I L5“) 5 l 55L—‘5! Il| 57555055155405.5555. 5-5 0-5 5 55555555 5‘55555'..5 “#5 I555 I5 W" 555555Kl "II 55555 5555“ 555555 55555 5 I I5. 5 5 5 555m 5 “I 5 ‘ 5555';- I55'5555* 5 ‘ 5555 5 :5: 5555555555555 555555555 5. 5' . I ,'~ .55555552-5555555v55 5 .5 5 55 5 5 '57 1? .5 5': 555, 5 5 5. I 5555 5 5 I555 5 I 5I III ' I5 55 . 5 555555555 " .5 55 -. I. I 5,5 555 55 I5: '55" , .1- 5.555 5, 55 .. 5, 'I .' '5' 5. 55.5 55 5...55‘,75'5‘555,5 5 55555 ‘ I555" "5555. .55 5.5 5555555 5555'” 5 555555555555 AI 555' 555 5 555 "‘ I.‘ 55 III, 5555.5 555555 5 '5 I |.I .5 I 5 :5 .‘V , 5 a 1“ 5." \5 5-‘ r‘ .—. ‘ I1. . ‘5 . "5‘. .'-- (.1: I 5 5555 5 "'! >- 5‘3“” 5. ".: .' I‘ ' . . - ¢ -: . . If. . : .'.II‘ - wee . s I: — In z.‘ . . - - - I: - - , "I ' ' ' 'xz: ‘3 5 l n: . y - I . '- "-1 . - .1 Ll' .._---. “- 'I - Q I .l - 7 v ..I."I_'"U. o"l : .._— . 1‘ {55% 5; 5.:5.' 5'55 . 5" ."._"I_ 5 - T - ' 55555.5 55’ r“ I'I‘l‘ " .355 '9--:‘~". «”RI' I ' 5' 5" ' '0 : -|_L. I5 _. 5 I -- " '_ I _5~ ’ . _ E I ) L5 53w 4 I5 III! 55555555 5555' 55 5 55 :5I5I'II'I 5I5I . . . 55555555555555‘III555'55555555555555555 55555555555555I55555 55 5555555 - . 5'I555555'5 5555555555551" I! IIIII'II‘ 552"" III.. 5 55555 555555 55555 5555555555 555555555555 5555555“ II. 55555. ‘5 5555555 5" 'ii' 5 ""5 II" 555 “.5555. 55555 5555555555 5.. 55555555 555555 555. 555551555555 55.5. 555555 55555555 55 555555555 555555 5 555 5555555555 5I555 55555 55r55'555 ~~5 "‘ II 55 5555.5 5555.555 555 5555555 55 l 55‘ 55.515555 55:. 5 5.55». 5.5.5: . ,, III 5.55:5 . 5 , , 5 I 5555 I55. 5'5 5555‘ 55 555555 55555555 .5555 55555 .5555 5555.1- ' may; .I ’8‘. .-.- u.~__" nan; . . ~,-_ ~ —... I5‘li555 5 5 5.55 .515 5.555555555555555 5n 5.5555555 '. .. :15 5'55 5' 555555515555 ‘55” 5 5 II III: III 55I 7'- M: .. ‘M :.m I «I; 'l-‘J‘f’ WaJAL 9 ., I. ‘II it I' I I 'I. I 55 .15 II; 55 - .. '_ _ . “5": . '5 .:.., '« — — NC 1‘1». WI... “:4? 555555 5 5555 I5 ‘55 5555555 I 555I555I5555555555I555 555.: 5” 555 55555 I5 51I"I5III 55'II55555II5II5555II4I 555'55 III” ‘III w will!“lllllflllllLlllmufllflllllflllljlflflll L9 LIBRARY This is to certify that the thesis entitled QUALITY MAGNITUDE ESTIMATION FUNCTIONS FOR DEGRADED SPEECH BY LISTENERS WITH NORMAL HEARING AND LISTENERS WITH SENSORINEURAL HEARING LOSS presented by Gary Dean Lawson has been accepted towards fulfillment of the requirements for Ph . D. degree in Audiology and Speech Sciences fifizggfi Major professor Date February 7; 1980 0-7639 OVERDUE FINES ARE 25¢ PER DAY PER ITEM Return to book drop to remove this checkout from your record. QUALITY MAGNITUDE ESTIMATION FUNCTIONS FOR DEGRADED SPEECH BY LISTENERS WITH NORMAL HEARING AND LISTENERS WITH SENSORINEURAL HEARING LOSS By Gary Dean Lawson A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Audiology and Speech Sciences 1980 © Copyright by GARY DEAN LAWSON 1980 ABSTRACT QUALITY MAGNITUDE ESTIMATION FUNCTIONS FOR DEGRADED SPEECH BY LISTENERS WITH NORMAL HEARING AND LISTENERS WITH SENSORINEURAL HEARING LOSS By Gary Dean Lawson Although traditional word discrimination tests continue to be widely used in clinical settings, they do not adequately predict lis- tener performance in everyday life. Speech quality judgments may be helpful in this respect, since they have sometimes differentiated among hearing aids when word discrimination tests did not. Insufficient re- search is available, however, to justify the clinical use of speech quality judgments on a routine basis. Clinical research on quality judgments has, for the most part, employed the method of paired com- parisons. Other methods, for example direct magnitude estimation, have received little or no attention. This study investigated speech quality magnitude estimates (SQMES) by 12 normal hearing listeners (Group 1) and 12 sensorineurally impaired hearing listeners (Group 2) as a function of seven degrees of three degradation types (low-pass filter bandwidth, high-pass filter band- width, and percent total harmonic distortion by linear rectification.) The purpose was to determine: (1) whether the psychophysical power law applies to the scaling of speech quality and (2) whether there are dif- ferences in the log SQME - log degree of degradation functions as a function of listener group, degradation type, and listener group-by- degradation type interaction. Gary Dean Lawson Prior to the listening tasks, each subject participated in visual magnitude estimation training and screening tasks. Dependent variables were (1) visual magnitude estimates of circle size and (2) slopes of the least squares lines of best fit which related log visual magnitude estimates to log circle size. The visual magnitude estimates and the slopes showed excellent within-session repeatability. The log-log func- tions were relatively linear and showed roughly equivalent mean slopes (about 0.7) for Groups 1 and 2. Both groups reliably produced expected data and appeared to have similar visual magnitude estimation skills. Dependent variables for the listening tasks included (1) log geo- metric mean SQMEs across trials for each degree of each degradation type and (2) the slopes of the least squares lines of best fit for the log SQME - log degree degradation functions. A subgroup of four sub- jects in each group repeated the listening tasks in a second session. Between-session reliability of log geometric mean SQMEs for individual subjects was very high for both subgroups under each degradation type, but between-session reliability of slopes for the two subgroups showed considerable variability as a function of group—by-degradation type interaction. Log geometric mean SQMEs for Groups 1 and 2 increased linearly as a function of decreasing log degree of degradation. The slopes of the log-log functions differed as a function of degradation types and group-by-degradation type interaction. The excellent reliability of the visual magnitude estimation data suggests that systematic differences in performance on the SQME tasks are probably due to perceptual differences. The linear relationship between log geometric mean SQMEs and log degree of each degradation Gary Dean Lawson type indicates that a power function exists in each case. Systematic slope differences among the log-log functions were attributed to percep- tual differences. Estimates of poor between-session reliability of slopes were attributed to perceptual difficulties. Collectively, the findings were sufficiently encouraging to warrant additional research. Possible areas of research include the application of SQMEs to the evaluation of communication systems, clinical practices in audiology, and how normal hearing and hearing impaired individuals process complex signals. To Becky iii ACKNOWLEDGEMENTS Appreciation is extended to the members of my dissertation guidance committee: Michael R. Chial, Herbert J. Oyer, Joseph R. Vorro, and Steven C. White. Special gratitude is owed to Michael R. Chial, who served as the dissertation advisor, provided extensive help in the de- velopment of the visual training task, provided perceptive comments on the manuscript, and gave support and encouragement during hard times. Special thanks are also due to Herbert J. Oyer for guidance through most of my academic program and for encouragement of early attempts at research. A debt is owed to the subjects who volunteered for the study and to a number of fellow students, friends, and associates. Michael L. Stouffer spent many hours, days, and nights writing computer programs; Lynn Waters was the talker for the experimental stimuli; Linda L. Smith provided materials for data collection; and Carol Goldschmidt and Nancy Brewer assisted in selecting stimulus materials. Thanks go to my colleagues at Western Michigan University for their patience and under- standing and to the Western Michigan University Computer Center for assistance in analyzing the data. The person who deserves the most thanks is my wife, Becky, for her- interminable patience, understanding, encouragement, and loving help through it all. Special thanks go to our parents for their continued support and encouragement and for understanding when we weren't there. 0 iv TABLE OF CONTENTS CHAPTER LIST OF TABLES . . . . . . . . . . . . . . LIST OF FIGURES . . . . . . . . . . . I INTRODUCTION . . . . . . . . . . . . . . . Background . . . . . . . . . . . . . . . General Approaches to Speech Quality Measurement . . . . . . . . . . . . . . Paired-Comparison Quality Judgments . Sensitivity to electroacoustic characteristics . . . . . . . . . . Sensitivity to other stimulus characteristics . . . . . . . . . . . . Reliability . . . . . . . . . . . . . Feasibility issues . . . . . . . . . . Quality Magnitude Estimation . . . . . . Relevant Constructs . . Statement of the Problem . . . . . . . . . Purpose . . . . . . . . . . . . . . . . . . II METHOD . . . . . . . . . . . . . . . . . . Subjects . . . . . . . . . . . . . . . . . Normal Hearing Listeners . . . . Listeners with Sensorineural Hearing Loss Stimuli . . . . . . . . . . . . . Speech Stimuli . . . . . . . PAGE xi 10 10 13 14 15 17 17 17 18 19 19 CHAPTER III Talker . . . . . . . . . . . . . . . . . Stimulus materials . . . . . . . . . . . Types of signal degradation . . . . . . . Degrees of degradation . . . . . . . . . Master recording of undegraded stimuli . Submaster recordings of filtered stimuli Submaster recordings of rectified stimuli Summary of submaster recordings . . . . Computer generated tapes . . . . . . . . Final test tapes . . . . . . . . . Effects of apparatus . . . . . . . . . . Visual Training and Screening Stimuli - . - Procedures . . . . . . . . . . . . . . Audiometric Screening . . . . . . . . . . . Visual Magnitude Estimation Training and screening 0 O I O O O O O O O O O O O O O 0 Listening Tasks . . . . . . . . . . . . . . Calibration of listening apparatus . . . SQME training . . . . . . . . . . . . . . SQME experiment . . . . . . . . . . . . Second Listening Session . . . . . . . . . Summary . . . . . . . . . . . . . . . . . . RESULTS . . . . . . . . . . . . . . . . . . . Introduction . . . . . . . . . . . . . . . . Data Reduction . . . . . . . . . . . . . . . Auditory Stimulus Magnitudes . . . . . . . SQMEs . . . . . . . . . . . . . . . . . vi PAGE l9 19 20 20 22 22 26 29 29 32 33 36 38 38 40 41 41 41 42 42 43 45 45 46 47 49 CHAPTER PAGE Statistical Procedures . . . . . . . . . . . . . . . . 49 Reliability Procedures for SQME Data . . . . . . . . 50 Within sessions . . . . . . . . . . . . . . . . . . 50 Between sessions . . . . . . . . . . . . . . . . . 50 Analysis Procedures for SQME Data . . . . . . . . . . 51 Procedures on Visual Training and Screening Data . . . . . . . . . . . . . . . . . . . . . . . . 52 Visual Magnitude Estimation Data . . . . . . . . . . . 53 Description . . . . . . . . . . . . . . . . . . . . . 53 Reliability . . . . . . . . . . . . . . . . . . . . . 57 Speech Quality Magnitude Estimation Data . . . . . . . 61 Description . . . . . . . . . . . . . . . . . . . . . 6l Reliability . . . . . . . . . . . . . . . . . . . . . 68 Within sessions . . . . . . . . . . . . . . . . . . 68 Between sessions . . . . . . . . . . . . . . . . . 72 Analysis . . . . . . . . . . . . . . . . . . . . . . 79 Presence of trends . . . . . . . . . . . . . . . . 79 Nature of trends . . . . . . . . . . . . . . . . . 79 Differences in slopes . . . . . . . . . . . . . . . 92 Summary . . . . . . . . . . . . . . . . . . . . . . . 98 Visual Magnitude Estimation Data . . . . . . . . . . 98 Speech Quality Magnitude Estimation Data . . . . . . 98 Reliability . . . . . . . . . . . . . . . . . . . . 98 Trend analyses . . . . . . . . . . . . . . . . . . 99 Slope analyses . . . . . . . . . . . . . . . . . . 99 IV DISCUSSION . . . . . . . . . . . . . . . . . . . . . . . .101 .Introduction . . . . . . . . . . . . . . . . . . . . . 101 CHAPTER Visual Magnitude Estimation Task . . . . . . . . Findings . . . . . . . . . . . . . . . . . . . Implications for the SQME Experiment . . . SQME Experiment . . . . . . . . . . . . Reliability . . . . . . . . . . . . . . . . . Within sessions . . . . . . . . . . . . . . . . Between sessions . . . . . . . Log Geometric Mean SQMEs SlOpes . . . . . . . . . . . . . Effects of degradation type for Group 1 (normal) . . . . . . . . . . . . . . . . . Effects of degradation type for Group 2 (impaired) . . . . . . . . . . . . Effects of degradation type for Groups 1 and 2 . . . . . . . . Theoretical Factors . . . . . . . . . . . . . . Information theory . . . . . . . . . . . Intelligibility theory . . . . . . . . . . . . Implications for Future Research . . . . . . . . Perception of Complex Signals . . . . . . . . . . Classification of SQME continua . . . . . . . . Matching perceptual experiences . . . . . . . . Determinants of speech quality . . . . . . . . Extensions of the current study . . . . . . . . Prediction of Perceptual Experience . . . . . Evaluation of Communication Systems . . . Clinical Application . . . . . . . . . . . . . . Description . . . . . . . . . . . . . . . . viii PAGE 102 102 102 103 103 103 104 105 109 110 110 110 111 111 114 115 116 116 116 118 118 118 119 119 119 CHAPTER PAGE Diagnosis . . . . . . . . . . . . . . . . . . . . . 120 Prognosis and progress . . . . . . . . . . . . . . 121 V SUMMARY AND CONCLUSIONS . . . . . . . . . . . . . . . . . 122 Introduction . . . . . . . . . . . . . . . . . . . . . 122 Background . . . . . . . . . . . . . . . . . . . . . 122 Purpose . . . . . . . . . . . . . . . . . . . . . . . 122 Experimental Design . . . . . . . . . . . . . . . . . . 123 Subjects . . . . . . . . . . . . . . . . . . . . . . 123 Stimuli . . . . . . . . . . . . . . . . . . . . . . . 123 Visual training and screening stimuli . . . . . . . 123 Auditory stimuli . . . . . . . . . . . . . . . . . 124 Procedures . . . . . . . . . . . . . . . . . . . . . 124 Hearing screening . . . . . . . . . . . . . . . . . 124 Visual magnitude estimation training and screening . . . . . . . . . . . . . . . . . . . . 125 SQME training . . . . . . . . . . . . . . . . . . . 125 SQME experiment . . . . . . . . . . . . . . . . . . 125 Dependent Variables . . . . . . . . . . . . . . . . . 126 Findings . . . . . . . . . . . . . . . . . . . . . . . 126 Conclusions . . . . . . . . . . . . . . . . . . . . . . 127 APPENDIX A TABULAR SUMMARY OF AGE AND AUDIOMETRIC DATA FOR INDIVIDUAL SUBJECTS AND GROUPS . . . . . . . . . . . . . 129 B TRANSCRIPT OF SIX STIMULUS PASSAGES . . . . . . . . . . . 132 C TABULAR DESCRIPTION OF THE SUBMASTER RECORDINGS, RECORDINGS PRODUCED BY THE COMPUTER SYSTEM, AND THE FINAL TEST TAPES . . . . . . . . . . . . . . . . . . 133 D GLADC: COMPUTER PROGRAM USED FOR ANALOG-TO- DIGITAL CONVERSIONS . . . . . . . . . . . . . . . . . . 138 ix APPENDIX PAGE E RPLAY: COMPUTER PROGRAM USED TO CONVERT DIGITAL SOUND FILES TO ANALOG FORM AND PLAY THEM IN WOM ORDER 0 O O O O O O O 0 O O O O C O O O 9 I O O O 146 F GSCALE: COMPUTER PROGRAM USED TO MAKE ADJUSTMENTS IN PIAYBACK LEVEL 0 O O O O C C O O I O C O O O O I O O O 160 G FREQUENCY RESPONSE MEASUREMENTS ON EQUIPMENT USED TO PREPARE AND PRESENT SPEECH STIMULI . . . . . . . . . . 167 H RUN PROTOCOL O O O O O I O O 0 O O O O O O O O O I O O O 176 I INFORMED CONSENT RELEASE FORM III . . . . . . . . . . . . 177 J AUDIOLOGICAL SCREENING FORM . . . . . . . . . . . . . . . 178 K SCRIPT FOR VISUAL TRAINING AND SCREENING TASK . . . . . . 179 L INSTRUCTIONS AND RESPONSE SHEET FOR VISUAL TASKS O C O O O O O O O I I O I I O O O O O O O O O O O 186 M PILOT STUDY OF A VISUAL MAGNITUDE ESTIMATION TASK O I I C O O O O O O O I O O O O O O O O I O O O O O 188 N INSTRUCTIONS AND RESPONSE SHEETS FOR LISTENING TASKS O I O O O I O O O I O O O O O O O O I O O O O O O O 19 7 REFERENCES 0 O O O O I O O I O O O O O C O O O O O O O O O O O O 201 TABLE LIST OF TABLES PAGE Harmonic amplitude measurements in millivolts and computed percentages (Z) of total harmonic distor- tion (THD) in the output of the variable rectifier at seven different settings. The variable rectifier was adjusted to achieve seven degrees of THD as measured by a direct method for a 1 kHz driving signal . . . . . . . . . . . . . . . . . . . . . . . . . 28 Summary of events and their time requirements . . . . . . 44 Degrees of degradation expressed as (l) stimulus cutoff frequencies and their respective bandwidths for the filtered stimuli and (2) percentages of total harmonic distortion (% THD) and their respec— tive percent undegraded values for linearly recti- fied stimuli. The seven levels of each degradation type are listed from left to right in order of in- creasing degradation . . . . . . . . . . . . . . . . . . 48 Mean slopes, standard deviations, and ranges for the pilot study group, Group 1, and Group 2. Slopes were obtained from least squares solutions for log geometric mean visual magnitude estimates as a function of log circle area (in2)- . . . . . . . . . 55 Mean slopes, standard deviations, and ranges for Groups 1 (normal) and 2 (impaired), Trials 2 and 3, and group-by-trial interactions. SlOpes were obtained from least squares solutions for log visual magnitude estimates as a function of log circle area (inz) . . . . . . . . . . . . . . . . . . . . 56 Pearson product-moment correlation coefficients (r) between the visual magnitude estimates of circle size (inz) for each subject's Trial 2 and Trial 3 stimuli. All coefficients were significant beyond 0.10.05 (df = 5; rcritical = 0.754) . . . . . . . . . . 58 xi TABLE 10 11 12 13 14 15 Analysis of variance in slopes as a function of trials (i.e., Trials 2 and 3) and groups (i.e., Groups 1 and 2). SIOpes were obtained from least squares solutions for log visual magnitude estimates as a function of log circle area (in2) . Analysis of variance in slopes as a function of groups (i.e., pilot group and Groups 1 and 2). Slopes were obtained from least squares solutions for log geometric mean visual magnitude estimates as a function of log circle area (inz) . . . . . Mean slopes, standard deviations, and ranges for Groups 1 (normal) and 2 (impaired) as a function of degradation types. Slopes were obtained from least squares solutions for log geometric mean SQMEs as a function of log stimulus values . . . Within subject correlation coefficients (Pearson r) between seven Trial 1 and Trial 2-SQMEs for each of 24 subjects (i.e., 2 groups of 12) in Session 1. "Average" coefficients were determined for groups within degradation types . . . . . . . . . . . . Within subject correlation coefficients (Pearson r) between seven Trial 1 and Trial 2 SQMEs for each of eight subjects (i.e., 2 groups of 4) within Session 1 and Session 2. "Average" coefficients were determined for groups within degradation types . Within subject correlation coefficients (Pearson r) between each of the eight subject's seven log geometric mean SQMEs for Sessions 1 and 2 under the three degradation types. "Average" coefficients were determined for groups within degradation types Test-retest correlation coefficients (Pearson r) and "average" coefficients between Session 1 and Session 2 slopes for two groups of four subjects under three degradation types. Slopes were obtained from least squares solutions for log geometric mean SQMEs as a function of log stimulus values . . . . . . . . . . . . . . . . . . . . . /Summary of analysis of variance in log geometric "mean SQMEs across trials for Group 1 as a function of log low-pass filtered bandwidth (Hz) . . . Summary of analysis of variance in log geometric mean SQMEs across trials for Group 1 as a function of log high—pass filtered bandwidth (Hz) . . . xii PAGE 59 60 66 7O 73 75 77 80 81 TABLE PAGE 16 Summary of analysis of variance in log geometric mean SQMEs across trials for Group 1 as a function of log percent of undegraded by linear rectifi- cation 0 O O O O O O O O O O O O O O O O O O O O O O O O 82 17 Summary of analysis of variance in log geometric mean SQMEs across trials for Group 2 as a function of log low-pass filtered bandwidth (Hz) . . . . . . . . . 83 18 Summary of analysis of variance in log geometric mean SQMEs across trials for Group 2 as a function of log high-pass filtered bandwidth (Hz) . . . . . . . . 84 19 Summary of analysis of variance in log geometric mean SQMEs across trials for Group 2 as a function of log percent undegraded by linear rectification . . . . 85 20 Results of test for linear trend in log geometric mean SQMEs for Group 1 as a function of log band- widths (Hz) of low-pass filtered stimuli . . . . . . . . 86 21 Results of test for linear trend in log geometric mean SQMEs for Group 1 as a function of log band— widths (Hz) of high-pass filtered stimuli . . . . . . . . 87 22 Results of test for linear trend in log geometric mean SQMEs for Group 1 as a function of log- percent (Z) undegraded by linear rectification . . . . . 88 23 Results of test for linear trend in log geometric mean SQMEs for Group 2 as a function of log band- widths (Hz) of low-pass filtered stimuli . . . . . . . . 89 24 Results of test for linear trend in log geometric mean SQMEs for Group 2 as a function of log band- width (Hz) of high-pass filtered stimuli . . . . . . . . 90 25 Results of test for linear trend in log geometric mean SQMEs for Group 2 as a function of log per- cent (Z) undegraded by linear rectification . . . . . . . 91 26 Approximate percentages of variance in log geo- metric mean SQMEs due to log degradation levels that could be accounted for by a linear equation. Percentages (Z) are shown as a function of group and degradation type . . . . . . . . . . . . . . . . . . 93 27 Results of a two-way analysis of variance in slopes as a function of two listener groups (Group 1 and Group 2) and three degradation types (low-pass fil— tering, high-pass filtering, and linear rectification). The lepes were obtained by applying the method of least squares to the log geometric mean SQMEs and log degrees of degradation . . . . . . . . . . . . . . . . . 94 xiii TABLE 28 29 30 C-1 C-2 C-3 PAGE Results of the Newman-Keuls specific comparison test on pairs of mean slopes for the three degrada- tion types. Critical values are given for all pos- sible ranges of means spanned (i.e., two or three means). A difference between any two means is significant when it exceeds the appropriate criti- cal value (CV) for a = 0.05. The number of means spanned is equal to k . . . . . . . . . . . . . . . . . . 95 Results of the Newman-Keuls specific comparison test on pairs of mean slopes for the group-by- degradation type interaction (2 groups X 3 degrada- tion types = 6 means). Critical values are given for all possible ranges spanning from two to six means. A difference between any two means is significant when it exceeds the appropriate criti- cal value (CV) for a = 0.05. The number of means is equal to k . . . . . . . . . . . . . . . . . . . . . . 96 Percentages of times the speech quality magnitude of the comparison stimulus equal to the standard stimulus was judged less than, equal to, or greater than that of the standard stimulus. Percentages are based on the magnitudes of the 24 SQMEs (2 passages judged by 12 subjects) produced by Groups 1 and 2 for the standard degradation level under each degradation type . . . . . . . . . . . . . . . . . . 