 0 39 5 M 25 19 0 0 1 0 671 13 3 0 0 0 0 0 19 2 0 G 2 16 0 C 32 1 N 0 2 2 0 0 2 11 467 0 2 1 0 0 0 6 11 0 0 4 3 0 1 25 3 F 35 18 1 1 1 0 1 0 604 11 0 15 0 1 5 0 G C 6 29 5 0 64 13 V 0 19 3 0 1 1 0 1 61 132 3 2 0 0 11 0 0 0 1 2 0 0. 23 10 S 1 0 0 0 1 0 0 0 2 0 1753 9 0 16 3 1 3 0 0 0 C 18 2 SH 0 0 18 7 0 0 0 0 1 0 0 47 0 402 0 0 18 8 1 0 0 26 12 R 0 0 0 0 0 0 0 0 2 0 0 1 0 0 828 18 0 0 10 9 41 0107 64 L 4 17 0 0 C 1 5 11 3 1 12 0 0 0 49 591 0 0 2 32 0 0 44 38 CH 0 0 57 0 2 0 0 0 0 0 1 9 0 8 0 0156 26 0 0 0 9 2 J* 0 0 85 38 4 3 0 0 1 0 0 0 0 0 C 2 3 316 1 1 1 3 52 30 H 13 4 0 0 1 6 1 0 3 1 1 0 0 0 16 4 0 0 649 0 1 0104 6 W 0 4 0 0 1 0 0 0 1 0 0 0 0 0 18 5 0 0 5 672 34 0 58 12 WH 2 0 0 0 0 0 0 0 0 0 0 0 0 0 16 5 0 0 6 28 197 0 12 4 Y 0 1 0 0 1 0 0 3 0 0 0 0 0 0 5 10 0 9 67 0 0 121 50 3 I 00 I

TOTAL FINAL CONFUSION MATRIX FOR ALL TAPES DR. AND MRS. RONALD TIKOFSKY ARE PROUD TO PRESENT THE FOLLOWING RESULTS P 8 T D P 681 25 22 2 b 20 171 4 13 T 2 01078 98 D 2 5 113 814 K G 3 0 0 4 7 3 1 4 M N NG V F 3 0 0 4 0 0 0 2 0 1 2 1 0 0 0 K 10 0 32 3 874 31 G 1 M 0 N 0 NG 0 V 1 F 2 1H 0 S 0 * 0 R C L 0 CH 0 J* 0 O 36 1 0 O 0 0 0 0 2 0 0 O 19 O 7 O 0 O 0 0 0 0 3 O 3 0 9 0 52 26 157 218 0 0 0 2 0 0 696 32 4 0 0 0 1031273 115 0 0 12 53 175 9 0 0 0 1 0 0 1 6 0 0 0 1 0 0 0 1 0 1 0 0' 0 0 C 0 0 0 1 0 0 0 0 0 0 0 I C 0 3 0 0 O 4 1 0 6 0 2 3 5; 9 0 2 0 0 03 0 6 0 ) 3 J 0 0 0 0 1 0 0 3 0 0 0 2 0 1 0 1 0 0 0 224 5 7 214 1 9 30 1 17 2 6 0 0 0 3 0 3 O0 TH S SH G R L CH J* BLANKUTHER O 0 0 0 0 C0 0 40 34 o 40 1 0 0 0 0 4 G 0 0 33 L5 (3 O 0 1 0 0 0 3 1 0 107 4P81 ) *1l Qv16) 2 0 1 O 4 0 0 C 0 94 33 7 7s53 8 2 0 0 0 0 1 J 91 2 C/,'o),/O 0 0 0 0 0 0 C a 95 5(3 9),io 0 0 0 0 0 2 1 0 0 61 10 0 $ (4 1 0 0 0C 1 C. 0 1l5 21 ( O o 0o o o o j 5 25 Co 6 0 0 0 1 C 0 Q 18 40 1 1 7 0 4 0 C C0 0 8 8 10) 5 113 8 11 u C 0 3 42 3a \ ) 4 676 7 3 0 0 0 3 0 39 52 %\) C 33 456 0 C C C 0 29 13 1 S3 l) 1 31 0 204 1 C 3 4 15 11( 0 13 84 0 129 0 U 3 34 3 Q1bl 2 1 0 0 0 697 12 0 0 61 ( 20 0 0 0 0 U 52 924 0 0 322 1 I 0 26 1 2 0 L 173 16 18 26 t 1 0 C C 31 1 4C 8 o -~~~~~~~~~~~~~~~~~~~~~~~~~~i 0 0 0 2

Appendix III Preliminary Report on the Study of the Acoustic Characteristics of Dysarthric Speech

Appendix III Preliminary report on the study of the aco.ustic characteristics of dysarthric speech July 21, 1960 Ilse Lehiste The following remakrs are intended to orovide a first approximation of what should eventually grow into an acoustic description of dysarthric speech. The findings are based on a study of two individuals; some of the symptoms occurred in both sets of data, while others seemed to be associated with one speaker only. It remains to be ascertained to what extent the symptoms are characteristic of dysarthric speech in general. The findings are presented under five general headings; neither the number of classification nor their order are significant at the present moment. 1. Lack of coordination between phonation and articulation. This lack of coordination was very general in the speech of both patients. It was manifested in various ways. A rather common way of manifestation was the late onset of voicing during initial voiced resonants, such as /1/ /r/ /m/ /n/. In most instances, a voiceless period was followed by a brief voiced period; however, instances occurred where the total duration of the resonant was voiceless, and voicing started abruptly with the onset -A13

of the vowel. In final position, cessation of voicing during a final voiced resonant was also observed, but in final position this need not be a symptom of dysarthria. A very general symptom is the devoicing of voiced plosives in both initial and final position. In initial position it was noticed that the secondary characteristics of voiceless and voiced plosives were also often confused: the voiceless plosives were often articulated with very scant aspiration, whereas the supposedly voiced plosives were sometimes followed by an aspiration that was comparable to that used by the speaker with voiceless initial plosives. It is suggested that the problem of phonemic distinctions be investigated in these cases. In other words, listening tests are suggested to check whether the patient distinguishes p/b, t/d, k/g in his own speech and more generally in the speech of other speakers of the same dialect. A further interesting problem is presented by the presence of noise during the gaps of the plosives. A purely mechanical problem may have caused this; however, the possibility might be considered that during the articulation of some of the plosives the velurn may have been lowered (nasalization is often present -A14

throughout the utterance), and that the noise represents an emission of air through the nose, with a certain amount of friction produced during this emission. The acoustic records are insufficient for determining this at the moment. In addition, the lack of coordination between onset of phonation and articulatory movements suggests that we may find some evidence of "voiceless vibration." A recent French investigation has shown that in the case of two dysarthric patients, vocal fold vibration (without emission of sound) took place during the hold of voiceless plosives. It is suggested that we repeat the experiment (the instrumentation appears to be fairly simple) to determine whether the kind of lack of coordination that we have noticed can indeed be studied by such techniques. 2. Difficulty in performing controlled movements. The acoustic patterns show a great deal of irregularity in the articulation of such consonants and syllable nuclei that involve movements in which the rate of articulatory change is significant. Initial and final /r/ is a particularly defective sound; the dialect background of the speakers lets us expect a retroflexed Midwestern type of /r/, which cannot be found in the speech of the patients. Initial /w/ is practically -A15

always absent; initial /y/ is often articulated as /i/, with a slow change from the /i/ position to that of the following vowel. Another characteristic is the occasional occurrence of extremely rapid changes in the total formant pattern, without any measurable transitions between the steady state positions. Such jerky patterns suggest an inability to maintain a steady configuration for an extended period of time. In a previous study of syllable nuclei it was found that a tense-lax contrast could be associated with the relative duration of the target and the offglide. It is suggested that the material be studied to determine whether this distinction is maintained in dysarthric speech. Subjectively, it is noticed that the tense-lax distinction appears to be absent in the initial plosives; it is hoped that a technique may be developed for measuring the relative intensity of the explosions to check this observation (the continuous amplitude display is insufficient, since the energy of the explosions may be expected to appear at frequencies higher than 1500 cps). 3. Difficulty in kinesthetic monitoring of point and manner of articulation. It was observed that the speakers mA16

confused stops and fricatives at the same point of articulation. This occurred both initially and finally. In final position, it appeared that fricatives were more often substituted for plosives; in initial position, a labiodental plosive or affricate was substituted for /v/. The final /1/ appeared to be articulated without closure on several occasions; however, the acoustic records cannot be intepreted unambiguously at this moment. A further symptom that was present quite prominently in the speech of one of the patients was the apparent inability to select points of articulation between the alveolar ridge and the velum to provide necessary distinctions between t,d,w and kg I. It appeared that the speaker was fairly certain at the extremes of the possible points of articulation the labial series and some of the velar allophones of /k/ were properly produced —, but that he was quite uncertain about intermediate position. Both the /t/ series and certain of the allophones of /k/ appeared to be articulated at a general middle range, so that the word gull sounded like dull and get sounded like gyet. The speaker repeated some of these utterances until he produced an acceptable form; a change in the formant transitions reflects -A17

the successive approximations to the "correct" point of articulation. Occasionally both point and manner of articulation were affected: successive productions of hut had first a final voiced velar fricative, then a voiced velar stop, a voiced cluster gd, and finally a voiceless t. The speaker was able to understand the word he heard, but seemed to have difficulty in monitoring his own production. 4. The problem of clusters needs to be investigated more fully. It appears that some of the clusters offer greater problems than others. An order of difficulty should be established. There may be an upper limit in the number of consonants that may constitute the cluster; it needs to be investigated which member of the cluster is dropped in such instances, and whether this is determined by the position within the cluster or by the class of consonants involved, The possibility should be checked that no regular patterns may appear in these instances. 5. Further problems to investigate concern the suprasegmental components of speech: problems of intonation (one of the speakers uses rising inflections in every utterance), problems of timing; stress; distribution of intonation contours over individual words and larger -A1.8

utterances. It may be advisable to limit the present investigation to a study of the segmental components of speech. The test material devised for the study was mainly oriented toward the investigation of segmental sounds. 6. An intelligibility study of the speech of the pateints is presently being carried out by other members of the research group. -A19