108 A summary of ages, two-frequency average thresh- olds, test ear discrimination scores, and the means and standard deviations for all subjects . . . . . . . . 130 Pure tone thresholds (dB) and median thresholds as a function of ear (R, L) and frequency (Hz) for the hearing impaired subjects . . . . . . . . . . . . . . 131 Summary of the nine submaster recordings . . . . . . . . 134 Crossbreak matrix showing the makeup of the re- cordings produced by the computer system . . . . . . . . 135 Crossbreak matrix: Presentation orders for de- gradation types, passages, and random orders for comparison degradation levels . . . . . . . . . . . . . . 136 Summary of analysis of variance in log geometric mean (G.M.) magnitude estimates (Mag. Est.) as a function of circle size . . . . . . . . . . . . . . . . . 192 Summary of test for linear trend in log geometric mean magnitude estimates of circle size . . . . . . . . . 193 xiv TABLE PAGE Slopes for the least squares lines of best fit for three functions: (1) log magnitude estimates for Trial 2 as a function of log circle size (inz), (2) log magnitude estimates for Trial 3 as a func- tion of log circle size (inz), and (3) log geo- metric mean magnitude estimates across Trials 2 and 3 as a function of log circle size (inz) . . . . . . 195 FIGURE 1 10 11 LIST OF FIGURES Apparatus used to calibrate and verify filter cutoff frequencies . . . . . . . . . . . . . Frequency response curves for the Krohn-Hite Filter set at selected low-pass cutoff frequencies: (A) 600 Hz, (B) 1000 Hz, (C) 1350 Hz, (D) 1700 Hz, (E) 2000 Hz, (F) 3000 Hz, and (G) "none" . . . Frequency response curves for the Krohn-Hite Filter set at selected high-pass cutoff fre- quencies: (A) "none", (B) 1400 Hz, (C) 1850 Hz, (D) 2300 Hz, (E) 2800 Hz, (F) 3000 Hz, and (G) 3500 Hz . . . . . . . . . . . . . . . . . Apparatus used to calibrate the variable rectifier I I I I I I I I I I I I I I I I I I I I Calibration waveforms used to obtain desired percentages of total harmonic distortion . Computer system used to generate duplicate stimuli and play them back in random orders . . . Listening apparatus . . . . . . . . . . . . . Composite frequency response curves for the com- puter system (dashed lines) and the Grason-Stadler 162 speech audiometer with TDH-49 earphones (solid lines for E-l and E—2) ‘. . . . . . . . . . . . . Example of a stimulus slide used in the visual training and screening task . . . . . . . . . . Procedural flowchart . . . . . . . . . . . . . . Mean log geometric means of modulus-free visual magnitude estimates (VMEs) plotted as a function of log circle area (inz) for the pilot group and Groups 1 and 2 I I I I I I I I I I I I I I I I xvi PAGE 23 24 25 27 30 31 34 35 37 39 54 FIGURE "12 13 14 15 17 18 PAGE Mean log geometric mean modulus-free SQMEs for Groups 1 and 2 plotted as a function of log fre- quency bandwidth (Hz) for low-pass filtered stimuli. Lines of best fit were obtained from least square solutions . . . . . . . . . . . . . . . . . 62 Mean log geometric mean modulus-free SQMEs for Groups 1 and 2 plotted as a function of log fre— quency bandwidth (Hz) for high-pass filtered stimuli. Lines of best fit were obtaifiéa"ffom least squares solutions . . . . . . . . . . . . . . . . . 63 Mean log geometric mean modulus-free SQMEs for Groups 1 and 2 plotted as a function of log per- cent (Z) undegraded by linear rectification. Lines of best fit were Obtained from least squares solutions . . . . . . . . . . . . . . . . . . . . 64 Lines of best fit for modulus-equalized mean log geometric mean SQMEs for Groups 1 (normal) and 2 (impaired) plotted as a function of log stimulus values for low-pass filtering (L-PF), high-pass filtering (H-PF), and linear rectification (LR). Lines of best fit were obtained from least squares solutions. Modulus equalization was accomplished graphically by assigning the same arbitrary value to the points on the lines of best fit which represent the mean log geometric means SQME for the middle (i.e., the standard) stimulus . . . . . . . . 65 Mean slopes for groups plotted as a function of degradation types. Slopes were obtained from least squares solutions for log geometric mean SQMEs as a function of log degrees of degradation . . . . 67 Mean SIOpes for degradation types plotted as a function of groups. Slopes were obtained from least squares solutions for log geometric mean SQMEs as a function of log degrees of degradation . . . . 69 "Average" within-subject test-retest correlation coefficients between seven Trial 1 and Trial 2 SQMEs for two groups of 12 subjects under three degradation types in Session 1. "Average" cor— relations are Fisher's Z to r transformations interpolated from mean Fisher's r to Z trans- formations (Hays, 1963, pp. 680-681). The dashed horizontal line denotes the significance criterion (r = 0.754) . . . . . . . . . . . . . . . . . . . . . . . 71 xvii FIGURE 19 20 c-4 G-S G-6 PAGE "Average" within-subject test-retest correlation coefficients between seven Trial 1 and Trial 2 SQMEs for two groups of four subjects under three degradation types in Session 1 and Session 2. "Average" correlations are Fisher's Z to r trans- formations interpolated from mean Fisher's r to Z transformations (Hays, 1963, pp. 680-681). The dashed horizontal line denotes the significance criterion (r = 0.754) . . . . . . . . . . . . . . . . . . 74 Test-retest correlation coefficients (Pearson r) between Session 1 and Session 2 slopes for two groups of four subjects under three degradation types. The dashed horizontal line denotes the significance criterion (r = 0.95) . . . . . . . . . . . . 78 Apparatus used for measuring the frequency response of the tape recorders . . . . . . . . . . . . . 168 Lower portion of the frequency responses for the Ampex AG-500 recorder (A) and the Nakamichi 700 II recorder (N) I I I I I I I I I I I I I I I I I I I I I I 169 Apparatus used for measuring the frequency response of the computer system . . . . . . . . . . . . . 171 Frequency response for the computer system . . . . . . . 172 Apparatus used for measuring the combined fre- quency response of the speech audiometer and earphones . . . . . . . . . . . . . . . . . . . . . . . . 173 Frequency response for the Grason-Stadler 162 speech audiometer and the TDH-49 earphones (E—l and E-Z) o o o o o o o o o o o o o o o o o o o o o o 175 Mean log geometric mean visual magnitude estimates across Trials 2 and 3 for the visual pilot group (N = 12) plotted as a function of log circle size (inz). The solid line represents the least squares line of best fit . . . . . . . . . . . . . . . . . . . . 191 xviii CHAPTER I INTRODUCTION Several reviewers (e.g., Chial and Hayes, 1974; Oyer and Frankman, 1975; Millin, 1975; and Berger, 1978) have indicated that traditional word discrimination tests do not predict a listener's communicative effectiveness in the real world and therefore do not measure handicap. Communicative effectiveness in the real world may be more closely re- lated to the magnitude of "goodness" or overall quality of the speech a listener perceives than to the mere intelligibility of it. Licklider (1946) concluded that "amplitude distortion affects quality somewhat more severely than it does intelligibility" (p. 432). This is not to say, however, that quality and intelligibility are unrelated. Weldele (1973) and Weldele and Millin (1975) reported a significant relation- ship between preference-based ratings and discrimination-based ratings of hearing aids. It seems reasonable to assume that intelligibility contributes to the "goodness" or overall quality of one's perception of speech. Although there appears to be some interest in the use of speech quality judgments as a clinical tool, clinical methods based upon qual- ity judgments have not been well researched. In spite of the frequent complaints about traditional monosyllabic word tests, they continue to be widely used in clinical settings (Burney, 1972; Martin and Pennington, .1971; Martin and Forbis, 1978). Although Weldele and Millin (1975) en- couraged the use of quality judgments in hearing aid evaluations, they did not suggest that the use of discrimination tests be discontinued. As noted by Punch (1978), the difficulties with traditional methods. dictate that basic procedural issues and assumptions be carefully 1 2 evaluated before audiologists adopt any new clinical strategies. The first step is to examine what has already been done. Background Although there are a number of approaches to the assessment of speech quality, clinical research has emphasized the method of paired comparisons. Relatively little attention has been given to other psy- chophysical methods (e.g., magnitude estimation) or to theoretical con- siderations. General Approaches_£g Speech Quality Measurement The work on speech quality measurement was surveyed by Munson and Karlin (1962) and by Hecker and Guttman (1967). Munson and Karlin di- vided methods of measurement into "indirect comparisons" by which trans- mission systems are assessed singly and "direct comparisons" by which systems are assessed in pairs, as in paired comparisons. Hecker and Guttman (1967) categorized methods as (l) analytic approaches which aim to discover the psychological attributes of the speech signal and (2) utilitarian approaches which are concerned with determining speech quality by prior assumption of psychological attributes and reduction of measures to a unidimensional scale. The Institute of Electronics and Electrical Engineers (IEEE, 1969) recommended procedures for using subjective preference measurements to estimate speech quality. Utilitarian methods were said to be best suited for engineering practice, and three such methods were outlined: (l) the Isopreference Method, (2) the Relative Preference Method, and (3) the Category-Judgment Method. The Isopreference Method involves the comparison of a test signal to a referent signal subjected to 3 varying degrees of degradation. The isopreference level is the signal- to-noise ratio of the test and reference signal at which the test and reference signals are preferred an equal number of times. The Relative Preference Method seeks to determine the quality of the test signal by locating it on a quality continuum, which is defined by reference sig- nals representing different types of speech distortion. The test sig- nal is positioned on the continuum by considering how often it is pre- ferred to any reference signal. In the Category-Judgment Method listeners describe their impression of the quality of a Speech signal by assigning it to one of several simple categories (e.g., Unsatisfac- tory, Poor, Fair, Good, Excellent). The study of speech quality by audiologists has involved a "util- itarian" approach to the evaluation of signals transduced by hearing aids. Apparently, every study to date has employed a direct paired- comparison paradigm. Paired—Comparison Quality Judgments The first experiment on the clinical use of speech quality judgments was a paired-comparison study by Jeffers (1960). Today, at least eight studies have included paired-comparison quality judgments. Three studies included only hearing-impaired listeners. Jeffers (1960) asked 32 subjects with conductive hearing losses to give prefer- ences for the quality of speech transduced by five hearing aids arranged in pairs. Zerlin (1962) asked 21 subjects with sensorineural hearing losses to state preferences for the quality of speech transduced by six hearing aids arranged in pairs. Weldele and Millin (1975) obtained pre- ference judgments from 10 listeners with sensorineural hearing loss on pairings of four hearing aids. 4 Three additional studies included only normal hearing listeners. Witter and Goldstein (1971) obtained quality preference judgments from 30 normal hearing listeners on pairings of five hearing aids. Smaldino (1974) obtained quality preferences from 10 normal hearing subjects on pairings of stimuli transduced by 10 hearing aids. Yonovitz, Bickford, Lozar, and Ferrell (1978) obtained paired—comparison judgments and dis- similarity ratings of 12 hearing aids from 20 normal hearing listeners. Each of the two remaining studies included a group of listeners with normal hearing and a group of listeners with sensorineural hearing loss. Punch and Ciechanowski (1977) obtained preferences from 10 sub- jects in each group on pairings of stimuli transduced by five hearing aids. Chial and Daniel (1977) obtained preferences from 18 subjects in each group, using a magnitude estimation procedure as well as a paired comparison procedure on stimuli transduced by four hearing aids. It is helpful to examine the paired-comparison studies in terms of the sensitivity of quality judgments to electroacoustic and other stim- ulus characteristics, quality judgment reliability, and feasibility issues. Sensitivity £2_electroac0ustic characteristics. Several studies have suggested that paired-comparison quality judgments are related to the electroacoustic characteristics of hearing aids (e.g., Jeffers, 1960; Zerlin, 1962; Witter and Goldstein, 1971; Smaldino, 1974; and Yonovitz 35 31-: 1978). In general, these studies found that hearing aids with better electroacoustic characteristics are preferred over those with poorer electroacoustic characteristics. Jeffers (1960) and Zerlin (1962) found that preference tests differ- entiated among hearing aids when monosyllabic word discrimination tests did not. Unfortunately, neither investigator measured the electroacoustic 5 characteristics of the experimental hearing aids. Although Jeffers categorized aids on the basis of the manufacturer's specifications, Kasten and Revoile (1965) showed that the actual electroacoustic char- acteristics of hearing aids may differ significantly from the manufac- turer's design specifications. Apparently, Zerlin simply assumed that the differences in preferences for hearing aids were due to differences in electroacoustic characteristics. Weldele and Millin (1975) found that when hearing aids were rated by word discrimination scores obtained at 40 dB HTL, discrimination- based ratings and preference based ratings were significantly related. The authors reported only general descriptions of the frequency response and gain characteristics of the hearing aids. They did not comment spe- cifically on the relationship of quality judgments and electroacoustic characteristics. Witter and Goldstein (1971) and Smaldino (1974) were more system- atic in that they measured the electroacoustic characteristics of their hearing aids. Considering male and female voice stimuli separately, Witter and Goldstein found strong Spearman rank-order correlations (0.60 to 1.00) among modal preference rankings for hearing aids and rankings of transient response, frequency range, and harmonic distortion measures. Preference judgments were correlated positively with the high cut-off point for frequency response and negatively with the low cut-off point. Very low correlations were observed for preference judgments and intermodulation distortion. Smaldino (1974) found that quality judgments were (1) negatively correlated (Pearson product-moment correlations) with harmonic distortion measures at 800 Hz (-0.56), 1200 Hz (-0.59), and 1600 Hz (-0.54) with an input at 400 Hz and (2) positively correlated with the Houston Speech and Hearing Center (H.S.H.C.) bandwidth above 1000 Hz 6 (0.53). The H.S.H.C. bandwidth (Jerger and Thelin, 1968) is determined by drawing a line parallel to the frequency axis 10 dB below the highest point on a hearing aid response curve. The upper limit is the frequency at which the line intersects the curve above 1000 Hz; the lower limit is the frequency at which the line intersects the curve below 1000 Hz. Smaldino examined bandwidth above 1000 Hz, below 1000 Hz, and total bandwidth. Three other studies, Punch and Ciechanowski (1977), Chial and Daniel (1977) and Yonovitz 93 a1. (1978), also included measures of electroacoustic characteristics. Although Punch and Ciechanowski reported listener preferences among the five aids employed, they did not comment specifically on the relationship of quality judgments to electroacoustic characteristics. In the Chial and Daniel study, correlations among various electroacoustic measures and quality judgments were not signifi- cant; this finding was interpreted as failing to confirm or refute pre- vious claims about the sensitivity of quality judgments to measurable electroacoustic differences. Yonovitz gt a1. (1978) found that: (1) frequency response and bandwidth affected the perception of speech and music, (2) third harmonic distortion and internal noise were specific to speech perception, and (3) transient distortion and phase distortion were specific to music perception. Some insight into the sensitivity factor in subjects with sensori- neural hearing loss may be gained from examining reliability data. For example, Punch and Ciechanowski (1977) found that the quality preferences of normal and sensorineurally impaired listeners were highly correlated (Pearson r = 0.98), but noted that fewer hearing-impaired listeners were able to replicate their first-preference judgments. Chial and Daniel (1977) reported that although normal and dysacusic listeners expressed similar preferences for better quality signals, dysacusic listeners were 7 less consistent in expressing preferences when overall signal quality was low. These results suggest that the sensitivity of quality judgments to differences in electroacoustic characteristics may vary as a function of hearing acuity. This cannot be confirmed, however, on the basis of previous research. Jeffers (1960) failed to measure electroacoustic characteristics and used only conductively impaired listeners who probably responded much like normal listeners. Zerlin (1962) used sensorineurally impaired listeners, but failed to measure the electroacoustic character- istics of his hearing aids. The more systematic studies of Witter and Goldstein (1971) and Smaldino (1974) included only normal listeners. Sensitivity tg_other stimulus characteristics. Studies of speech quality have taken a number of approaches to the selection of stimulus materials. Jeffers (1960, p. 261) used eight one-minute tape recorded paragraphs ...believed to have little innate appeal... The source of the paragraphs and the sex of the speaker were unidentified. Zerlin (1962) used 30—second passages of Reader's Digest material tape recorded by a speaker whose dialect was General American and whose sex was un- identified. Witter and Goldstein (1971) used a single 10-second para- graph from a Thurber short story. The passage was tape recorded by a male and a female speaker of unspecified dialect. Smaldino (1974) also used a single passage. Weldele and Millin (1975) used word discrimination lists (CID Auditory Test W—22). In addition to music stimuli, Punch and Ciechanowski (1977) used 30—second passages from Mark Twain's Tom Sawyer recorded by a male and a female speaker, while Yonovitz g£_al. (1978) used a 30—second paragraph from The Rainbow Passage which was read by a male speaker of the General American dialect. Chial and Daniel (1977) used four passages tape recorded by a female speaker of General American dialect. These passages, originally intended for reading by high school 8 and junior high students, were previously used in Chial's (1973) dis- sertation. Chial judged these passages to be ...approximately equivalent along the dimensions of abstraction, grammatical complexity, vocabulary, intrinsic interest, and controversiality" (p. 19). Witter and Goldstein (1971), Smaldino (1974), and Yonovitz 35 31. (1978) attempted to control for differences in stimulus materials by using a single passage for all measurements. Chial (1973) judged his stimuli to be approximately equivalent along certain linguistic and literary dimensions. 0n the other hand, Jeffers (1960) used no discern- ible control. Apparently, Zerlin (1962) and Punch and Ciechanowski (1977) assumed their passages to be equivalent in reading difficulty. The passages were presented, however, as listening rather than reading tasks, and the degree to which readability predicts listenability is unclear (Klare, 1963). Although no study has investigated the sensitivity of quality judgments to differences in stimulus materials, some have failed to establish approximate equivalency of stimulus materials. The effects of talker sex differences upon quality judgments were discussed by Witter and Goldstein (1971) and Punch and Ciechanowski (1977). Witter and Goldstein reported differences in the proportion of prefer- ences for a given hearing aid as a function of whether the stimuli were presented by a female voice or a male voice. It was noted, however, that the proportion of preferences also depended upon the particular pair of hearing aids being compared, raising the issue of hearing aid-voice interaction. Punch and Ciechanowski (1977) reported statistically signi- ficant (p < 0.05) Pearson product-moment correlations between quality preferences for (1) male and female voices (0.72 for normal listeners and 0.89 for sensorineurally impaired listeners), (2) male voice and music (0.92 for normal listeners and 0.89 for sensorineurally impaired 9 listeners), and (3) female voice and music (0.89 for normal listeners and 0.94 for sensorineurally impaired listeners). Each correlation was said to account for an acceptable preportion of variance, and it was concluded that overall hearing aid-stimulus interaction was absent. Reliability. Jeffers (1960) concluded that quality judgments made by conductively impaired listeners are reliable. However, this conclu— sion was based upon intersubject agreement rather than test-retest agree- ment. In this case intersubject agreement probably represents the sensitivity of quality judgments to differences in hearing aid performance more than the repeatability of quality judgments. Zerlin (1962) performed no statistical test of paired-comparison judgment reliability. However, on the basis of a table of test-retest comparisons of hearing aid prefer- ence ranks by individual sensorineurally impaired listeners, he noted that reliability appeared to be "encouraging." Witter and Goldstein (1971) and Yonovitz 35 El- (1978) did not report specific reliability data. Witter and Goldstein (1971) did note that a partial replication of their experiment yielded similar results. Weldele and Millin (1975) also did not report test-retest data. Smaldino (1974) reported a fairly strong Pearson product-moment correlation (r = 0.827) between test-retest quality judgments, indicating good reliability with normal hearing listeners. Punch and Ciechanowski (1977) obtained statistically significant (p < 0.05) Pearson product- moment correlation coefficients for the test-retest comparisons made by the normal hearing listeners and the dysacusic listeners on each stimulus condition. The reliability coefficients for the normal hearing group were: 0.86 for the male voice, 0.69 for the female voice, and 0.56 for the music stimuli; coefficients for the dysacusic listening group were: 0.85 for the male voice, 0.54 for the female voice, and 0.35 for the music. 