Appendix IV,An Acoustic Description of Dysarthric Speech

An Acoust. c Description of Dysarthric Speech Ilse Lehiste, RS. Tikofsky, and R.P. Tikofsky Paper presented at the 62nd meeting of the Acoustical Society of America in Cincinnati on Nov. 11, 1961 For the past two years, a group of u.s at The University of Michigan have been engaged in an acoul.stic-phonetic study of dysarthric speech. Ou0. ten dysarthric subjects receive a neuirologi cal examination, their speech problems are diagnosed by a speech pathologist, an acoustic analysis of 160 test words produced by each subject is carried oi.t, and an intelligibility study of the dysarthric speech provides data for correlating the results of the acoustic analysis wi.th listener reactions Some results of the intelligibility stui'dy will be renorted separately; the present paper offers exampl.es of the types of dysarthric symptoms that can be best observed by snectrographic analysis. We have classified the symptoms according to their accol.stic characteristics. Some of them occu.r in the speech of all dysarthric subjects; others occur in the material rrodul.ced by smaller su.bgroups of the population, and we are in the process of comparing the subgroupings determined by the nerrologi.cal diagnosis of the patients. Usually several symptoms occur in each test word: therefore, I cannot present illustrations of each symptom seoarately, whi.ch woi.ld also require mi.ch more snace than is avail.able -A23

However, the following slides contain a representative collection. Slide 1, One of the general characteristics of dysarthric speech is a lack of coordination between articulation, phonation, and nasalization. The first slide contains broad band spectrograms of the word smiled, produced by a normal speaker (the first in the series), and two dysarthric subjects. If you observe the segment corresponding to /m/, you will notice that it is fully voiced in the speech of the normal subject. One of the dysarthric speakers produces a voiceless nasal, which becomes voiced about three centiseconds before the onset of the vowel.; the second dysarthric speaker produces an entirely voiceless. segment...The transitions from this segment to the following syllable nucleus indicate that although both the features of voicing and nasality are lost. the bilabial articulation has remained. Slide 2 contains spectrograms of the words gap and pike, spoken by a normal and a dysarthric subject. The contrast between a voiced and a voiceless initial nlosive is quite apparent in the normal articlation. The dysarthric subject produces a voiceless plosive in each case; the aspiration is csec for gap and 20 csec for pike. This type of neutralization of contrast between voiced and voiceless initial plosives is also a fairly general characteristic, although it does not always take this form- there are instances where voiceless plosives lose their aspiration, so that the leveling is in the other direction. Slide 3 presents a still different form in which the lack of coordination between phonation and articulation may appear, -A24.

-A25 N S6 S8 SMILED Slide 1

N N SIO GAP I ro SIbO PIKE PIKE Slide 2

N N N GULL S9b S9a S9a GULL I' —. VOTE VOTE LAKE LAKE Slide 3

The three words gull, vote, and lake are presented on the slide, pronounced by a normal speaker and a dysarthric speaker- Siubject 9 was recorded twice, once before therapy an- again after a year of speech therapy. In the word gull, the dysarthric subject is producing a voiceless plosive, followed by a laryngealized onset of the vowel. Three broad bands apnear preceding the vowel onset; in a narrow band spectrogram, these appear as harmonics rather than formants. The auditory impression is that of an implosive consonant; if the harmonics represent a highpitched soi.nd, there is no auditory impression of voicing. In the word vote the segment /v/ is voiceless. The auditory impression is of a voiceless /v/ rather than /f/. It may be interesting to note here that out of 32 listeners who judged that word, 13 identified the word vote correctly, while 19 substituted words beginning with /f/. A voiceless /v/ is not part of the English sound system, so the listeners identify it with one of the two phonetically most similar sounds, The final /t/ in this word contains noise in formant-like bands, and the release is heavily affricated, The auditory impression is of a voiceless plosive. The word lake starts with a voiceless /1/: again a sound that does not occur normally rih.nglish in,ini.tial position, The substitutions were mostly words beginning with the voicee less interdental fricative. The final /k/ presents a series of problems. The auditory imnression is that of a voiceless plosive with a heavily affricated release. Examination of the spectrogram suggest that there are actlually four segments in the /k/; first voice, then voiceless with formant-like bars, then voiced again, and lastly voiceless. The formant-like bars are similar to those observed in intitial position in gull. We,A28

of problems. The auditory impression is that of a voiceless plosive with a heavily affricated release. Examination of the,spectrogram.suggest'.that there are actually four segments in the /k/: first voice, then voiceless with formant-.ike bars, then voiced again, and lastly voiceless, The formant-like bars are similar to those observed in initial position in gull, We have not solved the problem of the intermittent voicing during the plosive; this is one reason why we are planning to extend this stu.dy to an investigation of the phonatory patterns of dysarthric speakers. Several of the other speakers also produce this noise during what should be plosive gaps. Slide 4 introduces several other symptoms, This slide contains spectrograms of four utterances of the word bou.nce. The utterance in the left top corner was produ.ced by a normal speaker. The lower spectrogram on the left hand side represents a pronunciation without an initial plosive segment; the auditory impression is that of a glottal ston. The transition from the syllable nucleu.is toward the final cluster is voiceless; final /n/ is completely Voiceless, and the whole syllable nucleus is laryngealized. The speaker whose utterance is shown in the right top hand corner had a voiceless /b/, followed by a segment of breathy phonation. A brief voiced /n/ is followed by a voiceless /n/ and an aspirated /s/. The lower right spectrogram contains an utterance by a speaker who has inserted an extra segment between the voiceless initial /b/ and the syllable nucleus, The auditory impression is of an /r/; however, the important -A29

It GPTTS 33Nn08 33Nno0 -'s S:: gS 0 <;4 " N

feature to notice here is the abrupt change from the /r/ position to the /a/ position. There is no-'transition between the /r/ segment and the vocalic segment; the artic.ulatory movement producing this change in the resonances mu.st have been very rapid indeed. This type of rapid change in formant structure is another feature of dysarthric sneech that occurs fairly consistently in the speech of all dysarthrics. The abrupt change of phonation.pattern, wi.th resultant breathy segments and laryngealized segments, is also a general feature; however, laryngealization has been observed also in normal speakers and does not by itself constitute a symptom. There are other dysarthric symptoms which anpear on spectrograms and which we expect to describe in a more systematic fashion when the report is presented in final form. I would like to show now some of the results of the analysis of the syllable nuclei. Slide 5 shows two acolistical vowel diagrams. The solid line connects points representing average formant positions of ten vowels produced by a dysarthric subject, the dashed line connects pbints representing the same syllable nuclei produced by a normal speaker. It is apparent that the series of front vowels are quite comparable; the back vowels, however, are markedly different. The second formant is evidently much higher for the dysarthric speaker than for the normal speaker; /a/ and /a/ are both -A31

F2 IN CYCLES PER SECOND 4000 2000 1000 500 200 "rl 300 400 o /u/_ u/ 0 I z^^~-^^f — \ -— /~ri ~o,, --- 600 /ae -4 /: (] - 800 1000 Slide 5

central and so close that a contrast based on formant structure is probably absent in this subjectts speech. Slide 6. The next slide shows two vowel diagrams representing the vowels used by a dysarthric speaker before and after therapy. The solid line indicates the average formant positions before therapy, the dashed line connects points indicating average formant positions in the same set of words after the patient had received speech therapy for a year. This latter diagram is very similar tohe normal diagram which was inclu.ded on the previous slide. It is evident that d.ur.ng the first recording most vowels had very small first formant values, so that the whole pattern was somewhat shifted up in addition to several other distortions. This study of dysarthric speech has provided us with many new insights. We are planning to carry through a componential analysis of the significant phonetic features, to state to what degree they may vary independently, and attempt to relate these findings to the diagnosed central nervous system impairment of the speakers. We are also studying the reactions of the listeners to the distorted speech, and are correlating the si.bstitutions in the responses with the acoustically analyzed symptoms. Knowledge of the lingui-stic system of the language, coupled with the available acoustic data, often makes it possible to predict -A35