10 The authors concluded that only the male voice stimulus resulted in acceptably reliable data for clinical purposes with dysacusic patients. Chial and Daniel (1977) reported test-retest correlation coefficients (Spearman rank order correlations) of 0.98 for normal listeners and 0.94 for dysacusic listeners. Feasibility issues. The use of paired-comparison quality judgments to determine the rank order of hearing aids can be an excessively time consuming affair. Punch and Ciechanowski (1977) indicated that reli- ability may be affected by the talker, while Chial and Daniel (1977) suggested that the judgments of dysacusic listeners are less consistent when signal quality is low. Thus, it is probably wise to obtain multiple observations for each paired comparison. In addition, the determination of rank order required the comparison of each aid with every other aid. Quality Magnitude Estimation One alternative to paired comparisons is quality magnitude estimation. Magnitude estimation is a procedure by which the quality of signals may be assessed singly. The listener's task is to assign a numerical value to each of several signals. Sometimes the stimulus set includes a pre- selected reference signal to which a specific value (i.e., a modulus) has been assigned. In other cases the listener might simply be asked to assign his own numerical value to a given stimulus and then rate the other stimuli relative to that one. Magnitude estimation is a procedure sug- gested by S. S. Stevens (1957) for obtaining ratio data. According to Stevens, discrimination of sensations can be accomplished by less strin— gent scaling methods (e.g., ordinal scaling), but sensation magnitude is directly and validly measured only by ratio scaling. ' Chial and Daniel (1977) employed a quality magnitude estimation ll procedure which included three preselected reference signals. Since the listeners used a graphic equal-appearing interval scale to assign integer values from zero to 10 to each of the stimuli, they produced interval data. Chial and Daniel compared the reliability and sensitivity of a paired-comparison method and their "magnitude estimation method" for measuring the quality of hearing aid transduced speech as perceived by normal and dysacusic listeners. It was found that both methods produced reliable data. Normal and dysacusic listeners expressed similar prefer— ences and similar quality magnitude estimates for better quality signals, but when signal quality was low, dysacusic listeners were obviously less consistent in expressing preferences and in estimating quality magnitude. Correlations among various electroacoustic measures and group performance on the two quality measurement tasks were not significant; this finding was interpreted as failing to confirm or refute claims about the sensitiv- ity of the paired comparison and magnitude estimation methods to measur- able electroacoustic differences. However, it was suggested that direct quality magnitude estimation procedures may provide information useful to the understanding of how impaired listeners process complex signals. Quality judgments represent psychological responses to physical stimuli. The problem is one of discovering a simple equation which de- scribes the relationship between the physical parameters of stimuli and subjective reactions to them. S. S. Stevens (1975) reviewed many of his earlier magnitude estimation experiments in which psychophysical functions were often displayed as straight lines when both the sensation and stim- ulus magnitudes were plotted as logarithmic coordinates. Stevens' data conformed to a straight line equation, log V = 8 log ¢ + log K where w represents the estimated psychological magnitude and o represents 12 the stimulus magnitude. Beta (8) represents the slope of a line and K is a constant. Changes in different physical stimuli lead to different sub- jective reactions and therefore, to different beta terms. Thus, beta becomes a dependent variable suggesting differences in perceptual events for different physical parameters. Taking the antilogarithms, the equa- tion becomes ¢=K¢B where K is a scaling factor equal to the intercept. In this form the equation represents a power function. S. 8. Stevens (1957) proposed that this relationship represents a psychophysical law which states that equal stimulus ratios produce equal sensation ratios. S. S. Stevens and Galanter (1957) described two general classes of perceptual continua of sensory magnitude, "prothetic" and "metathetic" continua. How continua are classified is determined by how they behave in psychophysical experiments. Prothetic and metathetic continua are thought to be mediated by different physiological processes. Prothetic continua are thought to be associated with changes in sensation magnitude resulting from the addition or subtraction of neural excitation, whereas metathetic continua are thought to be associated with changes in the quality or spatial location of sensation resulting from the substitution of one form of neural excitation for another. Perceptual continua on which subjects make judgments of "how much" (e.g., heaviness, brightness, loudness) belong to the prothetic class of continua. Perceptual continua on which subjects make qualitative judgments of "what kind" and "where" (e.g., pitch, apparent position) belong to the metathetic class of con— tinua. In general, prothetic processes produce perceptual judgment scales in accordance with Stevens' power law, whereas, metathetic proc- esses generally result in judgments that are less orderly (S. S. Stevens, 13 1957). While loudness sensation may be represented as a power function of stimulus intensity, a scale of pitch sensation assumes a curvilinear form. Relevant Constructs Regardless of the psychophysical method employed, quality judgments have normally been used to evaluate speech transmission systems. The signal system, however, is only part of the auditory communication proc- ess. Distortion at any point in the transmission or "reception" of a speech signal may influence the signal's auditory perception to some de— gree. This concept is conveyed by two models. According to information theory (Shannon and Weaver, 1949, 1963), I (amount of information) = 2 t w log (S + N/N) where t is the signal duration, w is the width of the usable frequency range, S is the maximum amplitude of the signal, and N is the minimum discernable intensity difference. Information may be viewed according to what is "received" as well as what is transmitted. Lassman (1964) considered a "noise interference" model in which a hearing aid is the signal transmission system. According to Lassman's model, S/N environment I: Nhearing aid + Nperipheral + N . central auditory auditory system system where I is intelligibility, S is signal magnitude, and N is noise in the information theory sense (i.e., anything that increases signal ambiguity). Neither the information theory model nor Lassman's model has been directly related to speech quality judgments, and no claim is made for the valid- ity of either of them. These models are presented only to suggest a re— lationship between the effects of degradation in the auditory system and 14 in signal transmission systems (e.g., changes in the amount of information passed, degree of intelligibility, and perhaps changes in speech quality). Apparently, there are no widely accepted theoretical models of speech quality. McGee (1965) discussed two "theories" of speech quality which appear to be insufficiently developed to warrant their appellation. One was a theory of intelligibility based on the articulation index described by French and Steinberg (1947), Fletcher and Galt (1950), and ANSI 53.5—1969. The articulation index involves the construction of a ratio scale from interval data, i.e., percentage scores on articulation tests. In apparent disdain for the intelligibility theory, McGee noted that a perceptual study of speech quality requires a different response by the subject than a written report of what is heard over a speech transmission circuit. The second theory referred to was that of Ochiai and Fukamura (1953, 1956), which McGee called a "vocalic voice" theory. Based upon a thorough analysis of five Japanese vowels, this theory was said to con- sider a naturalness factor ("vocalic quality") as well as an intelligi- bility factor ("phonal quality"). The perception of articulation quality was thought to be more directly associated with select portions of the speech spectrum, while the perception of naturalness quality was thought to have a more subtle association with the entire spectrum. Statement of the Problem There has been very little systematic research on how the overall quality of complex signals is processed by listeners with normal hearing and listeners with sensorineural hearing loss. The research by audiolo- gists has primarily involved hearing aid transduced connected discourse. Although several studies have demonstrated interest in the use of speech quality judgments as a clinical tool, there are only two systematic 15 studies, Punch and Ciechanowski (1977) and Chial and Daniel (1977), which included sensorineurally impaired listeners. Only one of these two, Chial and Daniel (1977), investigated an alternative psychophysical method to paired-comparisons. This study used ordinal scaling. No study has employed Stevensonian ratio scaling to examine speech quality magni- tude estimates as a function of subject groups, types of degradation, and degree of degradation. Purpose This study was designed to examine psychophysical functions obtained on listeners with normal hearing and listeners with sensorineurally im- paired hearing. Speech quality magnitude estimates (SQMEs) were obtained on the two groups of listeners as a function of three types and seven degrees of signal degradation. The log geometric means of two within- session SQMEs were plotted as a function of log degree of degradation for the two listener groups under the three degradation types. That is, the data were plotted in a log-log space as suggested by S. S. Stevens (1957). These log-log functions were examined to answer the following questions: 1. Is there a statistically significant trend for the log geometric mean SQMEs to be influenced by changes in degree of degradation for each listener group under each degradation type? 2. If a statistically significant trend is present, what is the low- est order equation required to provide a satisfactory (i.e., statistically significant) fit to the data obtained for each listener group under each degradation type? 3. Is there a statistically significant difference among the lepes of the log-log functions as a function of: a. listener group (i.e., hearing acuity), 16 b. degradation type, or c. the interaction of listener group and degradation type? CHAPTER 11 METHOD Subjects Twelve normal hearing listeners (Group 1) and 12 sensorineurally impaired hearing listeners (Group 2) participated in the study. Ages of the Group 1 listeners ranged from 22 to 28 years with a mean of 23.58 years; ages of the Group 2 listeners ranged from 18 to 49 years with a mean of 32.58 years. For Group 1 the average threshold for 500 and 1000 Hz ranged from 0 to 5 dB HTL (re: ANSI 83.6-1969) with a mean of 0.42 dB; for Group 2 the two-frequency average threshold (Fletcher, 1950) for the better ear ranged from 15 to 50 dB HTL with a mean of 33.50 dB. In gen- eral, the Group 2 listeners had hearing losses which were more severe for frequencies greater than 1000 Hz. Appendix A shows the age and audio- metric data for groups and individual subjects. Normal Hearing Listeners The listeners in Group 1 demonstrated normal hearing by: (1) passing a pure tone screening test at hearing levels of 15 dB (re: ANSI 83.5-1969) for 250, 2000, 4000, and 6000 Hz; (2) exhibiting hearing threshold levels better than 15 dB (re: ANSI 83.6-1969) at 500 and 1000 Hz; (3) exhibiting normally shaped tympanograms indicating normal middle ear pressure (Jerger, 1970); (4) exhibiting acoustic reflex thresholds at hearing levels greater than 60 dB and less than 110 dB (re: ANSI 83.6-1969) at 500, 1000, and 2000 Hz; (5) exhibiting the ability to sustain stable acoustic reflexes for 17 18 10 seconds at 500 and 1000 Hz (Anderson, Barr, and Wedenberg, 1970); (6) reporting no history of otologic surgery, family hearing loss, recent upper respiratory problems, vertigo, tinnitus, or hearing loss. In addition, a speech discrimination score of 90Z or better was required in the test ear on a commercial version (Auditec of St. Louis) of the Northwestern University Auditory Test Number 6 (NU Auditory Test No. 6). The test was presented at 40 dB above the two-tone average threshold (Fletcher, 1950). Wilson, Coley, Haenel, and Browning (1976) concluded that for clinical purposes the Auditec and original Northwestern versions of the test were equivalent. Listeners with Sensorineural Hearing Loss The impaired listeners (Group 2) were initially selected from the clinic records of a hearing and speech clinic. They exhibited sensori- neural hearing loss bilaterally, as indicated by: (1) hearing threshold levels greater than 15 dB at two or more test frequencies (250, 500, 1000, 2000, 4000, and 6000 Hz) as measured by pure tone air conduction tests (Carhart and Jerger, 1959); (2) normally shaped tympanograms with normal middle ear pressure (Jerger, 1970); (3) the absence of market tone decay on a tone decay test (Olsen and Noffsinger, 1974) at 1000 and 4000 Hz; (4) no history of otologic surgery, recent upper respiratory prob- lems, vertigo, or active tinnitus at the time of testing. In addition, the speech discrimination score in the ear with the better two-tone average threshold was within the 60-90Z range on the NU Auditory Test No. 6 presented at a sensation level of 40 dB above the two-frequency average threshold (Fletcher, 1950). l9 Stimuli Speech Stimuli The speech stimuli represented different types of signal degradation and different degrees of degradation within each type. Talker. Because female talkers seem to accentuate quality differences (Punch and Ciechanowski, 1977), the stimuli were spoken by a female speaker of General American dialect. The following instructions were given to the talker: 1. Use a normal inflectional pattern that is not flat or monotonous. 2. Use normal linguistic emphasis. 3. Do not follow a rhythmic pattern. 4. Use a normal speaking rate. 5. Speak within your normal fundamental frequency range. 6. Peak the VU meter at zero. 7. Always speak from the same physical position (i.e., sitting or standing). 8. Maintain a constant mic-mouth distance of one hand-span. Stimulus materials. Following the example of other studies involv- ing listening tasks (e.g., Zerlin, 1962; Powers and Speaks, 1973; Chial, 1973; Speaks and Trooien, 1974; Gray and Speaks, 1977; Chial and Daniel, 1977; Punch and Ciechanowski, 1977), this study used orally presented passages of continuous discourse exerpted from reading materials as stimuli. The passages were chosen from three consecutive chapters of a junior high school history test (Wilder, Ludhum, and Brown, 1954). It was felt that this would decrease the diversity in writing style and increase the probability that the passages would be roughly equivalent in readability on a fairly easy level. In addition to the procedure followed by previous 20 researchers, the passages were evaluated by Fang's (1966, 1967) "Easy Listening Formula" (ELF). An "average" ELF score equals the number of syllables above one per word in a sentence divided by the number of sen- tences. Citing differences between materials prepared for listening (newscasts) and materials prepared for reading (newspapers), Fang con- cluded that an "average" ELF score below 12 is considered desirable for mass listenability. In the present investigation the ELF was simply used as an index by which the variance in listening ease might be reduced. No claim is made for the reliability or validity of the ELF. Initially, the ELF was applied to a pool of 93 different passages. Since there is not universal agreement on the definition of a syllable, 17 of the passages were selected for evaluation by three independent raters. Selections were made on the basis of similar ELF scores and a spoken duration of about 10 seconds. At least two of the three indepen- dent raters assigned the same score to 14 (82Z) of the 17 passages. Of these 14, six were selected to serve as stimuli. Two of the three raters gave an ELF score of 8.5 to five of the six passages and an ELF score of 9.0 to the remaining passage. The absolute deviation among the three raters was 0.5 for the six passages selected. Appendix B is a transcript of the six stimulus passages. Types 2f_signal degradation. Three degradation types for which arti- culation and "immediate" intelligibility data are available were selected: (1) low-pass filtering (French and Steinberg, 1947; Chial, 1973), (2) high-pass filtering (French and Steinberg, 1947; Chial, 1973), and (3) linear rectification (Licklider, 1946; Chial, 1973). Degrees g£_degradation. Marks (1974) noted that the choice of stimu- lus range and stimulus spacing over a region of the stimulus scale should be based on the range and spacing of sensory magnitudes rather than 21 stimulus magnitudes. Insofar as possible, sensory data were employed to determine degrees of degradation. In the magnitude estimation functions reported by S. S. Stevens (1975), five to eight points were usually plotted. Thus, it was felt that seven degradation levels should provide sufficient data to permit estimation of appropriate functions. For each degradation type, one of the seven degradation levels was a "no" degrada- tion condition. French and Steinberg (1947) reported syllable articulation data which were obtained at constant intensities and plotted as a function of the cutoff frequency of a low-pass and a high-pass filter. In the pres- ent study an all-pass or "no" degradation condition was arbitrarily assigned an articulation score of lOOZ for the low-pass and high-pass "filtered" speech. Thereafter, cutoff frequencies were interpolated from French and Steinberg's data for syllable articulation decrements of 15Z (i.e., 85, 70, 55, 40, 25, and lOZ). The cutoff frequencies for low-pass filtering were 3000, 2000, 1700, 1350, 1000, and 600 Hz. The cutoff frequencies for high-pass filtering were 1400, 1850, 2300, 2800, 3000, and 3500 Hz. It was felt that the choice of cutoff frequencies which produce approximately equal decrements in articulation scores might lead to conclusions regarding the intelligibility "theory" of speech qual- ity referred to by McGee (1965). Degrees of linear rectification were chosen on the basis of percent total harmonic distortion (THD) produced by signal rectification. Start- ing with less than lZ THD (the "no" degradation condition), seven magni- tudes of THD were selected: lZ, 10%, 20Z, 30Z, 40Z, 50Z, and 60Z. Both Licklider (1946) and Chial (1973) included a half—wave rectification condition, which Chial determined to represent approximately 40Z THD. Thus, the selected THD values were directly related to previous data by 22 the 40Z THD value. Master recording gf_undegraded stimuli. The following apparatus was used to produce a master recording of the undegraded speech passages. A microphone (Electrovoice model RE-lS) and a VU meter were located in the doubledwalled test room of a sound suite (IAC 1200 Series). The output of the microphone was passed through the wall to one channel of a VU meter bridge (Teac MB-20) and associated audio mixer (Teac 2). The output of the mixer was routed to a two-track reel—to-reel magnetic tape recorder (Ampex AG-500). The output of the tape recorder was passed through another channel of the audio mixer and meter bridge to a remote VU meter in front of the speaker. The six passages selected as stimuli were spoken by a female speaker in the test room and recorded on a master tape at a speed of 7.5 inches per second. The same speaker also produced stimulus labels and headings (e.g., "Trial 1", "Item 3", etc.). Submaster recordings gf filtered stimuli. The master signal from the Ampex recorder was passed through a filter set (Krohn-Hite 3550) to a cassette recorder (Nakamichi 700 II) where it was re-recorded in vary- ing degrees of degradation by high— or low-pass filtering. Prior to recording a speech sample of any type and degree of degradation, however, the apparatus shown in Figure 1 was used to calibrate and verify the filter cutoff frequencies. The output of a swept sine generator (Bruel and Kjaer 1024) was passed through an attenuator set (Hewlett-Packard 350D) to the filter set. The input of the filter set was monitored by a frequency counter (Heath/Schlumberger SM 4100). The output of the filter set was monitored on an RMS voltmeter (Bruel and Kjaer 2607) and recorded by a graphic level recorder (Bruel and Kjaer 2305). Response curves are depicted in Figure 2 for the low cutoff settings and in Figure 3 for the high cutoff settings. 23 Swept Sine —’ Attenuator —-’- Filter Generator Set Set (Mechanical Linkage) Voltmeter I I Graphic I Level .J Recorder Figure 1. Apparatus used to calibrate and verify filter cutoff frequencies. 24 .=mco== Ace can .u: coon Ame .u: oooN Ame .Nm coed Ace .N: owns on .nx Good Amy .um coo Au=o uncommon >oao=vmum .N ou:m«m um :H zucoawmum ocoo~ oocm OOON coca Gem CON God on ON ofi _ _ L _ A _ _ L _ _ on- I 8.. TI. Ill CHI II. 0 II. o arm QMIQ.QU am 0< __ ___ ___ __ a. up ur Karsuanux aAraeIau 25 .nm cemm Auv can .Nx coon any .n: comm Amv .Nm cemu any .nm omm~ on .Nm ooe~ Amv .:w:oa: Au=u uncommon hocmsvoum .m muswwm a: ma mucosvmum cocoa coom ooo~ coo“ com com com an om 0H _ _ _ i _ _ _ _ _ _ use .m .n .o .m < on: out out o~+ gp ur Airsuanul earneraa 26 Submaster recordings of rectified stimuli. The master signal from the Ampex recorder was passed through the variable rectified portion of a custom-built CD-l speech distortion instrument described by Chial (1973). The output of the variable rectifier was passed to the Nakamichi cassette recorder. Prior to recording speech samples, however, the apparatus shown in Figure 4 was used to calibrate the rectifier for the desired percent total harmonic distortion values (Z THD). A 1 kHz signal from a sine generator (Bruel and Kjaer 1024) was fed through an attenuator (Hewlett-Packard 350D) to the variable rectifier. The output of the sine generator was monitored by a frequency counter (Heath/Schlumberger SM 4100) and by one channel of a dual beam bistable storage oscilloscope (Tektronix 5113). The output of the variable rectifier was monitored by the other channel of the oscilloscope, as it passed to a frequency analyzer (Bruel and Kjaer 2107). Thus, the undistorted and the distorted waveforms could be monitored simultaneously. The output of the variable rectifier was adjusted by approximation until the desired THD values were measured on the frequency analyzer. In addition to reading direct THD values from the frequency analyzer, the amplitudes of the first through the tenth harmonics of the 1 kHz driving signal were also obtained. This allowed computation of THD according to the standard formula suggested by ANSI 83.3-1960, 2 2 2 A2 + A3 + ... An Z THD = 100' ' 2 2 2 Al + A2 + ... AD The results of the harmonic distortion measurements (Table 1) showed good agreement between the computed THD and the direct THD. Once a desired THD value was obtained, the wave forms monitored on 27 Frequency Counter l I l I l l Sine ; Attenuator . Variable Generator I Set Rectifier I I l l l l l ‘l ' J L... 2-Channel Oscilloscope *‘1 l I I I I Frequency I Analyzer : F Figure 4. Apparatus used to calibrate the variable rectifier. 28 Table l. Harmonic amplitude measurements in millivolts and computed percentages (Z) of total harmonic distortion (THD) in the output of the variable rectifier at seven different settings. The variable rectifier was adjusted to achieve seven degrees of THD as measured by a direct method for a 1 kHz driving signal. Percent THD as Measured by the Direct Method "None" Harmonic (0.63) 10 20 30 40 50 60 1 820.00 800.00 800.00 740.00 740.00 690.00 620.00 2 6.40 75.00 160.00 225.00 310.00 390.00 460.00 3 2.90 3.40 4.90 6.70 8.50 10.50 12.50 4 2.00 15.50 32.00 46.00 63.00 78.00 92.00 5 1.80 2.40 3.20 4.50 5.50 7.00 8.50 6 1.60 6.60 13.50 19.00 26.50 33.00 39.00 7 1.50 1.90 2.80 4.00 5.00 6.20 7.50 8 1.45 4.00 7.80 11.00 15.50 15.50 22.50 9 1.40 1.70 2.60 3.50 4.50 5.60 6.70 10 1.35 2.70 5.10 7.10 10.00 12.00 14.50 Computed Z THD* 1.00 9.60 20.10 29.80 39.50 50.14 60.56 A +A +°°°A10 *Computed Z THD = 100° +000A A 10 +A l—‘N NN NN MON 29 the oscilloscope were photographed for future use. These permanent calibration waveforms (Figure 5) made it possible to reproduce the de- sired THD values. Subsequently, the variable rectifier was simply ad- justed to achieve the appropriate calibration waveform, and the percent THD was verified by the direct method on the frequency analyzer. Summary gf_submaster recordings. The master recording was used to produce nine cassette submasters which are described in Appendix C. Each of the first seven submasters consisted of comparison passages A, B, and C under one of the seven degrees of degradation for each degrada- tion type. The eighth submaster consisted of standard comparisons A, B, and C under a single degree of degradation for each degradation type. The ninth submaster consisted of labels (i.e., "Trial 1", Trial 2", "Item 1", etc.) which were dubbed directly from the master recording. Computer generated tapes. The cassette submaster recordings were used with the computer system shown in Figure 6 to generate reel-to-reel recordings. The cassette recorder fed the signal from the submaster tapes into a 3 Rivers Computer Corporation analog-to-digital converter (ADC). The digitized signal from the ADC was passed to a digital computer (Digital Equipment Corporation pdp 11/40) which was interfaced with two disk drives (Digital Equipment Corporation RKOS), a teletype terminal (Digital Equipment Corporation, decwriter II), and a video monitor. The digitized signals were processed by the computer to generate different random orders of stimuli and to control time intervals and stimulus out- put levels. The digital computer was patched to a 3 Rivers Computer Corporation digital-to-analog converter (DAC) and then to a reel-to-reel recorder. The computer system was controlled by three interactive programs: GLADC (Appendix D), RPLAY (Appendix E), and GSCALE (Appendix F). GLADC 30 10Z m /\ 50Z VV i 3“: 20% ‘\/I \\//~ 60Z Scale: 2v/unit 0.5 msec/unit Figure 5. Calibration waveforms used to obtain desired percentages of total harmonic distortion. 31 Cassette Analog-to- Tape Digital Recorder Converter j Disk Drive MViie° pdp 11/40 °“ t°r Computer T 1 t D' k D i e e ype 18 r ve Terminal Reel-to-Reel Digital-to- Tape ———" Analog Recorder Converter Figure 6. Computer system used to generate duplicate stimuli and play them back in random orders. 32 was used for analog-to-digital conversions, that is to create sound files, while RPLAY was used to convert the digital sound files to analog form and to play them in random order. A 10 kHz sample rate was used in the con- version programs. GSCALE was used to make any necessary adjustments in playback level required to achieve zero VU on the Ampex recorder. Permanent sound files were created for a 1 kHz calibration tone and for voiced headings and labels (e.g., "Trial 1", "Trial 2", Trial 3", "standard", "item", and spoken digits "1" through "7". Trial headings were used to announce the beginning of each series of seven stimulus pairs. The word "standard" followed each trial heading to introduce the standard passage for the upcoming series of stimuli. The word "item" always preceded a number (e.g., 1-7) used to designate a stimulus pair. Each pair consisted of the standard stimulus followed by a comparison stimulus. Temporary sound files were created for the standard passage and for each of the seven degradation levels of a given comparison pas- sage under a degradation type. Nine reel-to-reel recordings (Appendix C) were produced by the com- puter system. Each of these interim tapes represented a single degrada- tion type and consisted of six different random orders, or trials, of seven stimulus pairs. In each of these pairs, the standard passage always represented the middle or fourth degradation level; the compari— son passage represented one of the seven degradation levels. Final test tapes. Six final tapes were prepared by splicing timing (leader) tape and trial segments from the tapes generated by the computer. Appendix C shows the presentation orders for the three degradation types, six spoken passages, and seven degrees of degradation of the comparison passages. Each final test tape consisted of (l) a level calibration tone, (2) spoken instructions to the subject, (3) practice materials including 33 one seven-item trial for each of three degradation types, and (4) experi- mental materials including two seven-item trials for each of the three degradation types. Each trial lasted approximately 3.5 to 4.0 minutes. Approximately 15 seconds were required for the trial heading, standard heading, and standard passage. Approximately 3.5 minutes were required for the seven stimulus items (i.e., about 30 seconds for each of the 7 items). The standard and comparison passages lasted approximately 10 seconds each and were followed by a 5-second response interval. Effects pf apparatus. The final auditory stimuli received by each listener were subject to filtering imposed by all instruments used to generate and present the stimuli. The apparatus used in stimulus genera- tion was described earlier. A block diagram of the listening apparatus is shown in Figure 7. In the control room the output of a reel-to-reel tape recorder (Ampex AG-500) was routed to the tape input of a two-channel speech audiometer (Grason-Stadler 162), then through the wall to two pairs of TDH-49 earphones located in the test room. One earphone of each pair was a "dummy". Appendix G describes the measurement of frequency response curves of instruments used to generate and present auditory stimuli. Composites of these response curves (Figure 8) show that (a) the highest of the low frequency cutoffs (i.e., the 3-db down points) was imposed at 50 Hz by the TDH-49 earphones and (b) the lowest of the high frequency cutoffs was imposed at 4400 Hz by the computer system. The high frequency cutoff was due primarily to the use of an ADC sampling rate of 10 kHz. A higher sampling rate would have been desirable but was precluded by computer system limitations. These response curves show that all "undegraded" and linearly rectified speech stimuli had a low frequency cutoff of 50 Hz and 34 9 ‘\ ‘\.“\ ‘\ ‘\ V‘x ‘\\ ‘x. ‘\, ‘\ ‘\ ‘\t L\\\\\\\\\ \\ A .maumuwaam wcwcmumfia .n shaman ll ll.lu|l4 nouazm neuuoflom soaumuaaamo nouoaowv=< zoomem umvuouom mama 35 .Amlm can —Im so“ mused mfiaomv mmcozahmm mqnxoh cues uwumEofimsm nomwam New umfimmumncommuu one new Ammcwa monmmmv Eoumxm umusasoo ocu new mo>u=o mucoammu xocwaumum mufimoaeoc .w ouswam N: CH xocoacwum cocoa ooom OOON ooo~ com CON cod Om ON o~ _ I _ _ _ _ . . . . . . . Ill /\\ gp ur Aarsuanul earnerau 36 a high frequency cutoff of 4400 Hz. These cutoff frequencies, together with those imposed by the low-pass and high-pass degradation conditions, effectively determined bandwidths for all the filtered stimuli. Visual Training and Screening Stimuli Naive anul unpracticed subjects are apparently capable of yielding reliable and consistent results on magnitude estimation tasks (S. S. Stevens and Poulton, 1956; J. C. Stevens and Tulving, 1957). However, S. S. Stevens (1975) noted that it is sometimes helpful to initiate the new observer with an easy experiment such as the judgment of apparent line length or circle size. Also, it seemed important in this study to eliminate subjects who showed great difficulty with the method of magni- tude estimation. The goal was to eliminate those who had difficulty with the method, not those who had difficulty judging speech quality. Thus, a visual screening and training task seemed apprOpriate. Three sets of visual stimuli were used for the screening and training task. Each set represented a trial comprised of seven pairs of stimuli produced on 35-mm slides. Each stimulus pair consisted of two white geometric forms located side by side on a blue background. An example of how the stimuli appeared on slides is shown in Figure 9. Each slide had an identifying number centered beneath the two stimuli. The left member of the pair, the standard stimulus, was located below an "S" for standard. The right member of the pair, the comparison stimulus, was located below a "C" for comparison. The middle-sized comparison shape was the same size as the standard. The first trial consisted of squares of different areas. The second trial consisted of circles of different areas. The third trial consisted of a different ordering of the same size circles used in the second set. 37 Figure 9. Example of a stimulus slide used in the visual training and screening task. 38 Procedures Experimental procedures are summarized in Figure 10. Appendix H shows the protocol followed by the experimenter in running each subject. Audiometric Screening Each subject was asked to sign an informed consent release form (Appendix I) and to verify case history data taken earlier by telephone. Subjects were then evaluated to determine their qualifications for the study and to provide reference thresholds for the experimental tasks. Audiological test results and subject history information were recorded on a form devised for that purpose (Appendix J). With the exception of the speech discrimination tests, all hearing tests were administered to each ear. Normal hearing listeners received a pure tone screening test; hearing impaired listeners received a thresh- old test (Carhart and Jerger, 1959). Normal listeners were tested for reflex decay at a sensation level of 10 dB (re: acoustic reflex thresh- old) at 500 and 1000 Hz; hearing impaired listeners received an audio- metric tone decay test (Olsen and Noffsinger, 1974) at 1000 and 4000 Hz. Tympanograms were plotted manually for all subjects, taking compliance measurements at pressure increments of 100 mm H20. For each subject, the "better" ear was designated as that ear which produced the lowest two-frequency pure tone average threshold. In the absence of interaural difference, the right ear was arbitrarily selected as the test ear. A speech discrimination score was obtained for the test ear by presenting the NU Auditory Test No. 6 at 40 dB sensation level (re: the two-tone average threshold). Eipirimizc:r Subject Signs Audiometric Criteria n erv w Release Form Screening Mfit Subject 39 Dismiss Subject Screening Experimenter Criterion and Prepares Met Training Visual Task Stimuli Dismiss Subject ‘ l Experimenter Experimenter Subject Prepares Practices Calibrates SQME A aratus SQME Stimuli pp Task SQME Dismiss L335 Experimental ? Start at for Second Session. Figure 10. Procedural flowchart. 40 Visual Magnitude Estimation Training and Screening Following the audiometric testing, a lS-minute program of audio- taped instructions and stimulus slides was used to teach the subject how to perform a magnitude estimation task and to assess the consistency of the subject's performance in the task. The script of the audio-visual training task is given in Appendix K. The training and screening program was conducted individually in one room of a double-walled sound suite (IAC 1200 series). The audio signal was presented through earphones at a comfortable loudness level. Each subject estimated the apparent magnitude of various squares and circles which were projected on a rear-screen (9" x 9") slide viewer equipped with a synchronized tape player (Singer Caramate II SP). Instructions to the subject and a sample response sheet are given in Appendix L. The visual training task was intentionally similar in form and procedure to the auditory tasks. Following a practice trial with randomly ordered squares, each sub- ject made magnitude estimates of circle size in two subsequent trials which represented different randomizations of the same stimuli. Each subject retained for the experiment was required to produce a correlation (r) between estimates for the second and third trials that was equal to or greater than 0.90. Subjects also were retained only if they assigned the same estimate to a standard stimulus and its equivalent comparison stimulus. A pilot study (Appendix M) of a separate group of subjects who met these criteria yielded results similar to those reported by S. S. Stevens (1975). 41 Listening Tasks The listening tasks consisted of SQME training and the SQME experi- ment. Listening tasks followed the visual magnitude estimation training and screening procedures and were administered to the subjects in pairs. Two subjects were located in back-to-back writing desks located in the test room. Stimuli were presented at 40 dB above the two—frequency aver- age threshold (Fletcher, 1950). Calibration pf listenigg apparatus. The listening apparatus (Figure 7, p. 34) enabled the same signal to be delivered at different intensi- ties to each of two subjects. A calibration selector switch allowed the examiner to monitor the taped calibration tone routed to either experi- mental earphone. The signal level to the earphones was checked in this manner prior to each experimental session. In addition to the within-session calibration, the tape recorder, speech audiometer, and earphones were checked before and after the in- vestigation. On both occasions, the system performed within the toler- ances specified by ANSI 83.6-1969. SQME training. Instructions for the listening tasks were given orally (via tape recording) and in writing. The written instructions and samples of the accompanying response sheets are shown in Appendix N. Subjects were instructed to assign any numerical value which seemed appro- priate to the standard stimulus for each trial and to assign a related numerical value (i.e., a magnitude estimate of speech quality) to each comparison stimulus. In other words, the training task, like the experi- ment itself, employed a free modulus paradigm. The instructions were followed by 21 practice items (seven degrees of each of three types of degradation). The order of degradation types 42 was counterbalanced across pairs of subjects. The same standard passage and the same comparison passage were used for all practice trials, but the order of comparison degradation levels varied randomly within each trial so that a given order was never repeated. The SQME training lasted approximately 15 minutes and was followed by a 5—minute break. SQME experiment. Practice stimuli were followed by experimental stimuli on the same reel. The experimental tasks were conducted in the same manner as the practice task. Subjects performed speech quality magnitude estimates on 42 items (two seven-item trials of each of the three degradation types). The order of degradation types was counter- balanced across pairs of subjects, and the order of stimulus passages was counterbalanced across trials. Again, the order of degradation levels varied randomly within each trial so that a given order was never repeated. The total stimulus time was approximately 21 minutes (3.5 minutes for each of 6 stimulus sets). Subjects received a two-minute break after every two trials, that is, after listening for about 9 minutes. In other words, the subjects completed their judgments on all samples of a single degradation type prior to taking a break. Second Listening Session Eight listeners were randomly selected to participate in a second listening session. Four normal hearing listeners and four hearing impaired listeners returned within one to seven days to repeat the training and the SQME experiment. 43 Summary A summary of the experimental events and their time requirements is shown in Table 2. The primary screening, training, and listening session lasted approximately 50 minutes. 44 Table 2. Summary of events and their time requirements. Event Time Required (Minutes) Audiological Screening History (taken by phone) Pure tone air conduction testing Tone decay testing Discrimination testing in test ear Impedance testing Break Visual Magnitude Estimation Training and Screening *Speech Quality Magnitude Estimation Training Break Speech Quality Magnitude Estimation Experiment Stimulus sets 1 and 2 Break Stimulus sets 3 and 4 Break Stimulus sets 5 and 6 All events 12 15 \DNONQ 100 minutes *Second session begins with retraining for SQME. CHAPTER III RESULTS Introduction The purpose of this study was to examine speech quality magnitude estimation (SQME) functions of listeners with normal hearing and listen- ers with sensorineural hearing loss. The SQMEs were obtained as a func- tion of seven degrees of each of three types of degradation. Twelve normal hearing subjects and 12 hearing-impaired subjects met all selection criteria for the study. Two additional hearing-impaired subjects were eliminated because they did not demonstrate adequate con- sistency in the visual magnitude estimation task. Initially, the listeners were trained in the use of ratio scaling to perform magnitude estimations of square and circle size. They were then instructed in magnitude estimation of speech quality and presented with 21 practice stimuli (seven degrees of each of three types of de- gradation). Practice stimuli were followed by 42 experimental stimuli (seven degrees of each of three types of degradation for two trials). Both the practice and the experimental stimuli consisted of connected speech samples which had been degraded by low-pass filtering, high-pass filtering, and linear rectification. Four subjects were randomly selected from each group to participate in a second experimental seSsion. The second session was conducted in the same manner as the first, beginning with instruction in speech qual- ity magnitude estimation. Log geometric mean SQMEs obtained over trials were plotted as a function of log degree of degradation. The log-log data were examined to answer the following experimental questions: 45 46 1. Is there a statistically significant trend for log geometric mean SQMEs to be influenced by changes in degree of degradation for each listener group under each degradation type? 2. If a statistically significant trend is present, what is the lowest order equation required to provide a satisfactory (i.e., statis- tically significant) fit to the data obtained for each listener group under each degradation type? 3. Is there a statistically significant difference among the slopes of the log-log functions as a function of: a. listener group (i.e., hearing acuity), b. degradation types, or . c. the interaction of listener group and degradation type? Data Reduction The experimental questions deal with (a) the log geometric means of magnitude estimates obtained for two trials and (b) the slopes of the functions relating the log geometric mean estimates to log stimulus values. Thus, it was necessary to reduce individual magnitude estimates to log geometric magnitude estimates and slope terms. This was accom- plished through the use of MAGEST, a Fortran IV computer program for analyzing magnitude estimation data (Kerst, 1978). The inputs required by this program include the perceptual magnitude estimates and the values of the stimulus magnitudes. The next two sections describe operations on the auditory stimulus values and the SQMEs from the listening experiment. The procedures em- ployed with the SQMEs were also followed to reduce data from the visual training and screening task. 47 Auditory Stimulus Magnitudes It was desirable to express the stimulus magnitudes in a manner that required their numerical values to change in the same direction relative to changes in degradation across degradation types. Thus, auditory stim- ulus magnitudes were expressed as frequency bandwidths for the filtered stimuli and as percent undegraded values for the linearly rectified stim- uli. (Note that the values of the visual stimulus magnitudes were appro- priately expressed in their original form, square inches.) Table 3 summarizes the low-pass and high-pass cutoff frequencies with their respective bandwidths and the percentages of THD with their respective percent undegraded values. Bandwidths of low-pass filtered stimuli were determined by subtracting the lower frequency limit of the listening apparatus (50 Hz) from each low-pass cutoff frequency. The "no" degradation value was determined by subtracting 50 Hz from the upper frequency limit of the computer system used in stimulus preparation (4400 Hz). Bandwidths of high-pass filtered stimuli were determined by sub- tracting each high-pass cutoff frequency from 4400 Hz (the upper fre- quency limit of the computer system used in stimulus preparation). The "no" degradation value was determined just as it was for the low-pass filtered stimuli, that is, by subtracting 50 Hz from 4400 Hz. Values for percent undegraded by linear rectification were determined by subtracting the measured percentages of THD from 100Z THD. The "no" degradation level (0.63Z THD) was called lZ THD or 99% undegraded. Numerical values for degrees of degradation were entered into the MAGEST program separately for each degradation type. MAGEST transforms these values by taking the natural log of each. The log stimulus values are retained by the program as independent variables to be used in 48 Table 3. Degrees of degradation expressed as (l) stimulus cutoff fre- quencies and their respective bandwidths for the filtered stimuli and (2) percentages of total harmonic distortion (Z THD) and their respec- tive percent undegraded values for linearly rectified stimuli. The seven levels of each degradation type are listed from left to right in order of increasing degradation. Degrees pthegradation py_Low-Pass Filtering Low-pass cutoff frequency (Hz) 4400* 3000 2000 1700 1350 1000 600 Low-pass bandwidth (Hz) 4350 2950 1950 1650 1300 950 550 Degrees p£_Degradation.pyuflighjpass Filtering High-pass cutoff frequency (Hz) 50* 1400 1850 2300 2800 3000 3500 High-pass bandwidth (Hz) 4350 3000 2550 2100 1600 1400 900 Degrees pf Degradation py_Linear Rectification Percent (Z) THD 1Z* 10Z 20Z 30Z 40Z 50Z 60Z Percent (Z) undegraded 99Z 90Z 80Z 70Z 60Z 50Z 40Z it "no" degradation 49 subsequent processing of SQMEs obtained from individual subjects. SQMEs SQMEs were tabulated from individual response sheets for each sub- ject and each listening condition. They were then coded and logged on punch tape to create data files for subsequent processing by MAGEST. MAGEST computes geometric mean and log geometric mean SQMEs across trials for each subject and for each degree of signal degradation. The geo- metric mean (G.M.) is defined as G .M. = {fixl} (x2) . . .