F2 IN CYCLES PER SECOND 4000 2000 000 50 0 200 l/u l_ 9 l - - - 7j/ -\ ~ -- /~/-u/ - - -n 300 400 - rm CI r600: (n 0 800 1000 Slide 6

the substitu.tions. An illustration is presented on Slide 7. This slide contains a normal and two dysarthric nroductions of the word, hut. The normal pronunciation involves a short syllable nucleus and a long plosive gap. Subject 5 pronounced the wo-rd with a syllable nucleus that was more than twice as long. The final plosive, although voiceless, is considerably shorter than the voiceless plosive in the normal utterance. The formant positions are approximately what one would expectr In English, the long vowel most similar to the vowel in hut is /a/, The short voiceless plosive resembles a final /d/, which is very often devoiced in normal speech. It may be predicted that the sequence of the syllable nucleus and final consonant will be identified as odd. The first consonant ini.the dysarthric production is a voiced uvular fricative., Since the listeners may be expected to identify it with the phonetically most similar sound in English, their resnonses might inform us which of the English sounds annears most similar to a voiced uvu.llar fricative. The group of:30 listeners nrodi.ced no correct responses: there were two omissions. Words containing an /a/ as syllable nucleus in normal speech were substituted in 26 instances: the final consonant was heard as /d/ in 23 instances. In fol.ur instances the voiceless plosive was heard as /t/. The two -A755

Q z7 HZC: s CN I

most frequent substitutions were odd, with 11 occurences, and rod, with 8 occurrences. There were four occurrences of the word god and two of rot. The u.vular fricative was thus completely ignored in 11 instances. Those who heard the word as god, probably reacted primarily to the place of articulation cu.es present in the uvu.lar fricative: the majority of those who identified the initial consonant at all evidently felt that the uvular voiced fricative was most similar to an English /r/, In the third spectrogram on the slide the syllable nucl.e.us is neither long nor short, and the first and third formants are mu.ch lower than we would expect for a shwa. Nevertheless, 14 listeners identified the word correctly; there were ten substitutions of heart and hurt, indicating that an /r/-colored vowel was heard. Seven other substitutions involved /a/ (hot and lot). Thl.s in 17 instances the syllable nucleus was identified with a long vowel. This su.ggests-that the length of the syllable nucle.i-s probably constitutes a distinctive characteristic in English: when an intermediate length is produced, about half of the listeners hear a short vowel, whereas the other half identify the sound with a long vowel. Information of this type contributes also toward our understanding of the language.

Appendix V Phonetic Characteristics of Dysarthria

PHONETIC CHARACTERISTICS OF DYSARTHRIA Presented at The American Speech and Hearing Association 1961 Ronald S. Tikofsky, Ph.D. Ilse Lehiste, Ph.D. Rita P, Tikofsky, M.S, From the Communication Sciences Laboratory; Speech Clinic, Department of Speech of the College of Literature, Sciences and the Arts; and the Department of Neurology of The University of Michigan Medical Center, The University of Michigan. This investigation was supported by a PHS research grant, number B-2221, from the National Institutes of Health, Division of Neurological Diseases and Blindness, Public Health Services. The present paper represents an initial report of the first two years study of the phonetic characteristics of dysarthria through the u.se of acou.stic and instrumental phonetic techniques. The approach was selected in the hope that it will provide a more objective basis for the classification of some types of dysarthric speech, A further aim is to relate the obtained characteristics to the neurologic difficulty producing the dysarthric symptoms. A study of the intelligibility of the dysarthric speakers was also made. The test material developed for the study consists of -A41

four lists: 1. a set of fifty CNC words (consonant-syllable nucleus-consonant); 2. a set of sixty monosyllabic words contain-,ing -ini.tialI. and final consonant clu.sters; 3. a set of fifty spondees; and 4, a set of forty brief sentences, The lists were recorded by ten dysarthric speakers in the Communicati-o' —.Siences Laboratory —or the- Neurology.Re'search Laboratory.at the Universi-ty of Michigan Medical Center. The recordings were su.bjected to spectrographic analysis using the Bell Telephone Sound Spectrograph at the Communication Sciences Laboratory, The sound spectrograph is an electronic device which converts an accoustical signal into visual form, where the relative distribution of energy among the various frequencies of the spectrum appears as a function of time, Thus 160 spectrograms were made for each of the subjects, or a total of 1600 spectrograms for the dysarthric population. The spectrographic analysis of the sentences has not yet been completed, To study the intelligibility of the dysarthric speech, listening tapes were constructed from the original recordings Three such listening tapes were made for each subject; on each of the tapes the sequence of the lists apneared ir' a different order. Each listener was required to write the word that he thought the dysarthric uttered, No clues were present to aid the listener. Ten listeners were u.sed for each listeing tape for each dysarthric subject; a total of 49,600 responses forms the basis of the intelligibility -A42

study, The listeners were students at The University" of Michigan and members of the Communication Sciences Laboratory and the Speech Clinic who were not participants in this research. The essential biographical and clinical data for each of the ten dysarthric subjects are presented in the first slide. All of the subjects were native to the midwest or came to the middle west prior to the time they might have developed fixed dialect patterns. One of the subjects was recorded early in the course of speech therapy and rerecorded sometime after his discharge from therapy. All of the subjects were given complete neu.rologic examinations by the members of the staff of the Department of Neu.rology of The University of'Michigan Medical Center. The dysarthrics were all males between the ages of 20 and 68, and represent a variety of occupational, socio-economic and educational backgrounds. One normal native mid-western speaker was also recorded. The spectrographic analysis was designed to permit a detailed investigation of each of the utterances produced by the dysarthric speakers, as well as to provide data for comparisons between dysarthrics. In order to make the desired analysis the following measurements were made on the spectrograms obtained for each utterance made by the subject~

DYSARTHRIC SUBJECTS Subject Age Educational Occupation Neurological Etiology Date of Qualitative Amount of Number Level Diagnosis Onset Judgment of Speech Severity Therapy 1 42 M. D. 2 65 8th grade 3 68 8th grade 4 38 College (3 yrs.) 5 35 10th grade M. D. Salesman Farmer Sales Engineer Posttraumatic Brain and Cerebellar Deficit Progressive Bulbar Palsy C. V. A. Closed Head Injury Auto Accident 6/19/58 Moderate 2 mos. Unknown 11/60 Embolism 7/59 Severe Severe MildModerate None None 2 wks. Auto Accident 7/59 FactoryLaborer Postencephalitic Acute Encephalitis 1/60 Severe None I -p 6 20 College (2 yrs.) Student Posttraumatic Encephalopathy Auto Accident 8/2/58 ModerateSevere 3 yrs. 7 47 12th grade Oiler Unilateral Parkinsons Encephalitis (?) 1936 Severe None 8 38 High School Unemployed Fredric I s Ataxia Unknown 1929 Moderate None 9 25 College (2 1/2 yrs.) In service at time of accident Posttraumatic Brain Stem Damage Auto Accident 12/20/57 Moderate 1 1/2 yrs. 10 52 B. A. President of own Corporation C. V. A. C. V. A. 4/22/58 Severe 3 yrs. Slide 1

1. Position of the three formants in the syllable nucleus for all words at the target or steady state of the vowel. 2. The duration of the syllable nuicl.eu.s in the CNC words, 3. Formant position at the point of transition from the initial consonant to the syllable nu.cle.us and at the noint of transition from the syllable nucleus to the final consonant, 4. The duration of the transitions. 5. The duration of the initial and final consonants, 6. The formant positions within the resonant consonants as well as the energy concentrations in the fricatives and in the plosive releases were also measured, Intelligibility was measured in terms of the number of words correctly identified by a normal listener. Several intelligibility measures were available for sti.dy. The measures which were analyzed for the present paper were not concerned with the individual words but rather with overall performance in terms of dysarthrics, lists and orders, The following scores were computed from the original inte'.ligibility data: 1. A total intelligibility score, This score represents the total number of correct responses made by all of the listeners for a particular dysarthric siubject, 2. A total order intelligibility score. This score -A45

represents the total nu.mber of correct responses made by all of the listeners for a particular dysarthric su.bject for a given list without regard to order, 4. A total list by order intelligibility score, This score represents the total number of correct responses made by all of the listeners for a parti-cl.lar dysarthric subject for a given list in a given order. The spectrographic analysis of the test words has revealed several deviations from the expected normal patterns. Certain of the deviations occu.r in the speech of all dysarthric subjects; others occu.r in the material produced by smaller subgrou.ps of the nop.ulation, and we are in the process of comparing the subgroupings determined by the neurological diagnosis of the subjects with su.bgroupings determined on the basis of acoustically described symptoms, A few examples of the types of findings are presented on the following slides. Slide 2 contains three broad-band spectrograms of the word smiled from the cluster list. The first spectrogram represents a normal utterance. Notice the relatively short /s/ segment and the fact that /m/ is fully voiced, The following syllable nucleus contains two clearly defined targets, with smooth transitions between the targets; the final cl.uster consists of a long, voiced, velarized /1/, and a brief voiced /d/ -A46