(xn) where xn is the nth score. The geometric mean is the preferred index of central tendency (S. S. Stevens, 1975) because (a) it is consistent with the underlying scale of measurement (ratio) and (b) it is relatively insensitive to the effects of modulus differences across subjects or trials. MAGEST also applies the method of least squares to determine the slope and intercept of the linear equation which relates the log geo- metric mean SQMEs and log stimulus values. These computations were accomplished both for individual subjects across trials and for listener groups across subjects. Other MAGEST outputs are described by Kerst (1978). MAGEST was executed several times to produce log geometric mean SQMEs and slopes for individual subjects and for groups of subjects. The program was run separately for all combinations of degradation types, trials, and groups. Statistical Procedures In a magnitude estimation task, the values of magnitude estimates or log geometric mean magnitude estimates across trials are directly 50 affected by the numerical value of the standard stimulus, that is, the modulus. The slopes of the linear equations which relate the log magni- tude estimates and log geometric mean magnitude estimates to log stimu- lus values, however, are relatively independent of the modulus chosen. Since all magnitude estimates were made in a modulus-free manner, only within-subject analyses were done on the magnitude estimates and log geometric mean magnitude estimates across trials. Both within- and between-subject analyses were done on lepes. A significance level of 0.05 was used in all statistical tests. Reliability Procedures for SQME Data Pearson product—moment correlation coefficients (Linton and Gallo, 1975, pp. 347—352) were computed to assess reliability within and be- tween experimental listening sessions. Within sessions. Within-session correlation coefficients were ob- tained for variables underlying the dependent variables in Session 1 and Session 2. Correlation coefficients were obtained between each of the 24 subject's Trial 1 and Trial 2 SQMEs under each degradation type in Session 1. Thus, coefficients between Trial 1 and Trial 2 SQMEs for the two subgroups (n = 4) were available as subsets of the data for Session 1; additional subgroup coefficients had to be determined only for Ses- sion 2. Correlation coefficients for groups and degradation types were transformed to Fisher 2 scores (Hays, 1963, pp. 680-681), which were summed and divided by the appropriate number of scores to determine mean Z scores. The mean 2 scores were then transformed to correlation coef- ficients to obtain "average" coefficients. Between sessions. Between-session correlation coefficients were obtained for the dependent variables considered in the experimental 51 questions. The dependent variables were log geometric mean SQMEs across trials and the slopes of the lines relating the log geometric means to log stimulus values. Between~session data were available only for those four subjects in each group who returned to participate in a second lis- tening session. Each subject's seven log geometric mean SQMEs for Ses- sion 1 were correlated with those for Session 2 under each of the three types of degradation. The slopes obtained in Session 1 were correlated with those obtained in Session 2 for each listener group (n = 4) under each degradation type. "Average" between-session correlation coeffi- cients were determined by interpolating Z to r transformations from mean Fisher's r to Z transformations. Analysis Procedures for SQME Data Perceptions of speech quality were examined first by analyzing the log geometric mean SQMEs as a function of log stimulus values, and sec- ond, by analyzing the slopes of the log-log functions. The log geometric means were examined separately for each of the two listener groups within each degradation type. This plan called for six one-way mixed—effects analyses of variance (ANOVA) with repeated measures on the seven degradation levels within a degradation type (Winer, 1971, pp. 261-268). These analyses made it possible to determine whether mean log geometric mean SQMEs differed as a function of degrada- tion level. If an ANOVA showed a statistically significant trend for the log geometric means to be influenced by changes in degree of degrada- tion, a test for linear trend (Winer, 1971, pp. 296-300) was used. The orthogonal coefficients required for this test were computed according to the method described by Kirk (1968, pp. 513-517) since Winer did not describe a procedure for use with unequal intervals of the independent 52 variable. Tests for higher order trends were used if a linear equation failed to provide a statistically significant fit to the data, or if the linear equation failed to account for more than half the variance. The percentage (Z) of variance accounted for by the linear component was estimated roughly by the following formula, SSlin ). degradation level Z Variance = 100 ( The slopes for the log-log functions were examined for statisti- cally significant differences as a function of groups, degradation types, and interactions of groups and degradation types. This plan called for a two-way ANOVA (2 X 3) with repeated measures on degradation types (Winer, 1971, pp. 518-526). In addition to the usual ANOVA summary table, two additional computations were done. The exact probabilities of Type I errors were estimated for all F ratios. A strength of associa- tion statistic (eta squared) was computed for significant F ratios (Linton and Gallo, 1975, pp. 335-337). This statistic estimates the pro- portion of sample variance in the dependent variable (slopes) which can be "explained" by the independent variables (i.e., groups, degradation types, and interactions). Where appropriate, the Newman-Keuls specific comparisons test was used to determine statistically significant differ- ences among specific pairs of means (Linton and Gallo, 1975, pp. 324-327). Procedures pp_Visua1 Training and Screening Data Computations were done on the visual training and screening data to discover possible differences among groups and subjects with respect to magnitude estimation scaling abilities. This was important since it was desirable to attribute differences in the SQME data to perceptual differ- ences rather than to differences in sophistication with the method of 53 measurement. Pearson product-moment correlations (Linton and Gallo, 1975, pp. 347-352) were used to check the within-session test-retest reliability of the visual magnitude estimates. Pearson r's were obtained between the magnitude estimates for Trials 2 and 3 for each of the 24 subjects. To check within-session test-retest reliability of slopes, a two- way ANOVA with repeated measures on Trials 2 and 3 (Winer, 1971, pp. 518-526) was computed for Groups 1 and 2. The analysis made it possible to discover any statistically significant differences as a function of groups, trials, or group-by-trial interactions. Visual Magnitude Estimation Data Description Since subjects were not confined to using a common modulus in making their estimates, typical descriptive statistics were not computed on mag- nitude estimates or log geometric mean magnitude estimates of circle size. Figure 11, however, shows the mean log geometric means of the modulus- free magnitude estimates plotted as a function of log circle area (inz) for Group 1, Group 2, and a third group tested during a pilot study of the visual training task (Appendix M). Although it was possible for each subject to assign a different value to each standard stimulus, the mean log geometric means appear to be of similar value at each level of stimulus magnitude. The data points for the three groups also appear to form relatively straight lines with roughly equivalent slopes. The re- sults of the pilot study (Appendix M) showed that a linear equation does, in fact, provide a statistically significant fit of the pilot data. Table 4 summarizes the mean slopes, standard deviations, and ranges for Groups 1 and 2 and the pilot group. Table 5 summarizes the mean Mean Log Geometric Mean VMEs 54 E. .5 5 A E 1.5’ I : E l 8 : : OEI : 1.1.b B : 0.9:- A fig ": 0.7’ ‘ t z : 1 0.5 7' -: : A :3 0.3: Q . i 1 . .4 d ...\E I H 0‘ I H N I O m I O b 0.4 Log Circle Area (inz) C) Pilot Group B Group 1 (Normal) A Group 2 (Impaired) Figure 11. Mean log geometric means of modulus-free visual magnitude estimates (VMEs) plotted as a function of log circle area (inz) for the pilot group and Groups 1 and 2. 55 Table 4. Mean slopes, standard deviations, and ranges for the pilot study group, Group 1, and Group 2. Slopes were obtained from least squares solutions for log geometric mean visual magnitude estimates as a function of log circle area (in2). If S . D . Range Pilot Study Group .717 0.141 0.459 (n = 12) Group 1 (Normal) .687 0.118 0.417 (n = 12) Group 2 (Impaired) .790 0.212 0.667 (n = 12) 56 Table 55. Mean slopes, standard deviations, and ranges for Groups 1 (normal) and 2 (impaired), Trials 2 and 3, and group-by-trial inter- actions. Slopes were obtained from least squares solutions for log visual magnitude estimates as a function of log circle area (inz). RI S.D. Range Group 1 Trial 2 0.682 0.114 0.417 Group 1 Trial 3 0.691 0.131 0.417 Group 2 Trial 2 0.787 0.227 0.701 Group 2 Trial 3 0.794 0.203 0.632 Trial 2 0.734 0.184 0.705 Trial 3 0.743 0.176 0.683 Group 1 0.686 0.121 0.417 Group 2 0.790 0.211 0.701 57 slopes, standard deviations, and ranges for Groups 1 and 2, Trials 2 and 3, and the group-by-trial interactions. Reliability Table 6 displays the Pearson product-moment correlation coefficients (r) between each subject's Trial 2 and Trial 3 visual magnitude estimates of circle size (inz). All coefficients were significant beyond the 0.05 level (df = 5; r = 0.754). Coefficients ranged from 0.98 to 1.00 critical for the normal hearing subjects (Group 1) and from 0.94 to 1.00 for the hearing-impaired subjects (Group 2). The coefficient of determination for the lowest r in each group was 0.96 for the normal hearing group and 0.88 for the impaired group. Thus, a statistically significant, strong positive relationship existed between each subject's Trial 2 and Trial 3 magnitude estimates. Most of the variance in the magnitude estimates (i.e., over 96Z for Group 1 and over 88Z for Group 2) could be accounted for by the linear relationship between Trial 2 and Trial 3 estimates. Cumulatively, these results suggest a very high degree of within-subject reliability for both groups. Table 7 summarizes the results of a two-way analysis of variance in slopes for Groups 1 and 2 with repeated measures on Trials 2 and 3. None of the observed F ratios were significant at the 0.05 confidence level. These results suggest good within-session test-retest reliability of slopes. Table 8 summarizes the results of an incidental one-way analysis of variance in lepes as a function of groups (i.e., the pilot group and Groups 1 and 2). The observed F ratio was not significant. Thus, there were no systematic differences in the slopes of the three groups on the visual training and screening task. 58 Table 6. Pearson product-moment correlation coefficients (r) between the visual magnitude estimates of circle size (in2) for each subject's Trial 2 and Trial 3 stimuli. All coefficients were significant beyond as 0.05 (df = 5; rcritical a 0.754). Group 1 (Normal) Group 2 (Impaired) Subject r Subject r 1 1.00 13 0.980 2 0.997 14 1.000 3 0.996 15 0.953 4 1.000 16 0.999 5 1.000 17 0.990 6 0.981 18 1.000 7 0.980 19 0.990 8 0.990 20 0.960 9 1.000 21 0.980 10 0.980 22 1.000 11 1.000 23 0.990 12 1.000 24 0.940 59 Table '7. Analysis of variance in slopes as a function of trials (i.e., Trials 2 and 3) and groups (i.e., Groups 1 and 2). Slopes were obtained from least squares solutions for log visual magnitude estimates as a function of log circle area (inz). F for o( for Source SS df MS F cx=.05 fobserved Between subjects 1.4277 23 0.0621 Groups 0.1290 1 0.1290 2.185 4.30* 0.154 Subjects within 1.2987 22 0.0590 groups Within subjects 0.0570 24 0.0024 Trials 0.0008 1 0.0008 0.326 4.30* 0.574 Groups X trials 0.0000 1 0.0000 0.004 4.30* 0.948 Trials X 38bj5°ts 0.0561 22 0.0026 Wlthln groups Total 1.4847 47 *From F distribution table (Winer, 1971, pp. 864-869). 60 Table 23. Analysis of variance in slopes as a function of groups (i.e., pilot group and Groups 1 and 2). Slopes were obtained from least squares solutions for log geometric mean visual magnitude estimates as a function of log circle area (inz). Source of F for (Xfor Variance SS df MS F CX=.05 F obs. Between 0.6820536E-01 2 .34lOE-01 1.295 3.347* 0.2874 Within 0.8687792 33 .2633E-Ol Total 0.9369845 35 *From F distribution table (Winer, 1971, pp. 864-869). 61 Speech Quality Magnitude Estimation Data Description Mean log geometric mean modulus-free SQMEs for Groups 1 and 2 are shown in Figure 12 for low-pass filtered stimuli, Figure 13 for high- pass filtered stimuli, and Figure 14 for linearly rectified stimuli. Lines of best fit were plotted by applying the method of least squares. Figure 15 summarizes the lines of best fit for modulus-equalized mean log geometric mean SQMEs for Groups 1 and 2 plotted as a function of log stimulus values for low-pass filtering, high-pass filtering, and linear rectification. Modulus equalization was accomplished graphically by assigning the same arbitrary value to the points on the lines of best fit which represent the mean log geometric mean SQMEs for the standard stimulus. The individually plotted functions for each degradation type appear to be relatively linear in shape, and the magnitudes of the log geometric mean SQMEs are monotonically related to stimulus magnitude (i.e., decreasing stimulus degradation). The composite graph showing lines of best fit, however, suggests that the magnitude of quality per- ception grows at different rates for the two groups under different degradation types. Table 9 summarizes the mean slopes, standard deviations, and ranges for listener groups as a function of degradation types. As before, slopes were obtained by applying the method of least squares to the log geometric mean SQMEs and log stimulus values. The smallest mean slope was obtained for Group 2 on low-pass filtered stimuli; the highest mean slope was obtained for Group 1 on linearly rectified stimuli. Figure 16 shows the mean slopes for groups plotted as a function of degradation types. The group with the highest slope varies in a manner which results 62 1.50 1.40 1.30 1.20 Mean Log Geometric Mean SQMEs 11'1111 111111111 111fll1111 1111' 111 111111111 VUVUIVVIU 'U'Vl'V'U U'fl'U'I' U'U'II'U' 'V'U'U' 1.10 2%.} / 1.00 0.90 0.80 / 0.70 / O 0.60 l 1 I l 1 L 2.5 2.7 2.9 3.1 3.3 3.5 3.7 L08 BW(Hz) for L-PF Stimuli Figure 12. Mean log geometric mean modulus-free SQMEs for Groups 1 and 2 plotted as a function of log frequency bandwidth (Hz) for low- pass filtered stimuli. Lines of best fit were obtained from least square solutions. Mean Log Geometric Mean SQMEs 63 C] 1.5 \..1 1.4 \ 1.3 1.2 \ P 1111ll111 1111'4111 111L11111x11111 m1 .2... 5 1.1 / Cl/ 46 mi.) 1.0 0.9 0.8 0.7 ‘U'l'tt‘ jifr‘t'UU YfiU'UVVU VIIIIVUVU 'U'r‘UII' 0.6 2.7 2.9 3.1 3.3 3.5 7 Log BW(Hz) for H—PF Stimuli N U! Figure 13. Mean log geometric mean modulus-free SQMEs for Groups 1 and 2 plotted as a function of log frequency bandwidth (Hz) for high-pass filtered stimuli. Lines of best fit were obtained from least squares solutions. 64 1.4 r dcuw'z ': ' (11mm : m 1.3: l/Q/fi : g " ma 1 O L. (mus) . m 1.2 r' 1 5 7 : g 1.1 ' A/E : U : / d .2 I : u 1.0 — 1 m - . E ' . O : . 3 0.9 _ ‘6“ = 1 A 0.8E- C “... a: - : P 4 92’ 0.7 ’ . E 1 0.6 [- .5 0.5 ' ‘ 1.5 1.6 1.7 1.8 1.9 2.0 2.1 Log Percent Undegraded by LR Figure 14. Mean log geometric mean modulus-free SQMEs for Groups 1 and 2 plotted as a function of log percent (Z) undegraded by linear rectification. Lines of best fit were obtained from least squares solutions. 65 V‘U‘ "" g N .3. l 2 c" /": m I. /: a 2/ ‘1 g E 1 3 : / / 1: E. E / 0 no 3. 33 -S \\ 5 . \ I B E 1 111.1111 111111114 ““l‘.“_"‘ T'VV Log Stimulus Values Figure 15. Lines of best fit for modulus-equalized mean log geometric mean SQMEs for Groups 1 (normal) and 2 (impaired) plotted as a func- tion of log stimulus values for low-pass filtering (L-PF), high-pass filtering (H-PF), and linear rectification (LR). Lines of best fit were obtained from least squares solutions. Modulus equalization was accomplished graphically by assigning the same arbitrary value to the points on the lines of best fit which represent the mean log geometric mean SQME for the middle (i.e., the standard) stimulus. 66 oom.H mNm.o mma.o Nmo.H Hm~.o om~.o MMH.H mom.o 0mm.o N a H Nwo.H mmq.o Hmo.o mw0.H mom.o mmu.o oom.o omm.o Hmw.o mom.o omN.o on.o N com.H qu.o omw.o mmm.H qu.o «QH.H msHaeHum on mo GOHuocsw m mm mMXOm some oHuumaoow on now mcoHusHom mmumscm ammoH Eouw wochuno mums mmaon .mmahu GOHumvmuwmv mo GOHuocam 6 mm AcmuHmasHv N cam AHmEuocv H maaouo MOM mowcmu mam .maOHumH>ov wumvamum .mmaon :mmz .mw mHan Mean Slope 67 cowl / (m) CB~\\\\\\\‘ g::éfl 1f\ \ (D \O UV'U'U'IT ‘VVUIVUUI 'UII‘IIfV Vii'l'fl' 1111IL111 111111111 111111111 111111111 DJ 9‘ ‘1vvvlivvv C\ 11111111 L-PF H-PF LR Degradation Type Figure 16. Mean slopes for groups plotted as a function of degradation types. Slopes were obtained from least squares solutions for log geometric mean SQMEs as a function of log degrees of degradation. 68 in a double transverse interaction. Group 2 shows the largest slope only for high-pass filtered stimuli. In Figure 17 the mean slopes for de- gradation types are plotted as a function of groups. The lepes for linear rectification are larger than the slopes for low-pass filtering by about the same magnitude for Group 1 and Group 2; the slopes for both degradation types are higher for Group 1 than for Group 2. However, when the slopes for high-pass filtering are considered, a transverse inter- action again becomes apparent. For Group 1, the smallest slope among the three degradation types is attributed to high-pass filtering; for Group 2, the greatest slope among the three degradation types is attributed to high-pass filtering. Reliability The between-session analyses directly involved the dependent vari- ables, while the within-session analyses involved variables underlying the dependent variables. Within sessions. Table 10 lists individual subject and group (n = 12) "average" correlation coefficients between SQMEs for the first and second trials in Session 1. Group average correlations also are shown graphically in Figure 18. Significant coefficients were those which equaled or exceeded 0.754. For Group 1, significant correlations were found for all but six of the 36 individual subject degradation-type combinations (two under each degradation type). The "average" coeffi- cients for Group 1 ranged from 0.90 for high-pass filtering to 0.93 for low—pass filtering. These averages suggest high within-subject relia- bility of SQMEs for Group 1 under all degradation types in Session 1. For Group 2, nine of the 36 individual subject correlations were not significant: five under low-pass filtering, one under high-pass 69 . 1 1.1 r 4 '1 P d : I 100 p \ 1 0.9 5' 3 ““1"” 1 p d “o’- 0'8 : / \ : .3 » . m 0.75- (X 5 LR '3 c: . d (U .. 1 (D b fl I z 006 p “a N i 0.5 :' 1 I : )- u 0.4 _ \ 1 . 1 0.3 :- \C> L-PF '3 I- I t 1 Group 1 (Normal) Grouv 2 (Impaired) Groups Figure 17. Mean slopes for degradation types plotted as a function of groups. Slopes were obtained from least squares solutions for log geometric mean SQMEs as a function of log degrees of degradation. 70 Table 10. Within subject correlation coefficients (Pearson r) between seven Trial 1 and Trial 2 SQMEs for each of 24 subjects (i.e., 2 groups of 12) in Session 1. "Average" coefficients were determined for groups within degradation types. Degradation Type Subject L-PF H-PF LR if: Grou ‘1 (Normal) 1 0.99 1.00b 0.89 2 0.97 0.65 0.92 3 0.37 0.92 0.89 4 0.98 0.97 0.67 5 0.96 0.77 0.94 6 0.78 0.83 0.95 7 0.90 0.90 0.95 8 0.96 0.85 0.70 9 0.36 0.86 0.86 10 0.97 0.95 0.99 11 0.94 0.88 0.90 12 0.96 0.52 0.87 '2: 0.93 0.90 0.91 0.92 Group 2_(Impaired) 13 0.85 0.99 0.91 14 0.52 0.98 0.87 15 -0.54 0.75 0.65 16 -0.37 0.63 -0.23 17 0.68 0.93 0.77 18 0.82 0.98 0.88 19 0.11 0.97 0.80 20 0.89 0.89 0.93 21 0.93 0.86 0.96 22 0.98 0.997c 0.96 23 0.92 0.94 0.96 24 0.88 0.91 0.26 ‘73 0.74 0.95 0.84 0.87 r a"Average" correlations are Fisher's Z to r transformations interpolated from mean Fisher's r to Z transformations (Hays, 1963, pp. 680-681). bChanged to r . 0.998 for conservative transformation of r to 2 since there is no transformation for r - 1.00. cInterpolated from three decimal place coefficient for conservative trans- formation of r to 2 since there is no transformation for r rounded to 1.00. 71 "Average" Correlation Coefficients (r) L-PF H-PF LR 1.0 I w _ I __ r— T" —“ 1. 0.8 - r— 0.6 b 0.4 r' 0.2 b 0.0 l l l 2 l 2 1 2 Groups (1 - normal; 2 . impaired) Figure 18. "Average" within-subject test-retest correlation coefficients between seven Trial 1 and Trial 2 SQMEs for two groups of 12 subjects under three degradation types in Session 1. "Average" correlations are Fisher's Z to r trans- formations interpolated from mean Fisher's r to Z transforma- tions (Hays, 1963, pp. 680-681). The dashed horizontal line denotes the significance criterion (R - 0.754). 