-A47 S8 SMILED Slide 2

with a voiced release. The second spectrogram, produced by Subject 6, has an extra-long /s/ segment with a low cut-off frequency, which produces the auditory impression /f/. This sibilant-like segment is followed by aspiration, which is followed by a voiceless /m/. However, not the total duration of /m/ is devoiced; voicing starts before the /m/ is released, so that a very brief voiced nasal segment occurs before the velar closu.re takes place, The syllable nucleus has only one target,. /a/, The second element of the diphthong is omitted; the final /1/ becomes voiceless during its production, and there is no final /d/. All thirty listeners identified this word as smile, omitting the final /d/ whose absence was also noted on the spectrogram, buit evidently having no question about the initial /s/ plus /m/ sequenceo The significant contribution of the spectrographic approach is the discovery that the /m/ and /1/ were partly voiceless. The lack of coordination between phonation and articulation, illustrated on this spectrogram, is one of the general characteristics of dysarthric speecho It appears in the speech of all our subjects, although it takes different forms with different dysarthric sneakers, The third spectrogram, representing the pronunciation of the same word by Subject 8, shows a similar lack of coordination between articu.lation, phonation, and nasalization. In this -A48

instance a very brief /s/ segment is followed by an entirely voiceless segment in place of the expected voiced nasal /m/. The low Fl and F2 transitions to the syllable n.ucleus cause an auditory impression of a voiceless/ W/ and suggest that although both the features of voicing and nasality are lost, the biliabial articulation has remained. In the following syllable nucleus there are two segments, but the second segment is not an /I/, which woild be reflected in a rise of the second formant. The second formant is lowered instead, suggesting a rounded back vowel ts the second component of the diphthong. A short voiced /1/ is followed by a voiced /d/, a long voiced release, and a long asniration, In spite of the complete voicelessness and velar closuire, 16 out of 30 listeners heard the word as intended. Five listeners gave no answer; there were nine: substitutions of the words spoiled, spe —lled, and sprawled~ This points out the fact that listeners react to words rather than separate sounds~ the word most similar to the distorted stimulus word is recorded, and the exact natu.re of the distortion is disregarded. Sounds that are alien to the phonemic system of the listeners are identified with the phonetically most similar sounds in a given distributtion. -A49

The objectivity provided by spectrographic analysis enables the analyst to overcome the bias introduced by his own linguistic background. The spectrographic data have been employed in different kinds of analyses, some of which will be illustrated on the following three slides. This slide shows two acoustical vowel diagrams. The solid line connects points averaged from occurrences of ten syllable n.uclei in the sneech of Subject 9a. The dashed line connects comparable points derived from materials produced by a normal speaker. It is apparent that the first formant in the speech of the dysarthric subject has a consistently smaller value than expected; as a result, the whole pattern is shifted 1up and somewhat squeezed in the horizontal direction, The points representing occurrences-of /./ and /U/ are too close, suggesting that a contrast based on formant positions is unlikely in the speech of this s.ubject, The same applies to /a/ and /a/; both pairs are clearly distinct in the normal pattern.- S-lide 3. shows-'1<:the: same data for the same subject, after a yearts therapy. The normal pattern is the same as before. It is obvious that the vowel pattern is now much more similar to the normal pattern; the differences are probably no greater than -A50

F2 IN CYCLES PER SECONDS 4000 2000 1000 50 - -o I- I I I I i i 0 200 m I,,. j /i/%,,'! /Q**4 -/I>U//E\. // / \ / i I /ul/ I/ /a/ 300nn -. I /, /00-0-,,/ I/ 0 400 P ITI;3) 600,, rr 0 0 800 0 1000 I I V1 Slide 3

F2 IN CYCLES PER SECOND 2000 1000 4000 5C )0 200 - - _ _ __ \ ^ iA;! __ ___ __ o /0 - U/ /a/I -n 300 z Co C) 0) 400 rm -o m 600 m C z 8000 1000 I I Slide 4

between normals, SLide 5 shows a comparison between the patterns for the same subject before and after therapy. Of course, this type of vowel diagram does not bring out one characteristic deviation of dysarthric speech from normal speech~ within each set of points representing realizations of one syllable nucleus, the dispersion is much greater for a dysarthric speaker than for a normal speaker. Nevertheless, it appears that there was considerable improvement in the vowel production. It should be added, however, that the production of consonants deviated even more from the expected normal patternat.-;.the' time when therapy was ended than before therapy, and the resu.lts of the intelligibility study indicate that in snite of the improvement in the vowel pattern, the speaker was less intelligible when the second recording was made, It appears thus that consonants carry a greater amoiunt of the intelligibility information than vowels, and that vowel analysis is probably a less sensitive tool for judging the prospective intelligibility than consonant analysis. The statistical analysis of the intelligibility data yields information which has both clincial as well as theoretic significance. An analysis of variance revealed that there was a significant difference between the dys-A553

F2 IN CYCLES PER SECOND 4000 2000 1000 500 - - - --------— [- -- - 2 0 0 -/u/ 300 \ / // /u// ---- - 7/ —\/- ///u/ — 400 "''' \ \/se//)P//a1^ | _r---- -- -- ---- l I I o /al/ a 800: 1000 Slide 5

arthric speakers. There was an obvious difference between the most intelligible dysarthric and the normal sneaker However, the overall analysis failed to show any significant differences for lists or order. At first glance, it would appear that it wo.uld not be necessary to be concerned- aboelt the selection of a test word list by the diagnostician when he tries to determine the articulatory and phonatory charac~ teristics presented by the dysarthric, It would also seem that if more than one list of words were u.sed in the evalu.ation it would not matter in Which order the diagnostician heard the lists. Sulch a situ.ation would reduce the amou.nt of time necessary tc evaluate the dysarthricts speech, However, when the statistical analysis is carried fu.rther so as to examine the interactions between subjects and lists, subjects and apder, order and lists, and subjects, lists and orders we find that these interactions are all highly significant Only when we consider the entire population of dysarthrics is it possible to ignore the lists or order effects. This means that if we are conducting experiments in which we are not concerend with the individl.al it is possible to use ju.st ~ne list for the studyO However, if w-e are more interested in studying the individual, it becomes necessary to u.se more than one list and it is necessary to take into consideration the order in which the lists are heard. We can expect to find that our intelpretation of the symptoms varies according to -A55

the material selected for analysis and the order in which we make this analysis for a given speaker. Thus it is possible to miss some pertinent information if the influence of these interactions is not taken into account. In addition to the statistical analysis described above, we also ranked the subjects in terms of intelligibility based on listener responses. This procedure confirmed the results of the statistical analysis. Tt also suggests that it should be possible to develop scaling procedures for determining the severity of dysarthria on the basis of intelligibility scores. The research reported in this rpaer suggests that some new approaches to the diagnostic evaluation of dysartrhia are needed, It points out the lack of precision of the commonly used terms such as "substitutions, "distortions, " and "ommissions" for an adequate description of the dysarthrics articulatory effort, and demonstrates that it is possible to derive more precise descriptions through spectrographic analysis. These analyses plus information gleaned from intelligibility data will permit the development of methods of analysis and classification leading to more precise diagnosis and better understanding of the dysarthricfs disability, whith in turn, will result in improved therapeutic techniques~ -A56

Appendix VI A Study of the Intelligibility of Dysarthric Speech

I

A STUDY OF THE INTELLIGIBILITY OF DYSARTHRIC SPEECH presented at Meetings of Acou.stical Society of America Ronald S, Tikofsky, PhD. Ilse Lehiste, Ph D, Rita P. Tikofsky, MS. From the Communication Sciences Laboratory; Speech Cl.in.c, Department of Speech of the College of Literatui.re, Sciences and the Arts; and the Department of Neurology of The University of Michigan Medical Center, The University of Michigan, This investigation was supported by a PHS research grant, number B-2221, from the National Institutes of Health, Division of Neurological Diseases and Blindness, Public Health Services. The speech scientist and speech pathologist are both interested in the communicative act. Both are therefore concerned with problems relating to the efficient transmission of signals or information betweer speaker and listener, In the typical model information can be lost at several points: at the input the ear of the listener, at the output the speaker s articullatory act, or in the signal itself due to some external interference such as filtering. In most of the studies reported at these meetings the alterations in the signals used to study intelligibility have been experimentally induced, In the present.paper the alterations -A59

of the signals are the result of the subject's inability to produce normal articulatory and phonatory patterns because of damage to the nervous system which has impaired his ability to use the mechanism properly. Speech disorders oc'th'is typ'e are classified as dysarthrias, This paner reports an initial investigation of the intelligibility of a population of dysarthric speakers based on listener responses. The dysarthric population was composed of ten male subjects ranging in age from 20 to 69. The essential biographical and clinical data for these subjects are presented in Slide 1. Subject 9 was recorded early in the course of speech therapy and then some months after his discharge. All of the subjects were native to the middle west or came to the middle west prior to the time they might have developed fixed dialect patterns. One non-dysarthric male native midwestern speaker was also recorded. The test material developed for the study consists of four lists: 1. a set of fifty CNC words; 2. a set of sixty monosyllabic words containing initial and final consonant clusters; 3. a set of fifty spondees; and 4, a set of forty brief sentences. The lists were recorded by the speakers in the recording room at the Commu.nication -A60