72 filtering, and three under linear rectification. "Average" coefficients for Group 2 ranged from 0.74 for low-pass filtering to 0.95 for high— pass filtering. These averages suggest moderate to high within-subject reliability of SQMEs for Group 2 in Session 1. The grand mean coeffi- cients were 0.97 for Group 1 and 0.87 for Group 2. The grand grand mean was 0.89. These results suggest good "average" individual subject relia— bility of SQMEs for both experimental groups. Table 11 contains the test-retest correlation coefficients between the Trial 1 and Trial 2 SQMEs for each subject who was tested in two ses- sions. ”Average" coefficients are displayed in Table 11 and in Figure 19 for groups within sessions and degradation types. Significant correla- tions were those which exceeded 0.754. For Group 1, all individual sub- ject correlations were significant except two for a single subject under low-pass filtering and two for another subject under linear rectification. "Average" coefficients for Group 1 ranged from 0.88 to 0.97, suggesting high within-subject reliability of SQMEs for reliability Group 1 under all degradation types in Sessions 1 and 2. All but four of the Group 2 correlations, three for low—pass filtering and one for linear rectifica- tion, were statistically significant. The "average" Coefficients for Group 2 ranged from 0.78 to 0.95; these averages suggest that the within- subject, within-session reliability of SQMEs for Group 2 ranged from ade- quate to high. The grand mean coefficients were 0.93 for Group 1 and 0.90 for Group 2. The grand grand mean across groups was 0.92. These results suggest good average individual subject reliability of SQMEs in the two subgroups. Between sessions. Table 12 displays the Pearson correlation coeffi- cients between the seven log geometric mean SQMEs for Sessions 1 and 2 for each subject who was tested in two sessions. "Average" coefficients 73 Table 11. Within subject correlation coefficients (Pearson r) between seven Trial 1 and Trial 2 SQMEs for each of eight subjects (i.e., 2 groups of 4) within Session 1 and Session 2. "Average" coefficients were determined for groups within degradation types. Degradation Type L-PF H-PF LR Session Session Session Subject 1 2 1 2 1 2 1: Group 1 (Normal) 3 0.37 0.52 0.92 0.94 0.89 0.87 4 0.98 0.92 0.97 0.95b 0.67 0.62 10 0.97 0.98 0.95 1.00 0.99 0.92 11 0.94 0.97 0.88 0.89 0.90 0.95 'i? 0.93 0.93 0.94 0.97 0.92 0.88 0.93 Group 2_(Impaired) 17 0.68 0.51 0.93 0.97 0.77 0.98 18 0.82 0.65 0.98 0.94 0.88 0.89 20 0.89 0.91 0.89 0.95 0.93 0.97 24 0.88 0.86 0.91 0.92 0.26 0.91 i; 0.83 0.78 0.94 0.95 0.79 0.95 0.90 a"Average" correlations are Fisher's Z to r transformations interpolated from mean Fisher's r to Z transformations (Hays, 1963, pp. 680-681). b . . . Changed to r = 0.998 for conservative transformations of r to Z Since there is no transformation for r = l. 74 .Aenm.o I uv sowuouwuo cosmowmwawam onu mouocmc mafia Hmucouuuo; vmsmmv may .Admo Iowo.ma .moaa .mNm:v mcowumsuommamuu N ou u m.uo:mam some souu pounded Iguana acowumEHOmmamuu u cu N m.uo:m«m mum meowumaouuoo :owmum><: .N :ofimmom can H scammom cw moQNu nowumkuwov wounu nova: muoofinsm know mo mnsouw oau new mmzom N Hague can a Hague co>om coozuon mucowoww Imooo coaumaouuoo ummuoulumou woofinsmlafinuuz :owmuo><: .¢~ wuswam Avmuqmaaa a N “awake: I ~V mmaouo N N d N g N H N H _ _ l i J [I II F.1I E I." I l l' I U1] I, F F *1 E _ F f l _ ||_ _ _ m4 whim mmlg MA hml: mMIA N scammom H coammmm o.o N.o «.0 9.6 m.o o.H (1) s:ua;o;;;aog uoIIBIaxzoo “azazaavu 75 Table 12. Within subject correlation coefficients (Pearson r) between each of the eight subject's seven log geometric mean SQMEs for Sessions 1 and 2 under the three degradation types. "Average" coefficients were determined for groups within degradation types. Degradation Type Subject L-PF H-PF LR 'i: Group 1 (Normal) 3 0.97 b 0.88 0.94 4 0.997 0.93 0.67 10 0.99 b 0.96 0.96 11 0.996 0.96 0.98 'i: 0.99 0.94 0.94 0.97 Group 2 (Impaired) 17 0.80 0.97 0.97 18 0.93 0.96 0.94 20 0.95 0.99 0.98 24 0.99 0.91 0.80 ‘i: 0.95 0.97 0.95 0.96 a"Average" correlations are Fisher's Z to r transformations interpolated from mean Fisher's r to Z transformations (Hays, 1963, pp. 680-681). b . Derived from three decimal place coefficient for conservative transfor- mation of r to Z since there is no transformation for r rounded to 1.00. 76 are shown for groups and for groups under degradation types. Significant correlations were those which equaled or exceeded 0.754. All correla- tions were statistically significant, except one for a single subject in Group 1 under linear rectification. "Average" coefficients ranged from 0.94 to 0.99 with a grand mean of 0.97 for Group 1 and from 0.95 to 0.97 with a grand mean of 0.96 for Group 2. The grand grand mean was 0.96. Thus, the between-session reliability of the log geometric mean SQMEs appears to be very high for both listener groups. Table 13 and Figure 20 show the test-retest correlation coefficients between Session 1 and Session 2 slepes for the two groups of four sub- jects under the three degradation types. "Average" coefficients are shown for groups across degradation types, degradation types across groups, and for all data. Statistically significant correlations were those which equaled or exceeded 0.95. However, significance or non- significance may be of questionable value when the number of correlated pairs is only four. Perhaps it is wiser to determine when 50 percent or more of the variance in one set of scores can be accounted for from the other set of scores. This situation exists when r2, the coefficient of determination (Linton and Gallo, 1975, pp. 329-332), is equal to or greater than 0.50 (i.e., r s 3_0.71). Under this criterion, meaningful ob correlations included the Group 1 correlations for low-pass filtering and high-pass filtering and the Group 2 correlations for low-pass filter- ing and linear rectification. Meaningful "average" correlations con- sisted of the one for Group 1 across degradation types and the one for low—pass filtered stimuli across groups. 77 Table 13. Test-retest correlation coefficients (Pearson r) and "average" coefficients between Session 1 and Session 2 slopes for two groups of four subjects under three degradation types. Slopes were obtained from least squares solutions for log geometric mean SQMEs as a function of log stimulus values. Degradation Type L-PF H-PF LR 3:: Group 1 (Normal) 0.96 0.73 0.09 0.76 (n = 4) Group 2 (Impaired) 0.88 -0.52 0.91 0.65 (n = 4) 1?: 0.93 0.17 0.67 0.71 a"Average" correlations are Fisher's Z to r transformations interpolated from mean Fisher's r to Z transformations (Hays, 1963, pp. 680—681). 78 1 0 L-PF H-PF LR ‘rx 0.8 b 1F1 0.6 - 3 m 0.4- U c: Q) ...... U ...4 U4 “4 8 0.2 - U a O w-i 4.! <6 I l H E: 0.0 . , H O U 1 2 l 2 l 2 Groups (1 = normal; 2 = impaired) Figure 20. Test-retest correlation coefficients (Pearson r) between Session 1 and Session 2 slopes for two groups of four subjects under three degradation types. The dashed horizontal line denotes the significance criterion (r - 0.95). 79 Analysis Results of the primary analyses included those on (1) the presence of trends in log geometric mean SQMEs as a function of log stimulus values, (2) the nature of trends, and (3) the significance of differences in slopes. An alpha level of 0.05 was chosen as the criterion for deter- mining statistical significance. Presence of trends. Tables 14 through 19 summarize the results of analyses of variance in log geometric mean SQMEs across trials as a func— tion of log degradation levels for each degradation type. Each table shows the results for one of the two experimental groups under one of the three types of degradation. Group 1 data are shown for low-pass fil- tering (Table 14), high—pass filtering (Table 15), and linear rectifica- tion (Table 16); Group 2 data are also shown for low-pass filtering (Table 17), high-pass filtering (Table 18), and linear rectification (Table 19). In each case, the observed F ratio was significant, suggest- ing a statistically significant trend for the log geometric mean SQMEs to be influenced by changes in degree of degradation. Nature of trends. Tables 20 through 25 summarize the results of tests to determine whether a significant portion of the trends detected by the analyses of variance could be accounted for by a linear equation. Each table shows the results of a test for linear trend for one of the two experimental groups under one of the three types of degradation. First, Group 1 data are shown for low—pass filtering (Table 20), high— pass filtering (Table 21), and linear rectification (Table 22); then, Group 2 data are shown for low-pass filtering (Table 23), high-pass fil- tering (Table 24), and linear rectification (Table 25). In each case, the observed F ratio for linear trend was significant, suggesting that a 80 Table 14. Summary of analysis of variance in log geometric mean SQMEs across trials for Group 1 as a function of log low—pass filtered band- width (Hz). F for CX for Source 88 df MS F CX=.05 F . obs. Between subjects 10.7219 11 0.9747 Within subjects 4.5816 72 0.0636 Degrada- tion 3.7916 6 0.6319 52.795 2.242* ‘< 0.0001 levels Residual 0.7900 66 0.0120 Total 15.3035 83 *From F distribution table (Winer, 1971, pp. 864-869). Table 15. 81 Summary of analysis of variance in log geometric mean SQMEs across trials for Group 1 as a function of log high-pass filtered band- width (Hz). F for CXfor Source SS df MS F .05 obs. Between subjects 9.9774 11 0.9070 Within subjects 2.0433 72 0.0284 Degrada- tion 1.5396 6 0.2566 33.618 2.242* <20.0001 levels Residual 0.5038 66 0.0076 Total 12.0207 83 *From F distribution table (Winer, 1971, pp. 864-869). Table 16. 82 Summary of analysis of variance in log geometric mean SQMEs across trials for Group 1 as a function of log percent of undegraded by linear rectification. F for (X for Source SS df MS F CX=.05 F obs. Between subjects 9.0894 11 0.8263 Within subjects 2.4580 72 0.0341 Degrada- tion 1.9073 6 0.3179 38.096 2.242* <20.0001 levels Residual 0.5507 66 0.0083 Total 11.5474 83 *From F distribution table (Winer, 1971, pp. 864-869). 83 Table 17. Summary of analysis of variance in log geometric mean SQMEs across trials for Group 2 as a function of log low-pass filtered band- width (Hz). F for CXfor Source SS df MS F CX=.05 Fobs. Between subjects 8.2609 11 0.7510 Within subjects 1.2846 72 0.0178 Degrada- tion 0.6638 6 0.1106 11.763 2.242* ( 0.0001 levels Residual 0.6208 66 0.0094 Total 9.5455 83 *From F distribution table (Winer, 1971, pp. 864-869). 84 Table 18. Summary of analysis of variance in log geometric mean SQMEs across trials for Group 2 as a function of log high—pass filtered band- width (Hz). F for CXfor Source SS df MS F CX=.05 Fobs. Between subjects 8.2438 11 0.7494 Within subjects 3.5472 72 0.0493 Degrada- tion 2.8280 6 0.4713 43.250 2.242* <:0.0001 levels Residual 0.7193 66 0.0109 Total 11.7910 83 *From F distribution table (Winer, 1971, pp. 864-869). Table 19. 85 Summary of analysis of variance in log geometric mean SQMEs across trials for Group 2 as a function of log percent undegraded by linear rectification. F for (X for Source SS df MS F CX=.05 F obs. Between subjects .8678 11 0.7153 Within subjects .3085 72 0.0182 Degrada- tion .7709 6 0.1285 15.773 2.242* <:0.0001 levels Residual .5376 66 0.0081 Total .1764 83 *From F distribution table (Winer, 1971, pp. 864-869). 86 Table 20. Results of test for linear trend in log geometric mean SQMEs for Group 1 as a function of log bandwidths (Hz) of low-pass filtered stimuli. Source of F for «for Variance SS df MS F CX=0.05 F obs. Linear trend 3.62 1 3.62 362 4.0147 <10.0001 Deviation from linear 0.96 71 0.01 trend *Interpolated from F distribution table (Winer, 1971, pp. 864-869). 87 Table 21. Results of test for linear trend in log geometric mean SQMEs for Group 1 as a function of log bandwidths (Hz) of high-pass filtered stimuli. Source of F for CXfor Variance SS df MS F CX?0.05 F obs. Linear trend 1.43 1 1.43 143 4.0147 ‘< 0.0001 Deviation from linear 0.61 71 0.01 trend *Interpolated from F distribution table (Winer, 1971, pp. 864-869). 88 Table 22. Results of test for linear trend in log geometric mean SQMEs for Group 1 as a function of log-percent (Z) undegraded by linear recti- fication. Source of F for CXfor Variance SS df MS F CX=0. 05 F obs. Linear trend 1.89 1 1.89 189 4.0147* <10.0001 Deviation from linear 0.57 71 0.01 trend *Interpolated from F distribution table (Winer, 1971, pp. 864-869). 89 Table 23. Results of test for linear trend in log geometric mean SQMEs for Group 2 as a function of log bandwidths (Hz) of lowhpass filtered stimuli. Source of F for' CXfor Variance SS df MS F CX=0.05 F obs. Linear trend 0. 65 1 0. 65 65 4.0147* < 0.0001 Deviation from linear 0.63 71 0.01 trend *Interpolated from F distribution table (Winer, 1971, pp. 864-869). 90 Table 24. Results of test for linear trend in log geometric mean SQMEs for Group 2 as a function of log bandwidth (Hz) of high-pass filtered stimuli. Source of F for CKfor Variance SS df MS F CX=0.05 F obs. Linear trend 2.69 1 2.69 269 4.0147* <20.0001 Deviation from linear 0.86 71 0.01 trend *Interpolated from F distribution table (Winer, 1971, pp. 864-869). 91 Table 25. Results of test for linear trend in log geometric mean SQMEs for Group 2 as a function of log percent (2) undegraded by linear recti- fication. Source of F for (X for Variance SS df MS F CXFO-OS F obs. Linear trend 0.76 1 0.76 76 4.0147* < 0.0001 Deviation from linear 0.55 71 0.01 trend *Interpolated from F distribution table (Winer, 1971, pp. 864-869). 92 linear equation provides a statistically significant fit to the data. Also, the linear component of the trend in each case accounted for a very high percentage (93-99%) of the variance in log geometric mean SQMEs due to log degradation levels. Thus, tests for higher order trends were not employed. Table 26 summarizes the approximate percentages of variance that can be attributed to the linear component of each trend. Differences in_lepes. Table 27 shows the results of a two-way analysis of variance in $10pes as a function of the two listener groups (Group 1 and Group 2) and three degradation types (low-pass filtering, high-pass filtering, and linear rectification). Measures were repeated on degradation types. Significant F ratios are those which exceed the critical values of F at an alpha level of 0.05. The main effect of groups was not significant. The main effect for degradation types, however, was significant, suggesting a statistically significant difference in slopes as a function of degradation types. The group-by-degradation type inter- action was also significant, indicating that the effects of degradation types and groups were interdependent. Significant effects in the analy- sis of variance were followed by a Newman-Keuls specific comparison test to evaluate differences within specific pairs of means. Table 28 shows the results of a Newman-Keuls specific comparison test on pairs of mean slopes for the three degradation types. All dif- ferences among the mean slopes for low-pass filtering, high-pass filter- ing, and linear rectification were statistically significant. Table 29 shows the results of a Newman-Keuls specific comparison test on pairs of mean slapes for the group—by-degradation type interac- tion. Nine comparisons revealed differences that were significant beyond the 0.05 level. The pattern of the important differences may be described as follows. The mean slopes for Group 1 differed from those for Group 2 93 Table 26. Approximate percentages of variance in log geometric mean SQMEs due to log degradation levels that could be accounted for by a linear equation. degradation type. Percentages (Z) are shown as a function of group and Group 1 (Normal) 1 1 2 (Impaired) 2 2 Degradation Type Low-pass filtering High-pass filtering Linear rectification Low-pass filtering High-pass filtering Linear rectification Z_Variance 96 93 99 98 95 99 94 Table 27. Results of a two-way analysis of variance in slopes as a function of two listener groups (Group 1 and Group 2) and three de- gradation types (1ow-pass filtering, high-pass filtering, and linear The slopes were obtained by applying the method of least squares to the log geometric mean SQMEs and log degrees of de- rectification). gradation. Source SS df MS F for F CX=0. 05 cxfor obs. Between subjects Groups Subjects within groups Within subjects Degradation types Groups X degradation types Degradation types X subjects within groups Total 7.1099 0.7612 6.3487 6.0174 1.9660 1.7129 2.3385 13.1273 23 22 48 44 71 0.3091 0.7612 0.2886 0.1254 0.9830 0.8565 0.0531 2.638 4.300* 18.496 3.214* 16.115 3.214* 0.119 < 0.0001 (0.0001 \ *From F distribution table (Winer, 1971, pp. 864-869). 95 .mcmoa mo game m :mmBumn monouowwwv ucmofiwficwwm m mmuoamas mHsH.o u m>o saaa.o mule mqu.- u ~>o «mos.o «oo~.o um-u x>u mmm.o om~.o omm.o mzamz MA mmlm mmla .x ou Hmsvm ma poacmmm momma mo hopes: one .mo.onqu pom A>ov m=Hm> Hmo Iauwuo ouwwumounam oSu mwoooxm ufi c053 unmowmficwwm ow mamms o3u Nam coozuwn mucouowmwv < .Amamoa mouzu no oau ..m.«v pmccmam momma mo momcmu mHnHmwoa Ham Mom cm>Hw mum moaaw> Hmowuwuo .moaxu cowumvmuwwv manna osu you moaoam :moa mo muwmm :o ummu :owwummaoo oamaomam maamxlcmasmz may mo muaammm .wN magma 96 .mamoa mo uHma m comauon mucoumwva unmoHWchHm m monocons owN.o u o>u umsN.o ems: x N usouo NGN.o u m>o «oos.o NON.o um-g x N uaouo HmN.o u «>0 «aHs.o oNH.o mHo.o an x N usouo NNN.o u m>o *SNm.o HNN.o sNfi.o mON.o emu: x N uaouu omN.o u N>o «me.o «mmm.o *mNs.o «mos.o *som.o mans x N asouo x>o sqN.N Nmm.o «SN.o ANN.o oNo.o cam.c mzam: 5 x El: x mans x 5 x Elm x m: x H asouo N nacho H moouw N macho H onouu N anouo .3 ON Hmavm mH momma mo Hanan: one .mo.o uxu qu A>Uv 03Hm> HmoHuHuo mumHuaouaam mnu mwomoxo uH cons unmoHMchHm mH momma oau mom cmmsuon mucouomme < .mcmma me Ou oau Bouw wchamaw mowcmu mHnHmmom HHm HON co>Hw mum mmSHm> HmoHuHuo .Amamoa c u mma%u aoHumcmuwow m x masouw NV GOHuomumucH wmhu aOHumvmuwovlzn Imoouw mnu pom mmaon cams mo muHma no umou somemaaoo oHMHomam mHsoMIameaoz man mo muHsmmm .aN oHan 97 as a function of low—pass filtering and linear rectification, but not as a function of high—pass filtering. Within Group 1, the mean slope for low-pass filtering differed from the mean for linear rectification, but not the mean for high-pass filtering; the mean SIOpes for high-pass fil- tering and linear rectification, however, also differed. Within Group 2, the mean slopes for low-pass filtering differed from the means for high- pass filtering and linear rectification; the mean slopes for high—pass filtering and linear rectification, however, did not differ. Mean slopes for groups (Figure 16, p. 67) varied as a function of degradation types in a manner which resulted in a double transverse interaction. The mean Group 1 slope was higher than the mean Group 2 slope for low-pass filter- ing, lower than the mean Group 2 slope for high-pass filtering, and higher than the mean Group 2 slope for linear rectification. This pat- tern may partially account for the failure to find a significant main effect for groups. Effects found to be significant by the analysis of variance were also followed by a strength of association statistic. Eta squared (n2) was computed to estimate the proportion of variance in slopes that could be accounted for by main effects and the group-by-degradation type inter- action effect. For degradation types, n2 was 0.1498, suggesting that approximately 15% of the variance in slopes could be attributed to de- gradation type. For the group-by-degradation type interaction, n2 was 0.1305, suggesting that approximately 13% of the variance in slopes could be attributed to the interaction. The relatively low percentages of variance in slopes accounted for the significant effects suggests a rela- vtively mild experimental effect in both cases. 98 Summary Visual Magnitude Estimation Data The log-log plots for modulus-free magnitude estimates as a func- tion of circle size were relatively linear and had roughly equivalent slopes. High within-session test-retest correlation coefficients were obtained for the magnitude estimates. The nonsignificant results of a two-way analysis of variance in slopes as a function of groups and trials suggested good within-session reliability of slopes. The results of a one-way analysis of variance showed no statistically significant differ- ences in slopes for the normal hearing experimental group (Group 1), the sensorineural hearing loss group (Group 2), and a separate pilot group of normal hearing subjects. Speech Quality Magnitude Estimation Data Reliability. The dependent variables considered in the experimental questions were log geometric means and slopes. Since log geometric means were obtained for SQMEs across trials, their test-retest reliability was dependent upon the raw data. Since individual 310pes were obtained for the lines which relate log geometric mean SQMEs to log degradation levels, their reliability was dependent upon log geometric mean SQMEs. It should be noted that while individual subject correlations were obtained for SQMEs and log geometric mean SQMEs, within—subject correlations were obtained for slopes. "Average" within-session reliability of individual subject SQMEs for groups (n 8 12) within degradation types was generally high for Group 1 and ranged from moderate to high for Group 2. This same pattern pre- vailed for the subgroups (n = 4) in Session 1 and Session 2. Overall, 99 the "average" individual subject reliability of SQMEs was high for both data sets (n 8 12, n = 4). Between-session reliability of log geometric mean SQMEs for individ- ual subjects was very high for both subgroups (n = 4) within each degrada— tion type. The between-session reliability of slopes showed considerable vari- ability as a function of group-by-degradation type interaction. The most reliable slopes appeared to be the Group 1 slopes for low-pass and high- pass filtering and the Group 2 slopes for low-pass filtering and linear rectification. Trend analyses. The results of six one-way analyses of variance showed statistically significant differences in the log geometric mean SQMEs for each of the two groups as a function of degradation levels for each of the three degradation types. After each analysis of variance, results of tests for linear trend showed that a linear equation provides a statistically significant fit to the data and accounts for a large portion of the variance due to log degradation levels. Thus, a statisti- cally significant linear trend exists for each of the two groups under each of three degradation types. Slope analyses. The results of a two-way analysis of variance in slopes as a function of listener groups and degradation types showed statistically significant differences for degradation types and group- by-degradation type interaction, but not for groups. Differences among means for degradation types were significant for all comparisons. Mean lepes for groups varied as a function of degradation types such that a double transverse interaction occurred. The mean slope for Group 1 was greater than the mean slope for Group 2 for low-pass filtering and lin- ear rectification but not for high-pass filtering. Mean slopes for 100 degradation types varied as a function of groups such that a transverse interaction resulted. Mean slopes for low-pass filtering and linear rectification were higher for Group 1 than for Group 2. The mean slope for low-pass filtering was less than the mean slope for linear rectifica- tion by approximately the same amount for Group 1 and Group 2; so, no interaction was evident when only these two types of degradation were considered. When all three types of degradation were considered, how- ever, high-pass filtering yielded the lowest mean for Group 1 and the highest mean for Group 2, resulting in a transverse interaction. There were no statistically significant differences among the mean slopes for Group 1 - low-pass filtering, Group 1 - high-pass filtering, Group 2 - high-pass filtering, and Group 2 - linear rectification. CHAPTER IV DISCUSSION Introduction Chapter III described the results produced by 12 normal hearing subjects (Group 1) and 12 sensorineurally impaired hearing subjects(Group 2) on a visual training and screening task and in a speech quality mag- nitude estimation (SQME) experiment. 