DYSARTHRIC SUBJECTS Subject Age Educational Occupation Neurological Etiology Date of Qualitative Amount of Number Level Diagnosis Onset Judgment of Speech Severity Therapy 1 42 M. D. 2 65 8th grade M. D. Salesman Posttraumatic Brain and Cerebellar Deficit Progressive Bulbar Palsy Auto Accident 6/19/58 Moderate Severe 2 mos. None Unknown 11/60 3 4 5 68 8th grade 38 College (3 yrs.) 35 10th grade Farmer Sales Engineer FactoryLaborer C. V. A. Closed Head Injury Postencephalitic Embolism 7/59 Auto 7/59 Accident Severe MildModerate Severe None 2 wks. None Acute Encephalitis 1/60 6 20 College (2 yrs.) Student Posttraumatic Encephalopathy Auto Accident 8/2/58 ModerateSevere 3 yrs. 7 47 12th grade Oiler Unilateral Parkinsons Fredricts Ataxia Encephalitis (?) 1936 Severe None 8 38 High School Unemployed Unknown 1929 Moderate None 9 25 College (2 1/2 yrs.) In service at time of accident President of own Corporation Posttraumatic Brain Stem Damage Auto Accident 12/20/57 Moderate 1 1/2 yrs. 10 52 B. A. C. V. A. C. V. A. 4/22/58 Severe 3 yrs. Slide 1

Sciences Laboratory or in the Neurology Research Laboratory at The University of Michigan Medical Center, A set of three listening tapes containing the three word lists was constructed for each of the dysarthric speakers, and one for the normal speaker, The sequence of word lists appeared in a different order on each of the listening tanes, The tapes were played to listeners through a single TDH-39 earphone. Each listener was required to write the word that he thought the dysarthric u.ttered. No clu.es were present to aid the listener. Then listeners were used for each listening tape for each dysarthric subject, so that a total of 30 listeners responded to each speaker. Only one tape was used for the normal speaker with ten listeners, A total of 49,600 responses form the basis of the intelligibility data, The listeners were students at The University of Michigan and members of the Communication Sciences Laboratory staff who were not participants in this research., Intelligibility was measured in terms of the number of words correctly indentified by a normal listener, The present paper is not concerned with an individual word analysis but rather with overall intelligibility. The following scores were computed from the listener responses: 1. A total intelligibility score. This score represents the total number of correct responses made by all of the listeners for a particular dysarthric subject. -A62

2. A total order intelligibility score. This score represents the total number of correct responses made by all of the listeners for a particular dysarthric subject for a given order. Thus there are three total order intelligibility scores. 3. A total list intelligibility score, This score represents the total number of correct responses made by all of the listeners for a particu.lar dysarthric subject for a given list without regard to order, 4. A total list by order intelligibility scoreo This score represents the total number of correct responses made by all of the listeners for a particular dysarthric subject for a given list in a given order, Two approaches were employed in the analysis of these data. An analysis of variance was performed to evaluate differences between dysarthric speakers, lists and orders and their respective interactions. In addition, the subjects were rank-ordered according to theiir intelligibility scoresi The scores made:)by the listeners for nine of the ten dysarthric subjects were used in the analysis of variance. The scores for subject 8 were so low and demonstrated so little variability that they would have exerted an iu.ndue influence on the analysis. In fact, the total nu.mber of correct responses made by all of the listeners was 21 out of a possible 4800. Slide 2 presents the summary of the -A63

Summary of analysis of variance to evaluate differences between dysarthric speakers, lists, and orders Source df ms F Subjects (A) 8 12293.6 105.1* Lists (B) 2 583.7 1.4 Orders (C) 2 85.5 2.6 AB 16 404.4 34.6* AC 16 32.2 2.8* BC 4 2599.4 10.9* ABC 32 239.2 20.4* within Cells 729 11.7 Total 809 I. I I I I I, I I *Significant Slide 2

analysis of variance, As can be seen from the slide there was only one significant main effect, that for subjects None of the other main effects were significant. This is interpreted as indicating that for the popu.lation of dysarthrics taken as a whole neither the list used nor the order of presentation of the lists was important. All of the interactions, however, were significant at the 01 level of confidence. This finding is interpreted as indicating that for a given dysarthric the listeners' ability to identify the stimulus correctly is a function of the list and the order in which it was presented. In addition, the significant lists-order interaction indicates that listeners' performance will vary with respect to list and order in the sequence with regard to the particular speaker. The least significant interaction was between subjects and order. It is felt that order plays a lesser role in producing variation in listener performance than either the subject list or the combined speaker list order effects. It may be concluded from these findings that if one is interested only in population data, list and order may be ignored. However, if information about a particular speaker is desired, then a variety of lists must be employed and the sequenc in which the lists are presented to listeners must also be-varied. -A65

The dysarthric subjects were rank-ordered on the basis of listener responses in four ways: 1. by lists without regard to position; 2, according to position on the listening tapes; 3. by list and position on the listening tape, and 4. total intelligibility. The scores for all ten sub.bjects were used in this analysis. In addition, the scores made by the listeners on the second recording of subject 9 were also included, Slide 3 shows the rank order of the dysarthric subjects based on the order of presentation of lists, list type and total intelligibility score, The su.bjects identified by numbers are ranked in each column in descending order according to the number of correct listener responses. The first section of this slide shows the rank order of the subjects by list and position on the listening tapes. Su.bjects 4, 10, 5, and2 are always ranked in the same position with respect to the other subjects, This occurs for all lists wi.th respect to the other su.bjects. Some of the s.ubjects, however, do show shifts in rank order by as much as three positions. These shifts occur more frequently for the CNC lists than for either clusters or spondees, When the intelligibility scores are collapsed over order, as shown in the second section of this slide, the rank order of the subjects remains stable except for subjects 9a and 7. The third section of the slide presents -A66

Rank order of dysarthric subjects based on the order of presentation of lists, list type and total intelligibility scores. Rank Order by List and Rank Order by Position on Listening Tape List without regard for Position on Rank Order Listening Tape in terms of CNC CL SP Total Score 1 2 3 1 2 3 1 2 3 CNC CL SP 4 4 4 4 4 4 4 4 4 4 4 4 4 1 6 6 8 6 6 1 6 6 6 6 6 6 6 9a 1 6 8 8 6 1 1 I 1 1 1 8 1 7 1 1 1 7 7 7 8 8 8 8 9a 7 3 9a 9a 7 9a 9a 9a 9a 7 7 7 3 8 8 7 7 9a 8 8 9b 7 9a 9a 9a 7 3 9a 9b 9b 9b 3 3 3 3 3 3 3 9b 9b 9b 3 3 3 9b 9b 8 9b 9b 9b 9b 10 10 10 10 10 10 10 10 10 10 10 10 10 5 5 5 5 5 5 5 5 5 5 5 5 5 2 2 2 2 2 2 2 2 2 2 2 2 2 Slide 3

the rank order for the total intelligibility scores. It can be observed that the rank order for the subjects here is the same as that for the total list scores with the exception of subjects 9a and 7. The information offered by rank ordering supports the results on the analysis of variancee An additional rank order analysis was performed according to the position of the lists on the listening tape. This analysis is shown in slide 4. Examination of these data reveals that the greatest stability of rank order occurs in position one. Six of the subjects do not shift rank order between lists when they occur in the same position on the listening tapes, However, when shifts in rank order do take place, they tend to be of greater magnitude in this analysis than at any other pointO Shifting of rank order also occurs wi.th greater frequency in this analysis than it does with respect to the criteria described in Slide 3; These findings confirm the observations In order to evaluate fully the speech of the dysarthric with respect to intelligibility, it will be necessary to use several word lists and vary the order in which the lists are presented to listeners. When compared to a normal speaker, the intelligibility scores of even the most intelligible dysarthric were significantly lower. Listeners did not get every word right for the normal speaker. However, only 34 misidentifications occurred in 1600 responses, -A68

Rank order of subjects according to position of the lists on the listening tapes Position on Listening Tape Position 1 Position 2 Position 3 List List List CNC CL SP CNC CL SP CNC CL SP 4 4 4 4 4 4 4 4 4 1 8 1 6 6 6 6 6 6 6 6 6 9a 8 1 1 8 1 8 1 7 1 1 7 7 1 8 9a 9a 9a 7 9a 9a 3 7 7 3 7 8 8 7 8 8 9a 9a 7 9b 3 3 9b 3 9a 9b 3 9b 3 9b 9b 3 9b 9b 3 9b 10 10 10 10 10 10 10 10 -10 5 5 5 5 5 5 5 5 5 2 2 2 2 2 2 2 2 2 Slide 4

Based on the total number of correct responses, it appears that listeners have less difficulty with spondees than either clusters or CNC's, Obviously, snondee words contain contextual information which is not present on the other two lists, The most difficult list was the cluster list. This was true for the normal as well as for the dysarthric speakers, Although it would be of great interest to relate the intelligibility data to the acoustic-phonetic findings, it does not seem feasible when the analysis is limited to overall scores. It is necessary that further analysis of intelligibility be carried out with respect to the individual words before a clear relation to the acoustic data may be established. However, such overall scores permit the development of techniques by which scales of dysarthric impairment with respect to intelligibility may be developed. -A70

Appendix VII Intelligibility of Dysarthric Speech

Intelligibility of Dysarthric Speech presented at Meetings of the American Speech and Hearing Association November, 1963 R. P. Tikofsky T. Glattke G,L. Reynolds The dysarthrias present a wide range of interesting and challenging problems for the speech pathologist and neurologist For the purposes of the research reported in this paper, we shall define dysarthria as a disorder of articulation and/or phonation due to a lesion in those parts-of the nervous system which control and regulate the articulatory and phonatory mechanism. Such lesions result in distortions, substitutions and omissions of the phonetic elements of the speech signal, as alternations in rhythm, rate and vo-alc;quality. One parameter of dysarthric speech which is of great interest to the speech pathologist is intelligibility. If one of the aims of treatment is to increase the patient's intelligibility, then an objective measure of intelligibility would provide the therapist with a somewhat more accurate means for determining progress in treatment. Such measures, coupled with data obtained from acoustic-phonetic analyses of dysarthric speech, might also -A73

result'in more precise systems of classification. The primary aim of this research was to compare the intelligibility of dyarthric and normal sneakers, based on listener responses to tape recorded utterances, Listener; responses were analyzed to determine whether there were differences in the overall intelligibility among three lists of recorded words. In addition, we were interested in the effects of charging-the —order in which the lists were heard by the listeners, Comparisons of the intelligibility scores for each su.bject on each list were made to determine if his relative position across the lists remained constant, Listener responses were scored so as to yield overall scores for each list as well as compiled scores for each word. The data present in this paper are based on the list scores. Three word lists were recorded by each of the speakers: fifty CNC words; sixty monosyllabic words containing initial and final clusters; and fifty randomly selected spondee words were used. In all, each speaker recorded one hundred and sixty words, The recordings were made in the recording studio of the Communication Sciences Laboratory in a sound treated room at The University of Michigan Medical Center. The recording environment of the two rooms was -A74

approximately equal, and the recording equipment was equivalent in both situations. Only two of the speakers were recorded at the Medical Center. Each speaker was instructed to repeat the word spoken by the experimenter. None of the speakers demonstrated a hearing loss as determined by audiometric screening examination. As fat as could be determined from the neurological and speech examination, none of the speakers had any perceptive impairment. A set of three listening tapes containing each of the word lists was prepared from the master tape made by each speaker. Each list appeared in the first, second, or third position on a given listening tape. No list appeared in the same position on any two tapes. The instructions to the listeners were also recorded on the tapes, However, the experimenter's instructions to the speakers did not appear on the listening tapes. The tapes were played to the listeners through a single TDH-39 earphone at a comfortable intensity level. Each listener had his own headset, and one of the phones was a dummyd In most cases, we had five listeners resparidilinaa listeniing session. The listeners were required to write the word they heard; if they could not understand a word they were told to draw a line in the space for the word on the response forms. Intelligibility was measured in terms of the number of words correctly identifed by the listeners, Four separate -A75

intelligibility scores were computed: a) total intelligibility,. the total number of correct responses made by all of the listeners for.a given speaker over all of the lists; b) total order intelligibility, the total number of correct responses made by all the listeners for a speaker for each order, that is, listening tape; c) list intelligibility, the total. number of correct responses made by all of the listeners for a speaker for a list,. without regard to the position of the list on the listening tapes; and d) list by order intelligibility, the total number of' correct responses made by all of the listeners for a speaker on a list in a given order on the listening tape. The dysarthric population was composed of ten males ranging in age from 20 to 65 years. The educational as well as occupational background of these subjects varied considerably. Post-traumatic brain damage produced the cerebral insult which resulted in the dysarthria in four cases. Localization of the lesion in all ten cases was gross when-it could be established. Three of the patients had received extensive speech therapy. Ten normal speakers approximated the dysarthric population in terms of age,. education, and occupational status. Although an attempt was made to achieve a precise pairing between dysarthrics and normals, this was not always accomplished. Therefore, the matches between individueal subjects was sometimes gross, -A76

particularly with respect to education and occupational status. On the whole, however, there was a fairly even match between groups, All of the speakers were native to the middle west or came to the area before they might have developed the dialect pattern of their place of birth, Ten listeners responded to each of the listening tapes, making a total of thirty listeners for each speaker, None of the listeners had a history of a hearing loss. The hundred listeners who participated in this study were students in speech courses at The University of Michigan, A summary of the mean number of correct responses to each speaker for each list (without regard to order) and across all lists is given in Table 1. A maximum score for a single listener on the CNC's and Spondees is fifty, and for the Clusters, sixty. One dysarthric subject's intelligibility scores were not included in the analysis, His scores were so low, that it was felt that he was more characteristic of a true anarthria, rather than dysarthria, In all, there were only 21 correct responses out of a possible 4800 responses for this subject. Because the matches between groups was gross, and the fact that separate analyses were carried out, this subject's match was included in the analysis of the normal subjects.

Examination of the data reveals a marked difference between the dysarthrics and normals. The two nopu.lations are clearly different from each other. In only one instance was there any overlap in scores, and some of this overlan may be accounted for by the fact that the two s.ubjects were not a matched, pair. The dysarthrics come closest to the normal. on the spondees, with a mean of 47.2 correct responses across subjects, as against a mean of 49.5 for the normals. Although the difference between means for the two groups is on.'.y 12.3, the range in individual means for the dysarthrics was mi.ch greater than that for the normals. The lowest mean for the dysarthrics is.6.8 and the highest 47.6, a range of 30.8, wh'le the lowest mean for the normals is 49.2 and the highest 49.8, a range of only.6, I'n the case of the CNC's and Clusters the differences between the two groulps is more obvious, However, the most significant information to be derived from this data is not in the absolute differences, bu.t rather in the distribution of scores within each group. It is obvious from these data that, although there are differences between the normals, these differences, on the whole, are slight. The distribution of scores among these subjects is not very great. On the other hand, there is a mu.ch greater difference between the dysarthrics, Several other analyses -A78

were carried out to examine the differences between the dysarthric speakers, Kendall's coefficient of concordance was employTed to compare the subjects' rankings- on-'the'three word lists and on the total test intelligibility score'fdr the three orders of presentation. Comparisons were also made across the three lists, and across the lists plus the total test score without regard to order of presentation, The obtained coefficients are all significant. This suggests that it should be possible to predict a given dysarthric's rank on one of the word lists from his total intelligibility score, or from one of the other lists. An analysis of variance was carried out to determine whether there are significant differences between subjects, lists, and orders, as we'.l as the interactions between subjects and lists, anA,between order and lists, In order to account for the difference in maximum score between the Cluster list and the other two lists, the raw scores were converted to standard scores, A senarate score was derived for each list. The only values of F significant beyond the five ner cent level were for subjects and for all of the interactions, These findings suggest that there are significant differences between subjects. However, when the groun is taken as a whole, the order in which the lists are presented and, for that matter, -A79

which of the three lists is used, has little effect on the obtained intelligibility scores, Although all of the interactions are significant, the ones which are the most interesting are the subject by list and subject by orders In the case of the former, the significant interaction suggests some subjects may have more difficulty in producing intelligible speech on one list than they might have on another, This difference may be related to the nature of the articulatory act required to produce-the words on a particular list. That is to say, a particul.ar su.bject'is dysarthric impairment may make it more difficult for him to produce one set of words as against another. In like fashion, the significant subject by order interaction suggests that the listeners may derive some benefit from having heard some of the words before others. This goes along with significant list by order interactions The results indicate that there is a great difference between the intelligibility of dysarthric and normal sneakers, as measured in this research. This wide separation does not allow us to consider the dysarthric's impaired intelligibility as an extension of the low end of the distri.bui.tion of normal speakers on these measures of intelligibility. Thus it seems more fruitful to study the two populations independentlyThere appears to be no direct relation of the degree of impaired intelligibility based on a subject's rank and the nature of the etiology of the lesion. Nor does there seem to be a direct relation between age, t'.me since onset, level of education, and gross localization, to the degree of intelli -A80

gibility impairment, There is only fair agreement between the subjective judgement of a speech pathologist as to the severity of the dysarthria and the subject's rank and intelligibility, as determined in this stui.dy. Although it cannot be otained directly from the data, it is our impression that the res.ultant decrement in intelligibility seen in the dysarthric is most likely a fu.inction of the speaker's previou.s level of intelligibility and the locu.s and extent of the lesion. Our findings do, however, indicate that it is nossibl.e to differentiate between dysarthrics in terms of intelligibility, as measured in this research. It.is our impression that many of the words contained in the three lists contribu.te little to making this estimate. A more detailed examination of the words contained in the lists is necessary, This will lead to a reduction of the length of the lists, and possibly, to one list which contains those words which best discriminate between dysarthrics. It does seem reasonable to conclude from our results that the variability seen in our dysarthrics is characteristic of the dysarthric no.nu.lat on as a whole. We also feel that it is possible to develop objective measures of some aspects of dysarthric intelligibility based on listener judgements. If su.ch objectivity cluld be achieved, we could avoid the use of such terms as'mild, " or'Imoderate" in the evaluation of the dys'arthric imnairment -A81