0n the visual task modulus-free magnitude estimates were obtained as a function of circle and square size. In the SQME experiment, modulus-free SQMEs were obtained as a function of seven degrees of three types of signal degradation (low- pass filtering, high-pass filtering, and linear rectification). Two dependent variables were derived from the visual magnitude estimates: (1) log visual magnitude estimates and (2) the slopes of the lines relating log visual magnitude estimates to log circle size. The visual magnitude estimates were analyzed to determine the strength of the relationship between the Trial 2 and Trial 3 estimates. The mean log visual magnitude estimates for Group 1 and Group 2 were plotted as a function of log circle size for comparison with similar data from a pilot study (Appendix M). The slopes of the log-log functions for Groups 1 and 2 were analyzed for significant differences as a function of groups, trials, and group-by-trial interaction. The slopes of the log-log func- tions for Groups 1 and 2 and the pilot group were analyzed for statis- tical significance as a function of groups. Two dependent variables were also derived from the individual SQMEs: (1) log geometric mean SQMEs and (2) slopes for the lines relating log geometric mean SQMEs to log stimulus values (degrees of degradation). 101 102 The log geometric mean SQMEs were examined for statistical significance as a function of degrees of degradation for each listener group and degradation type. When differences were found, these data were further analyzed to identify the lowest order equation required to provide a statistically significant fit to the data. The slopes were analyzed to find statistical significance as a function of listener group, degradation type, and group-by-degradation type interaction. The findings and their implications are discussed below. Visual Magnitude Estimation Task Findings A statistically significant, strong positive correlation (r = 0.94) existed between each subject's Trial 2 and Trial 3 magnitude estimates during the screening task. The slopes relating the log magnitude estimates to log circle size did not differ as a function of groups, trials, or group-by-trial interaction, suggesting good within-session reliability of slopes. The slopes relating log geometric mean magnitude estimates across trials to log circle size did not differ for Group 1, Group 2, or the normal hearing pilot group. The log-log functions for these three groups were similar and were linear in nature. The mean slopes for the three groups were in good agreement with the 0.7 slope reported for sim- ilar data by S. 3. Stevens (1975). Overall, the reliability of the visual data may be characterized as excellent. Implications for the SQME Experiment The strong individual subject correlations suggest that each subject was able to reliably estimate visual magnitude during the visual training and screening tasks. The failure to find within-session differences in 103 slopes as a function of groups, trials, and group-by-trial interaction suggests that: (l) slopes are also replicable within a session and (2) there were no systematic differences in the performance of Group 1 and Group 2 on the visual magnitude estimation task. The failure to find differences in slopes for Group 1, Group 2, and a normal hearing pilot group suggests that slopes are replicable across time as well as groups. This implication is supported by the agreement of the group slopes with results reported by S. S. Stevens (1975). The fact that the log-log functions are linear suggests that the data represent a power function (8. S. Stevens, 1957). Collectively, the visual magnitude estimation data suggest that any systematic differences in performance on magnitude estimation tasks which immediately follow the visual tasks are probably due to factors other than lack of skill in magnitude estimation. SQME Experiment Reliability Within sessions. Pearson product-moment correlation coefficients (r) were used to assess the within-session reliability of the raw data (SQMEs). Pearson r's were determined for each'of the 12 subjects in each experimental group under each degradation type in Session 1 and Session 2. For the normal hearing listeners "average" individual subject cor— relations were consistently high (r = 0.90 to r = 0.93) for the experi- mental group (n = 12) and ranged from high (r = 0.88) to extremely high (r - 0.97) for the subgroup (n = 4). For the hearing impaired listeners "average" individual subject correlations ranged from moderate to high for the experimental group (r = 0.74 to r = 0.95) and for the subgroup (r = 0.78 to r = 0.95). 104 The "average" within-session reliability of SQMEs for individual subjects appears to be good. This reliability is important primarily because the dependent variables (i.e., the log geometric mean SQMEs and slopes) were derived from the SQMEs. The between-session agreement of the log geometric means reflects the reliability of the SQMEs. Between sessions. The between-session agreement of log geometric mean SQMEs was checked for each of the four subjects in the two subgroups that returned to participate in Session 2. "Average" individual subject correlations between log geometric mean SQMEs for Session 1 and Session 2 were positive for all degradation types; the "average" coefficients ranged from high to extremely high for Group 1 (r = 0.94 to r = 0.99) and for Group 2 (r = 0.95 to r = 0.97). Assuming that the individual ratio scales for the SQMEs did not differ drastically from Session 1 to Session 2, it seems reasonable that the between-session correlations of the log geometric mean SQMEs are stronger than the correlations of SQMEs. This is to be expected since the within- session correlations of SQMEs were strong and since the log geometric mean reduces the effects of extreme scores. The between-session correla- tions of log geometric mean SQMEs appear to reflect excellent "average" individual subject agreement over time. The between-session relationship of Session 1 and Session 2 slopes was checked for each subgroup (n = 4) under each of the three types of degradation. For Group 1, the strength of the relationship was essenti- ally nil (r 8 0.09) for linearly rectified stimuli, moderate (r = 0.73) for high-pass filtered stimuli, and very high (r = 0.96) for low-pass filtered stimuli. For Group 2, the relationship was a weak to moderate one (r = -0.52) for high-pass filtered stimuli and a strong positive one for linearly rectified stimuli (r = 0.91) and low-pass filtered stimuli 105 (r = 0.88). "Average" correlations across degradation types were moder- ate for Group 1 (r = 0.76) and Group 2 (r = 0.65). "Average" correla- tions across groups were nil (r = 0.17) for high-pass filtered stimuli, moderate (r = 0.67) for linearly rectified stimuli, and strong (r = 0.93) for low-pass filtered stimuli. Why did between-session correlations show poor test-retest agreement for Group 1 slopes for linearly rectified stimuli and of Group 2 slopes for high-pass filtered stimuli? All subjects were trained in magnitude estimation and demonstrated skill in using this procedure with visual stimuli and the remaining auditory stimuli. The order of stimulus presen~ tations was counterbalanced to reduce order effects. Thus, it seems rea- sonable to speculate that the poor test—retest agreement is due to per— ceptual difficulty during one or both listening sessions. It is likely that Group 2 listeners had perceptual difficulty (re- flected in poor reliability) with high-pass filtered stimuli because they had high frequency hearing losses and were unable to benefit from much of the high frequency portions of the signals presented. 0n the other hand, it is possible that the normal hearing listeners had diffi- culty with linearly rectified stimuli because the amount of information processed and the signal intelligibility were, for them, relatively un- affected by changes in percentage of total harmonic distortion (THD). Log Geometric Mean SQMEs The first experimental question asked whether there was a statistically significant trend for the log geometric mean SQMEs to be influenced by changes in degrees of degradation for each listener group under each degradation type. On the basis of the results obtained, it can be said that such a trend does exist. Statistically significant differences were 106 found in the log geometric means for Group 1 and Group 2 as a function of degree of degradation for each of the three degradation types. The second experimental question was concerned with determining the lowest order equation required to provide a satisfactory fit to the data for each listener group under each degradation type. Examination of test results for linear trend suggested that a first order equation provides a statistically significant fit to and accounts for a large portion of the variance in log geometric means as a function of log degradation levels. Thus, it can be said that a linear trend exists for Group 1 and Group 2 under each type of degradation. In general, the log geometric mean SQMEs are positively correlated with changes in bandwidth due to low-pass and high-pass filter cutoffs and with percent undegraded by linear rectification (% undegraded = 100% - % THD). As a group phenomenon, at least, log geometric mean SQMEs appear to be linearly sensitive to changes in the log electroacoustic characteristics of the signal. These results tend to support those of paired comparison studies (Jeffers, 1960; Zerlin, 1962; Witter and Goldstein, 1971; Smaldino, 1974; and Chial and Daniel, 1977) which found that hearing aids with better electroacoustic characteristics are pre- ferred over those with poorer electroacoustic characteristics. The linear nature of the log-log functions indicates that the data represent a power function (S. 8. Stevens, 1957). S. S. Stevens (1975) suggested that the slopes of such functions represent a dependent varia- ble which may indicate different perceptual events for different stimuli. According to the power law (S. 8. Stevens, 1957), SQMEs should increase as the power or exponent of the signal magnitude (i.e., degree of degrada- tion) increases. In other words, til = K¢fl where d! = subjective magni- tude (SQME), K = a scaling factor equal to the intercept, (b = stimulus 107 magnitude (i.e., degree of degradation), and B represents the slope of a log-log function. Do the observed power functions represent prothetic or metathetic continua of perceptual magnitude? Prothetic continua are generally con- cerned with decisions of quantity, degree, or how much, whereas metathetic continua are generally concerned with decisions of quality, place, what, or where (S. S. Stevens, 1957, 1975). Although power-law behavior is gen- erally exhibited by prothetic continua and not by metathetic continua, the existence of power functions is not independently sufficient to dis- tinguish between the two types of continua (Marks, 1974; S. S. Stevens, 1975; and Gescheider, 1976). In order to draw a distinction between pro- thetic and metathetic continua, S. S. Stevens (1957) suggested four func- tional criteria concerned with (l) subjective size of the just noticeable difference (JND), (2) the form of category rating-scales, (3) the time- order error, and (4) hysteresis effects due to stimulus presentation or- der. In the current study, neither JNDs nor category judgments were ob- tained, and hysteresis effects were minimized by randomizing degrees of degradation. Thus, the power functions obtained in the present study cannot be readily characterized in terms of these criteria. According to S. S. Stevens (1957, 1975), a time-order error exists on prothetic con- tinua such that a comparison stimulus (second) is judged to be greater than the standard stimulus (first) when the two are equal. The time- order error does not exist on metathetic continua. Table 30 shows the per-' centages of times the speech quality magnitude of the comparison stimulus equal to the standard stimulus was judged to be less than, equal to, and greater than that of the standard stimulus for each group under each degradation type. With the exception of Group 2 judgments under linear rectification, the quality of the comparison stimulus was judged to be Table 30. 108 Percentages of times the speech quality magnitude of the com— parison stimulus equal to the standard stimulus was judged less than, equal to, or greater than that of the standard stimulus. Percentages are based on the magnitudes of the 24 SQMEs (2 passages judged by 12 subjects) produced by Groups 1 and 2 for the standard degradation level under each degradation type. Perceived Quality of the Comparison.Stimulus Relative to that of an Degradation Types Equal Standard Stimulus L-PF H-PF LR Less 37.5% 37.5% 12.5% Group 1 Equal 45.8% 41.7% 62.5% Greater 16.7% 20.8% 25.05% Less 16.7% 33.3% 16.7% Group 2 Equal 45.8% 41.7% 41.7% Greater 37.5% 25.0% 41.7% 109 equal to that of the standard stimulus more often than it was judged to be less or greater. For Group 2 judgments under linear rectification, the quality of the comparison stimulus equal to that of the standard was judged to be equal to and greater than the quality of the standard an equal number of times. Only in the case of Group 1 under linear rectification, however, did the judged-equal category represent 50 per- cent or more of the judgments. Thus, the time-order error appears to be of little help in classifying the continua for SQMEs. Slopes Mathematically, the slope can be defined as change in the variable on the Y axis divided by the change in the variable on the X axis. When log geometric mean SQMEs are plotted as a function of log degree of de- gradation, the slope represents increase in perceptual values relative to increase in degree of degradation. If the value of the slope is one, change in log geometric mean SQMEs and change in log degree of degradation are equal and the function creates a 45° angle with the X axis. If the value of the slope is greater than one, change in log geometric mean SQMEs is greater than change in log degree of degradation, and the func- tion creates an angle greater than 45° with the X axis. If the value of the slope is less than one, change in log geometric mean SQMEs is less than change in log degree of degradation, and the function creates an angle of less than 45° with the X axis. The third experimental question related to the existence of statis- tically significant differences among the $10pes of the log-log functions as a function of listener group, degradation type, and the interaction of listener group and degradation type. The results of the study showed differences in slopes as a function of degradation types and group-by- degradation type interaction, but not as a function of groups. The means, 110 standard deviations, and ranges of the slopes are shown in Table 9 (p. 66) as a function of groups and degradation types. Effects gf_degradation type for Group_l (normal). Consider the mean slopes for Group 1 separately for each degradation type. The change in log geometric mean SQMEs relative to the change in log degree of degrada- tion was essentially the same for low-pass and high-pass filtered stimuli, but was significantly greater for linearly rectified stimuli. Effects gf_degradation type for Group 2_(impaired). Now consider the mean slopes for Group 2 separately for each degradation type. The change in log geometric mean SQMEs relative to the change in log degree of degradation was essentially the same for high-pass filtered and linearly rectified stimuli, but was significantly smaller for low-pass filtered stimuli. In other words, Group 2 required a relatively large change in degree of degradation by low-pass filtering to produce a given change in SQMEs. Effects 9f_degradation type for Groups 1 and 2, Inspection of all the mean SIOpes reveals that only the Group 1 mean for linearly rectified stimuli was greater than one. For Group 1, change in log geometric mean SQMEs was greater than change in log percent undegraded by linear recti- fication. As noted earlier, the between-session correlation of slopes (n = 4) for linearly rectified stimuli was nil (r = 0.09) for Group 1 and strong for Group 2. In general, change in log geometric mean SQMEs was less than change in log degree of degradation. Further inspection of the mean slopes shows that only in the case of high-pass filtered stimuli was the Group 2 mean greater than the Group 1 mean. Recall that the difference in the means was not statistically significant and that the between-session correlation of Group 2 slopes (n = 4) for high—pass filtered stimuli was a weak-to-moderate one (r = -0.52). 111 For low-pass filtered and linearly rectified stimuli the change in log geometric mean SQMEs relative to the change in log degree of degradation was significantly greater for Group 1 than for Group 2. In these two instances, Group 2 required more degradation to produce a given change in SQMEs than did Group 1. In the case of high-pass filtered stimuli, Group 1 required more degradation to produce a given change in SQMEs than did Group 2. One interpretation of these findings is that changes in low frequency energy have relatively smaller effects on speech quality magni- tude estimates and therefore that most listeners, particularly hearing impaired listeners, should perceive low-pass filtered signals as being of higher quality than high-pass filtered signals. This interpretation appears to be supported by recent observations (Punch and Ciechanowski, 1977; Punch and Beck, 1979; Punch and Parker, 1979; and Swartz, Walden, and Prosek, 1979) that low frequency signals are, in fact, preferred. Punch and Ciechanowski (1977) also noted better preference reliability with greater relatively low-frequency energy in the signal. Quality judgments were felt to be based mostly on low-frequency information, at least in subjects with good hearing for low frequencies. Theoretical Factors Perceived speech quality can be intuitively related to a variety of factors, including the amount of information transmitted and received, signal intelligibility, and a "naturalness" factor. Information theory. Information theory (Shannon and Weaver, 1949, 1963) states that I (amount of information) = 2 t w log (8 + N/N) where t is signal duration, w is width of the useable frequency range, 8 is the maximum amplitude of the signal, and N is minimum discernible 112 intensity difference. According to Lassman's (1964) "noise interference" model, S/N environment I: Nhearing aid + Nperipheral + Ncentral auditory auditory system system where I is intelligibility, S is signal magnitude, and N is noise in the information theory sense (i.e., anything that increases signal ambiguity). Thus, intelligibility appears to be inherently related to information trans— mission. Lassman (1964) was undecided about what to call his model and noted that the title almost calls for something like "a model of infor- mation flow". In any event, the most currently relevant variables in these two formulas appear to be frequency bandwidth and "noise". For the sake of argument, assume that information theory (Shannon and Weaver, 1949, 1963) and Lassman's model are applicable to quality judg- ments and that speech quality increases when the amount of information transmitted and received increases. Signal presentation levels for the impaired listeners ranged from 58 to 90 dB hearing level, while the thres- holds in the test ear ranged from 20 to 100 dB hearing level at 2000 Hz and from 35 dB to hearing levels beyond the intensity limits of the audio- meter at 4000 Hz. Thus, the normal hearing listeners received high fre- quency information at higher sensation levels relative to the thresholds for high frequencies than did the impaired listeners. Although the same signal bandwidths were presented to all listeners under the various filtering conditions, the "useable" bandwidths were probably smaller for hearing impaired listeners than for normal hearing listeners due to dif- ferences in auditory systems. Although the same "noise" levels were pre- sented to all listeners under the various filtering and linear rectification conditions, these "noise" levels may have interacted with larger "noise" 113 levels in the impaired auditory systems than in the normal auditory sys- tems. Different useable bandwidths and "noise" levels in the auditory system could affect the growth of perceived speech quality magnitude differently, depending upon the type of signal degradation. As low-pass filtered signals increased from minimum (550 Hz) to max— imum (4350 Hz) bandwidth, the mean geometric mean SQMEs underwent a 4.7 fold increase for Group 1 and a 1.9 fold increase for Group 2. The finding of a significant difference in the slopes obtained for Groups 1 and 2 suggests that Group 2 experienced an abnormally slow growth in perceived speech quality as a function of increasing bandwidth. As high-pass filtered signals increased from minimum (900 Hz) to max- imum (4350 Hz) bandwidth, the mean geometric mean SQMEs underwent a 2.7 fold increase for Group 1 and a 3.9 fold increase for Group 2. In this instance, it appears that Group 2 experienced an abnormally rapid growth in perceived speech quality magnitude. Recall, however, that there was no significant difference in the slopes obtained for Groups 1 and 2. Mean geometric mean SQMEs for linearly rectified stimuli underwent a 2.8 fold increase for Group 1 and a 1.9 fold increase for Group 2, as percentages of total harmonic distortion decreased from 60% (40% unde- graded) to 1% (99% undegraded). Since the observed slopes for Groups 1 and 2 were significantly different, Group 2 apparently experienced an abnormally slow growth in perceived speech quality magnitude as a function of decreasing total harmonic distortion. In general, it can be argued that more information is received and SQMEs increase as useable bandwidth increases and "noise" level decreases. The growth rate of perceived speech quality magnitude and information re- ceived, however, appears to vary as a function of group-by-degradation type interaction. The exact relationship between SQMEs and information 114 transmission is unknown. Intelligibility theory. Perceived speech quality has been related to the intelligibility of the signal as demonstrated on discrimination tests (McGee, 1965). The seven cutoff frequencies for the low-pass and high-pass filtered signals were interpolated from French and Steinberg's (1947) articulation scores for filtered CVC monosyllables. The seven cutoff frequencies were estimated from percent correct word discrimination scores ranging from 10 to 100% in 15% increments. As would be expected, SQME values in- creased as frequency bandwidth increased. As noted earlier, the growth of perceived speech quality magnitude as a function of low frequency bandwidth was significantly slower for Group 2 than for Group 1. This is reasonable, since the rate of change in Group 2's discrimination ability may also be slower. The growth of perceived speech quality magnitude with increases in high frequency bandwidth did not differ for Groups 1 and 2 in spite of the expectation that rate of change in discrimination ability would differ for the two groups. Licklider (1946) noted that 50% word discrimination is roughly equiva- lent to 90% sentence discrimination. Interpolations from French and Steinberg's (1947) syllable articulation data suggest scores of