We are quick to agree that measures of intelligibility based on single words probably do not offer the best estimate of overall intelligibility, but they do provide a starting point for further research. -A82

Table 1. Mean number of correct responses by listeners to nine dysarthric and ten normal speakers, on the CNC, Cluster, and Spondee word lists, and for all lists combined for each speaker, and for all speakers. Subject CNC* Cl** Sp Mean for all lists*** D-1 40.5 49.5 37.6 1.37.6 D-2 35.3 45.1 47.3 127.7 D-3 34.7 41.5 46.9 1 23.1. D-4 29.9 44.5 36.0.10.4 D-5 32.0 29.2 42.5 103.7 D-6 28.1 32.7 41..4 101.9 D-7 28.9 17.9 37.5 84.2 D-8 21.3 10.7 19.7 51. 2 D-9 10.2 6.9 16.8 33.9 Mean for all speakers 28.9 30.8 37.2 97.1

Table 1. (cont.) Subject CNC* Sp Mean for all lists*** N-1 N-2 N-3 N-4 N-5 N-6 N-7 N-8 N-9 N-10 47.7 47.7 49.1 47.2 48.8 48.4 42.8 49.0 42.1 48.2 58.5 58.6 58.9 56.4 59.3 57.5 55.7 59.0 48.7 58.5 49.8 49.5 49.5 49.8 49.5 49.5 49.2 49.6 49.2 49.8 156.0 156.2 157.7 153.5 157.6 155 4 147.7 157.7 140o, 2 156.5 Mean for all speakers 47.1 57.1 49.5 153.8 *Maximum number correct = 50 **Maximum number correct- 60 ***Maximum number correct =160

Intelligibility of Dysarthric Speech Table 2 - Kendall coefficients for CNC, C1, Sp, and Total intelligibility scores across three orders of presentation for dysarthrics List W P Dysarthrics CNC.926,010 C1.993.005 Sp.970.005 Total.993.005 Table 3 - Kendall coefficients across the CNC, Cl, and Sp lists, and across the CNC, C1, Sp, and Total lists for dysarthrics. Lists W P Dysarthri c s With Totals Without Totals.960.955.005.005 -A85

Intelligibility of Dysarthric Speech Table 4 - Summary of analysis of variance for the standard intelligibility scores of the dysarthric population, N=9 Source df ms F P Subjects (Sub) Lists (Lst) Orders (Ord) SubLst SubOrd LstOrd SubLstOrd Within 8 2 2 16 16 4 32 729 7307.57 788.71 70.00 294.66 29.64 86.19 15.95 7.66 953.99 2.68 2.36 38.47 3.87 5.40 2.08.001 NS NS.001.001.005.005 Total 809 -A86

Appendix VIII Listener Confusions of Dysarthric Speech

Listener Confusions of Dysarthric Speech Theodore J. Glattke Paper presented at the American Speech and Hearing Association, November, 1963 The previous paper reported wide individual differences among dysarthric speakers in terms of word intelligibility. We also demonstrated that there was essentially no overlap between the dysarthrics and normals on our intelligibility measures. In spite of the fact that the mean number of correct responses made to normal speakers was quite high, listeners did make errors. However, our initial study made no attempt to examine the nature of the perceptual confusions made by listeners to normal or dysarthric speech. In view of the results reported in the first paper, we felt thatsu-chaanc-anAlysiss might reveal important clues regarding the nature of dysarthric speech, as well as point to important differences between dysarthrics. The results of such analyses might be correlated with the type, locus, extent of lesion, and time since onset of the dysarthria. Further, comparisons of the perceptu.al errors made by the listeners in response to normal and dysarthric speech might be of significance to the speech pathologist who is attempting to treat the dysarthria. It has been well established that the perception of the normal speech signal may be interfered with by altering the signal before it reaches the listener's ear. -A89

A variety of techniques to accomplish this interference or distortion have been reported in the literature. Another technique used to stu.dy speech intelligibility has employed synthesized speech. In these experiments the listener attempts to identify the signal when one or more of the parameters of the signal have been manipulated in the production of the synthetic speech. However, these techniques employ "artificial " means to distort the speech signal. In this sense they do not permit any conclusions to be drawn regarding the amount and type of distortion produced by impaired articulation and phonation. Lesion in the nervous system which interferes with the control of the articulatory and phonatory mechanism produces distored speech signals. This is roughly analogous.to the experimental situation in which "noise, " in its broadest sense, is superimposed on an otherwise normal speech signal. The major difference here, of course, has to do with the experimenter's ability to control the "noise, " and the degree to which the parameters of the speech signal can be manipulated. Little control or manipulation of the signal can be achieved when one considers the dysarthric speaker as a generator of distorted speech signals. We recognize, of course, that it is possible to introduce some control by speech therapy, but our argument still holds for the hypothetical case. Therefore, the analysis of the errors made by listeners to speech signal.s produced by dysarthrics might tell us something abo.ut the nature of the noise. -A90

The purpose of the present paper is to present a preliminary description of the type and frequency of the consonant errors made by listeners responding to dysarthric and normal speech. Particular emphasis is placed on the analysis of consonant confusions. In addition to a descriptive analysis of the consonant errors, we were also interested in seeing if there were differences between the proportion of confusions made for initial and final consonants, if the pattern of listener confusions for initial and final consonants differed, and whether the patterns of listener confusions varied from speaker to speaker for either the initial or final consonants. The details of the experimental method were given in the preceding paper. To review briefly, we recorded ten male dysarthrics and matched ten normal speakers to them. Each speaker recorded three lists of words, fifty CNC's, sixty Clusters, and fifty Spondees, Thirty normal hearing listeners responded to each speaker by writing the word they thought the heard the speaker say. Only responses to the CNC words were analyzed for this study. The CNC words used in this research contained twenty intitial and twenty final consonants. The initial and final consonants were the same in sixteen instances. Some of the consonants appeared only once in the list, while some occurred as many as six times. Each time a consonant appeared it was responded to by thirty listeners for each speaker. The listeners' written responses to the fifty CNC words were scored as incorrect if any portion of a word -A91L

deviated from what was intended by the sneaker. However, a listener response was scored as correct if the word cou.d be pronounced in the manner intended by the sneaker,, Onl.y consonantal errors were evaluated in this research. All of the errors made by the listeners were tabul.ated independently for initial and final consonants, The tabllations were performed by an IBM 7090 compu.ter. A program was developed to analyze the total1 ni.umber of errors for each set of ten listeners for each order of presentation and across all thirty listeners for each speaker, These results were then combined for a..l sneakers and listeners for the intial and final consonants, In addition to these tabulations, the computer ou.tlpu..t prodllced confusion matrices which indicated the nature of the error made by the listeners. That is, the confusion matrices summarized the consonants which the listeners substituted for the one intended by the speaker. In this way, it was: possible for us to determine how frequently a particu.lar consonantal confu.sion occu.rred for a given speaker, as well as the total nu.mber of each consonantal confusion summed across all sneakers, One of the dysarthrics was almost totally u.nintelligible, and he and his control were excluded from the analysis. Table One of the handoul.t gives the number of nossible responses to each of twenty-four consonants which occurred in the CNC word list, In addition, the table also presents the per cent correct responses by listeners to the dyTsarthric and normal speakers. The range in per cent of correct responses -A92

to initial consonants for the dysarthrics is 92 9 for the /s/ to a low of.8 for /j/. For normal speakers the range is 99.8 for /1/ to a low of 77.l for /v/. It is interesting to note that no errors were made for seven of the nine normal speakers on the /v/, One error was recorded for one sneaker, 24 and 30 for two others. In fact, three normal sneakers contribu.ted 178 of the 230 listener confu.sions made by the listeners for initial consonants, For final consonants, the range in per cent correct responses for dysarthrics is 86.4 for the /1/ and a low of 41.0 for the /g/, Perfect recognition of final consonants for all normal speakers occurred on the /f/, //, and /tf/, The low was 88.3 per cent for /b/. Of the 366 confusions made by listeners, two speakers contributed 110 and 127 errors, Examination of the raw data for each dysarthric sneaker on each consonant did not indicate'- any clear-cu.t natterns which wo.uld permit distinctions between sneakers to be made with any degree of confidence. However, it did annear that certain classes of consonants, when produ.]ced by dysarthric speakers, evoked listener confu.sions in sneakers whose overall intelligibility was high for the dTysarthric group. This finding mu]st be interpreted broadly, The nattern does not always hold u.p. There apnears to be a uniquee combination of overall inte.ligibility and diffic.ulty in the artici.lation of some consonants for each dysarthric, Kendall's Coefficients of Concordance were compu.ted to examine the stability of rank between initial and final consonants for the dysarthrics, A coefficient of,69 was obtained, This suggests that, on the whole, the same consonants were as difficult for the listeners to perceive in the intitial and final positions. The most notabTe excentions -A953

Kendall's Coefficients of Concordance were compluted to examine the stability of rank between initial and final consonants for the dysarthricso A coefficient of.69 was obtained. This suggests that, on the who'.e, the same consonants were as difficult for the listeners to nerceive in the initial and final positions. The most notable exceptions to this were /v/, /d3/, /g/, /A/, and /n/, The singl.e most intelligible consonant in the initial position for the dysarthrics was the /s/, while the nasal /m/ and the sonorants /r/ and /1/ were the most intelligible final consonants for the dysarthrics, This may, in nart, be accounted for by the context in which the consonant was presented, The linguist..c constraints may have been such that the listeners could predict what the solnd, or, rather the word could be only on the basis of their knowledge of English. This suggests that, even if the sound is grossly distorted, it is correctly perceived because of these other factors. Spectrographic evidence tends to confirm these inferences When voiced and unvoiced consonants were comnared, both normals and dysarthrics yielded higher intelligibility scores for the voiceless consonants, The mean intelligibility scores for the voiceless consonants were 98.,1 and 75.6 ner cent for normals and dysarthrics resnectively,. while for the voiced consonants the mean intelligibility scores were 95.3 and 68.5 per cent for normals and dysarthrics. AltXhough these differences -A94

are not great it may be conclu.i.ded that listeners have somewhat more difficulty with voiced consonants than with unvoiced ones. A comparison of the consonants, based on npace of articil.ation, was al.so carried ou.t. The groui.ing of the consonants and the results of the analysis are gi.ven in Tabl.e Two, There is essentially no difference in mean intelligibility sc-res as expressed in per cent correct responses for the normals, However, there is a slight trend in ease of identification, going from the "front" to "back." consonants, There, the reverse is true for the dysarthrics, except for tte glottal /h/ which was most intelligible for both populations. The range in mean intelligibility for the dysarthrics is 45.0 to 80.1 per cent. Apparently, the distortion introduced by the dysarthria has a greater effect on the inte'.l.lgibi.ity of "back" consonants than on the "front" ones, We also examined the intelligibility scores in terms of manner of articulation, Figi..re One is a plot of the per cent correct responses for normals and dysarthrics, when the consonants are grouped in this way. Again, there is virtually no difference between groans for the normal speakers. The range of difference in intelligibility scores for the dysarthrics shows little variation when the affricates are omitted. The range of scores were 70,6 to 79.6 per cent correct responses. However, only 55,7 per cent of the listener responses were correct for the affricates. This indicates that the dysarthric probably has more difficulty with sounds which requ.ire re.ative.y complex artic.ula tory acts. The two affricates do not require a static artici-A95

latory position but, rather, controlled movement to achieve the desired resl.ult. Even the most intelligible of oiur dysarthric speakers had trouble with the /33/. This indicated that one of the major factors which impedes the accuracy of dysarthric intelligibility is a reduction in the control of graded synchronous movements. An analysis of the errors made by the listeners was also performed. We found that omissions accounted for 30 to 60 per cent of the listener errors on both initial and final consonants for normal speakers, Twenty ner cent of the errors made for the normals involved in the slbstitui.tion of another consonant which differed from the one intended by only one distinctive feature That is, a /d/ would be given by a listner when the speaker intended /t/ On the other hand, only about 5 to 12 ner cent of the errors made in response to the dysarthrics were omissions. Thirty per cent of the errors were si.bstitu.tions involving onl.y one distinctive feature. The remaining errors were s.ubstitutions involving shifts of more than one distinctive feature, The results of this stu.dy did not indicate any clearly distinctive pattersn of errors made by listeners to the.~ysarthrics. There were, of course, marked individual differences between speakers. However, it was not possible to separate the speakers only on the basis of consonants they misarticulated. Those sneakers whose overall intelligibility was high also had fewer consonantly misident;.fications However, careful examination of the individu.al differences do indicate that, on the average, some consonants are more -A96

difficult for dysarthrics to articulate than others. These difficulties seem to be related to the dysarthricc's ability to coordinate the artici.ilatory and phonatory acts as coincidental. behaviors, Spectrographic evidence s.uports sich a contention, This seems also to be the case for those consonants which require dynamic, rather than static articu.latory acts. It is perhaps interesting to note again, and attemnt to explain why the most intelligible consonant for the dysarthrics should be /s/, We feel that listeners had less trouble resnonding to this sound for two reasons: One is that there may have been less alternatives to choose from as su.bstitutes and, perhaps more important, because listeners tend to accent a fairly wide range of distortion of this consonant before making the substitution. In view of the findings we have described, it seems that there may be some relation between the natu.re of the tyne of consonants misidentified by listeners and some aspects of the locus and extent of the lesion, However, we do not feel that inte.ligibility data as described here or the typical phonemic analysis are a good predictor of the extent or site of the lesion produi.cing dysarthria BefOre such predictions can',be' made- wer-nedd'-tol know' mo..h. more,' bolut the neurologic control of the articu.latory and nhonatory mechanism. We do believe that analysis, such as the one described in this paper, does provide beginning stens for ful.rther -A97

research on the nat.ure of the phonetic characteristics of dysarthric speech -A98

Table 1. Number of possible correct responses to 24 consonants in the CNC word list, and percent correct responses made by listeners to dysarthric and normal speakers. Number of possible correct responses Phoneme Percent of Initials Dysarthric Normal Correct Responses Finals Dysarthric Normal Initials Finals /d/ /k/ /n/ /tf/ /d/ /h/ /3/ o080 810 810 810 540 81o 810 540 81o 270 810 540 1080 810 270 540 810 810 270 270 _ _ 810 270 1350 1080 1080 540 810 1620 270 270 270 810 270 810 1350 270 270 540 540 270 74.0 69.9 74.8 67.5 79.5 58.3 82.8 86.4 74.5 48.8 92.8 74.4 76.6 72.9 57.7 58.5 80.1 82.9 72.9 44.8 --- 98.3 94.4 98,1 97.5 97.2 94.9 99.7 99.3 99.1 77.1 98.8 99.3 98.6 99.8 98.6 99.7 99.2 98.2 86.7 96.0 _ on 84.0 63 3 79.8 75.3 81.0 41.0 85.9 78.5 64.8 79.2 82.9 83.4 75.5 86.9 86.4 62.9 44.0 56.4 84.4 47.7 95.6 88.4 95.5 99.1 98.6 98.3 98.2 98.1 96.2 100.0 99.6 96.4 100.0 92.0 91.7 100.0 92.5 95.1 99.4 92.2 -A99

Table 2. Mean intelligibility scores expressed in percent for consonants grouped by place of articulation made by listeners to dysarthric and normal speakers. Consonants grouped Mean percent correct intelligibility according to place Dysarthrlcs Normals of articulation Bilabials 76.9 94.9 /m, b, p, M/ Alveolars 78.9 97.2 /n, d, t, s, z, 1, r/ Velars 64.9 97.0 /k, g, D/ Labio-dentals 71.3 93.9 /f, v/ Inter-dental 56.4 95.1 /e/ Palatal 45.0 96.0 /j/ Alveolar-palatal 60.1 97.4 /f, 3, tf, d3/ Glottal 80.1 99.2 /h/ -A100

Table 3. Percent correct responses made by listeners to dysarthric and normal speakers when consonants are grouped by manner of articulation. Consonants grouped Percent correct responses by manner of articulation Dysarthrics Normals Plosives 70.7 96.3 /p, b, t, d, k, g/ Fricatives 70.6 93.8 /v, f, e, h, M/ Nasals 79.6 97.9 /m,n, j/ Sibilants 76.3 98.9 /z, s,, 3/ Sonorants 71.9 96.3 /r, 1, w, j/ Affricates 55.7 97.7 /d3, tf/ -AlO0

UNIVERSITY OF MICHIGAN III III901111 II0326 8807 3 9015 03526 8807