about 55 to 100% for four of the seven degrees of degradation used for the low-pass and high-pass filtered signals in the current study. Thus, it is probably safe to assume that four of the seven degradation levels for the lowbpass and high-pass filtered signals were at least 90% intelligible for normal hearing listeners. Intelligibility theory alone can account for relative- ly little of the change in quality related to filtering. Percent undegraded by linear rectification varied from 40% to greater than 99% in 10% increments. Perceived speech quality magnitude increased 115 as percent undegraded increased. The rate of increase was slower for Group 2 than for Group 1. Licklider and Held (1952) reported that their normal hearing subjects attained a 98% discrimination score on words sub- jected to half-wave rectification (40% THD or 60% undegraded). Since four of the seven degradation levels represent less than 40% THD, it is highly unlikely that Group 1's SQMEs were based entirely on discrimination ability. Group 2's SQMEs could be more closely related to discrimination ability since their discrimination ability should be poorer and show less improve- ment with changes in degree of degradation. In total, however, intelli- gibility theory appears to account for relatively little of the change in quality related to linear rectification. Implications for Future Research Speech quality magnitude estimation appears to be a manageable task for most people after minimal training. The visual training and screening task employed here appears to have served its purposes well. Since the visual data were reliable across time and groups, systematic differences in performance on the SQME tasks can be attributed to perceptual differ- ences rather than to lack of skill in magnitude estimation. Similar train- ing and screening procedures can be recommended for future studies involv- ing magnitude estimation tasks. The current study shows that log geometric mean SQMEs are positively and linearly related to log degree of degradation by lowbpass filtering, high-pass filtering, and linear rectification. The study also shows that change in log geometric mean SQMEs relative to change in log degree of degradation (i.e., slope) varies as a function of degradation type and group-by-degradation type interaction. These findings have implications for additional research on the evaluation of communications systems in 116 engineering, research on clinical practices in audiology, and further basic research to investigate how normal hearing and hearing impaired indivi- duals perceive complex signals. When perceptual phenomena exhibit power- law behavior, some degree of prediction becomes possible. Perception p£_Comp1ex Signals Classification pf_SQME continua. A ubiquitous issue in psychophysics is whether perceptual continua are of the prothetic or metathetic class. Since this issue was not resolved in the current study, a future study could be specifically designed to include functional criteria described by S. S. Stevens (1957) for distinguishing between classes of continua. Matching perceptual experiences. An interesting question in audiology is how to determine degrees of degradation which will allow a normal hearing listener to have a perceptual experience similar to that of a hearing impaired listener. Cross—modality matching procedures (S. S. Stevens, 1975) may be helpful in answering this question. Assume that each member of a normal hearing listener group and a hearing—impaired listener group performs magnitude estimates by assigning numbers to vibrations on the finger and to the quality of speech samples varying in degree of degrada- tion. Further assume that each of the same subjects adjusts a vibration on his finger until it matches the perceived quality of the same degraded speech samples. If the log perceptual values in each case are examined as a function of the log stimulus values, each of the functions should be linear, and the SIOpes of the matching function for each group should equal the ratio of the slopes for the two original functions for each group (S. S. Stevens, 1975). If the predicted slopes are verified in the matching functions, common vibration amplitudes for the two listener groups may represent speech samples that are perceived as being of equal 117 quality. If this proves to be the case, application of cross-modality matching procedures in this manner may enable the normal hearing listen- er to relate more realistically to the perception of an individual who is hearing impaired. Another interesting question relates to how much degradation of one type produces quality that is equivalent to that produced by a given de- gradation level of another type. Assume that a group of listeners assigns numerical magnitudes to perceived vibrations on the finger and to the perceived quality of speech samples degraded to varying degrees by low- pass filtering and high—pass filtering. Each subject would then adjust the vibration magnitude until it matches the perceived quality of the de- graded speech samples. Log perceptual values should grow linearly as a function of log degree of degradation, and the slape of the matching func- tion for each degradation type should equal the ratio of the slopes for the original vibration function and perceived speech quality function for the degradation type in question (S. S. Stevens, 1975). If the predicted slopes are observed in the matching functions, common vibration amplitudes for the two degradation types may represent speech samples that are per- ceived as being of equal quality. Another application of the cross-modality matching procedure might enable one to equate quality perceived in one domain with that perceived in another. Assume that a group of subjects assigns numerical magnitude estimates to: (l) the quality of a visually projected image subjected to varying degrees of distortion, (2) the quality of speech samples de- graded to varying degrees, and (3) a series of vibrations on the finger. Further assume that each subject adjusts vibration magnitude on the finger until it matches the perceived quality of the distorted visual images and the degraded speech samples. If the power law holds for each function, 1 118 the slopes of the two matching functions should equal the ratios of the vibration function and the respective original quality functions (S. S. Stevens, 1975). If this happens, equal magnitudes on the two matching functions may represent equivalent quality perception in the visual and auditory modalities. Determinants pf speech quality. Perception of speech quality appears to be related to a host of interacting factors. Future studies should consider various forms of multivariate analyses to discover the deter- minants of quality perception and their relative importance for normal hearing and hearing impaired listeners. Extensions p£_the current study. Studies similar to the current study should determine the effect of more closely defined type and de- gree of hearing loss and more severe hearing loss on quality perceptions. Such studies should be extended to include different degradation types, combinations of degradation types, and more degrees of degradation. Prediction pf_Perceptual Experience Scales developed from magnitude estimation and production of speech quality can be used in much the same way that the phon (Fletcher and Munson, 1933) and sone (S. S. Stevens, 1936) scales are used. For ex- ample, consider in the place of phone, SQUALs (speech quality levels). A SQUAL could be equated with a given dB level of a standard comparison signal which would be degraded to some specified degree in some specified manner (e.g., by adjusting cutoff frequencies). Equal quality contours could be developed for various dB levels of the standard comparison * signal. It is likely that speech quality will vary with presentation level as well as type and degree of degradation. Now, in place of sones, con- sider SQUALUs (speech quality units). One SQUALU would be equal to some 119 speech quality level (SQUAL). A speech sample of n SQUALUs would be perceived as having n times the quality of one SQUALU. 0n the basis of SQUALs and SQUALUs, it is likely that predicted speech quality for various degraded speech samples could be predicted for normal hearing listeners and that a given number of SQUALUs perceived by normal hearing listeners will be equivalent to the quality perceived by a hearing impaired listener. In this latter instance, it is likely that the number of SQUALUs indicated would suggest a reduced level of perceived quality. Evaluation p£_Communication Systems Rothauser, Urbanek, and Pachl (1968) noted that evaluation and optimi- zation criteria are needed during the design, development, and testing of speech handling and processing systems. Rothauser, Urbanek, and Pachl (1971) defined a preference unit (PU) scale based on the "Transmission Preference Units" (TPU) introduced by Munson and Karlin (1962). Rothauser pp a1. (1971) related data to the PU scale from four preference evaluation methods (the isopreference method, the category judgment method, the rela- tive preference method, and an absolute preference judgment method). They found it impossible to recommend any of the four as a single best method for all situations. Thus, speech quality magnitude estimation and pro- duction procedures may prove to be helpful in determining evaluation and optimization criteria for engineering purposes. Such an application would probably require the development of some sort of quality scale (e.g., the SQUAL and SQUALU scales). Clinical Application Description. Log SQME—log degree of degradation functions may be useful in describing clinical problems. The slopes of such functions 120 have been shown to vary as a function of degradation type and the inter- action of degradation type with hearing loss, suggesting the occurrence of different perceptual events. As more is learned about the processing of complex signals, measures like the SQUALs and SQUALUs described above may become useful in describing perceptual difficulty. Diagnosis. Audiological diagnosis typically provides information which contributes to determining site of lesion in the auditory system. The most difficult diagnostic problems are probably those which require differentiation among the cochlea, the eighth nerve, and the central audi- tory nervous system as the site of lesion. Direct magnitude scaling pro- cedures have been used on experimental bases to identify recruitment due to cochlear pathology. Using a cross-modality matching procedure to study the loudness growth of pure tones in patients with unilateral conductive hearing loss and patients with unilateral sensorineural hearing loss, Thalmann (1965) found that loudness balances of the right ear against the left ear were predicted by vibration matches on the fingers. S. S. Stevens (1966) indicated that abnormal loudness growth can best be described by two straight lines in a log-log plot. He postulatedtflun:whenever the ex- ponent (slope) of a sensory function is altered by pathology or some other circumstance, a power transformation occurs. In sensorineural hearing loss due to cochlear pathology, the power transformation is related to recruit- ment. In a similar vein, research on direct magnitude scaling of speech quality by sensorineurally impaired listeners may lead to even finer dis- crimination of hearing loss on the basis of slopes or power transforma- tions revealed in a log-log plot. It is generally accepted that speech signals are very helpful as test materials for distinguishing among dis- orderscnfthe cochlea, eighth nerve, and central auditory nervous system (Katz, 1962; Bocca and Calearo, 1963; and Jerger and Jerger, 1971, 1975). 121 Audiological diagnosis typically provides information about type and degree of hearing loss or deficit, but often provides very little infor- mation about hearing handicap (Oyer and Frankmann, 1975). Ixzmay be that communicative effectiveness in everyday life is closely related to the magnitude of the overall speech quality perceived by a listener. If a 1 suitably strong relation exists, clinical measures of quality perception may facilitate assessment of oral-aural communicative integrity. Prognosis and progress. The slopes of the log-log functions in the present study differed as a function of speech degradation type and listen- er group-by-degradation type interaction. If perceived speech quality magnitude can be improved by therapy, it may be that speech quality mag— nitude functions obtained at different points in time can serve as measures of progress in aural rehabilitation. Certain slope values might become valuable as predictors of success in improving perceptual skills. Where physical stimulus characteristics are determined by different hearing aids, the relative speech quality magnitudes assigned to Signals transduced by different aids may be helpful in hearing aid selection-Pro- cedures. Previous investigators have indicated that quality judgments are related to the electroacoustic characteristics of hearing aids 97h; (Jeffers, 1960; Zerlin, 1962; Witter and Goldstein, 1971; Smaldino’ it? 1. ‘13 Punch and Ciechanowski, 1977; and Chial and Daniel, 1977) and that 4 ea 0 at JUdgments may be more sensitive to electroacoustic differences thafl ‘11537’ 39% traditionally used word discrimination tests (Jeffers, 1960 and 7.8 09 1962). Quality judgments may represent a good predictor of succes9 amplif icat ion . CHAPTER V SUMMARY AND CONCLUSIONS Introduction Background Traditional word discrimination tests do not adequately predict listener performance in everyday life and therefore do not measure hand- icap (Chial and Hayes, 1974; Oyer and Frankmann, 1975; Millin, 1975; and Berger, 1978). Speech quality judgments have sometimes differenti- ated among hearing aids when word discrimination tests did not (Jeffers, 1960; Zerlin, 1962). Thus, judgments of speech quality may be more closely related to everyday communicative effectiveness than are word discrimina- tion scores. Although there appears to be some interest in the use of quality judgments in clinical audiology, there has been insufficient re- search with hearing impaired subjects to warrant their routine use. Furthermore, most of the available research has been limited to the method of paired comparisons. Although direct magnitude estimation procedures appear to offer some advantages over paired comparisons, no one has studied this method to determine whether Stevens' power law applies to the scaling of speech quality and whether there are differences in speech quality magnitude estimate-degree of degradation functions as a function of hearing status and signal degradation type. Purpose This study was designed to obtain speech quality magnitude estimates (SQMEs) from normal hearing and sensorineurally impaired hearing listeners as a function of seven degrees of degradation by low-pass filtering, 122 123 high-pass filtering, and linear rectification. The log geometric mean SQME-log degree of degradation functions were analyzed: 1. to determine whether log geometric mean SQMEs differed as a function of changes in log degree of degradation for each listener group under each degradation type; 2. to identify the lowest order equation required to provide a satisfactory fit to the log-log functions; and 3. to determine whether differences existed among the slopes of the log- log functions as a function of listener group, degradation type, or group- by-degradation type interaction. Experimental Design Subjects There were two groups of subjects. Group 1 consisted of 12 normal hearing listeners with a mean age of 23.58 years, a mean two-frequency average hearing threshold level of 0.42 dB in the test ear, and a mean speech discrimination score of 99.5 percent in the test ear. Group 2 consisted of 12 sensorineurally impaired hearing listeners with a mean age of 32.58 years, a mean two-frequency average hearing threshold of 33.50 dB in the test ear, and a mean speech discrimination score of 80.17 percent in the test ear. For the most part, the Group 2 hearing losses were most pronounced for high frequencies. Their hearing test results were con- sistent with a cochlear site of lesion. Stimuli Visual training and screenipg stimuli. The visual stimuli consisted of three sets of seven pairs of geometric forms located side by side.on 2" x 2" slides. The left member of each pair was always the standard 124 stimulus, whose size did not change; the right member of each pair was always one of the seven comparison stimuli which differed in size. The first set of seven pairs consisted of squares, while the last two sets of seven pairs consisted of circles. Each set represented a trial. Auditory stimuli. The auditory stimuli were audio recordings of six lO—second connected speech samples from a junior high history text, which were, in their undegraded form, approximately equal in listening diffi- culty. The final test tapes consisted of degraded versions of these sam- ples recorded in pairs. Each pair consisted of a standard stimulus fol- lowed by one of seven comparison stimuli. Each comparison stimulus repre- sented one of seven degradation levels, while the standard stimulus always represented the middle (fourth) of the seven degrees of degradation. The types of degradation were low-pass filtering, high-pass filtering, and linear rectification. Degrees of degradation were represented by: (l) bandwidths of 550, 950, 1300, 1650, 1950, 2950, and 4350 Hz for low-pass filtering; (2) bandwidths of 900, 1400, 1600, 2100, 2550, 3000, and 4350 Hz for high-pass filtering; and (3) percent undegraded values of 40%, 50%, 60%, 70%, 80%, 90%, and 99% for total harmonic distortion due to linear rectification. Procedures All subjects underwent: (1) a hearing screening, (2) visual magnitude estimation training and screening, (3) SQME training, and (4) the SQME experiment in that order. Four subjects in each listener group returned to repeat the SQME training and the SQME experiment at a later date. Hearing_screenipg. The hearing screening consisted of a brief history, pure tone air conduction testing, reflex decay or tone decay testing, word discrimination testing in the test ear, and impedance testing. 125 Visual magnitude estimation trainipg and screening. Each subject was presented with three sets of randomly ordered pairs of visual stimuli, which were accompanied by audio taped instructions. The first set con- sisted of squares which served as training stimuli; the second and third sets consisted of circles which provided a test—retest reliability check. Each subject assigned a numerical value to the magnitude of the standard stimulus in each set and then estimated the magnitudes of the comparison stimuli relative to that of the standard. The visual task served as a pro- cedural model for the SQME tasks. SQME training. Twenty—one tape recorded practice items (seven degrees of each of the three degradation types) preceded the SQME experiment. The same two passages, one for the standard stimulus and one for the comparison stimulus, were used for all practice items. All stimuli of a single degrad- ation type were presented in succession, with degradation levels being presented in random order. The order of presentation for degradation types was counterbalanced across pairs of listeners. Listeners assigned SQMEs to the auditory standard and comparison stimuli in the same manner used to make visual magnitude estimates in the visual tasks. SQME experiment. SQMEs were obtained on 42 experimental items (two seven-item trials of each of the three degradation types) following the same procedure used in the visual and SQME training tasks. In the experi- mental tasks, however, there were four, rather than two, stimulus passages. Two served as standard stimuli, and two served as comparison stimuli. The same two passages were always presented together, so that there were, in effect, only two different pairings of the four stimuli. The order of these pairs of passages was counterbalanced across trials. Just as in the SQME training task, the order of degradation levels was randomized *‘within each trial, and the order of degradation types was counterbalanced 126 across pairs of listeners. Dependent Variables Two dependent variables were derived from the visual magnitude estimates and the SQMEs. The visual magnitude estimates provided the basis for (1) log visual magnitude estimates and (2) the lepes of the lines of best fit which relate log visual magnitude estimates to log circle size. The SQMEs provided the basis for (1) log geometric mean SQMEs within subjects and across trials and (2) the slopes of the lines of best fit relating log geometric mean SQMEs to log degree of degrada- tion. Findings Findings of the study include the following. 1. Log visual magnitude estimates appear, as indicated by S. S. Stevens (1975), to be linearly related to log circle area and therefore, to represent a power function. 2. The mean slopes for the log visual magnitude - log area func- tions produced by a group of normal hearing listeners and a group of sensorineurally impaired hearing listeners are in agreement with the slope reported by S. S. Stevens (1975) for similar data. 3. Log geometric mean SQMEs (across trials) are positively and linearly related to log bandwidth of low-pass and high—pass filtered stimuli and to log percent undegraded by linear rectification, indicating that a power function applies in each case. 4. Slopes of the log-log functions for Group 1 differed from those for Group 2 under low—pass filtering and linear rectification, but not under high-pass filtering. 127 5. For Group 1 the slopes of the log-log functions for low-pass and high—pass filtered stimuli did not differ from each other, but did differ from the slopes of the log-log functions for linearly rectified stimuli. 6. For Group 2 the slopes of the log-log functions for high-pass filtered and linearly rectified stimuli did not differ from each other, but did differ from the slopes of the log-log functions for low-pass filtered stimuli. Conclusions The results of this study seem to provide the basis for the follow- ing tentative conclusions. 1. Excellent reliability of listener performance on the visual task within subjects, across groups, and across time suggests that systematic differences in performance on the SQME tasks are probably due to percep- tual differences. 2. Tasks similar to the visual training and screening task used in this study appear to be practical for use in future studies involving the method of magnitude estimation. 3. The slopes of the log geometric mean SQME - log degree of degrad- ation functions differ as a function of group-by-degradation type inter- action, suggesting systematic perceptual differences. 4. Estimates of poor between-session reliability of slopes suggested that Group 1 may have had perceptual difficulty with linearly rectified stimuli, while Group 2 may have had perceptual difficulty with high-pass filtered stimuli. 5. Factors other than discrimination ability and information trans- mission and reception appear to play a major role in the perception of speech quality. 128 6. The findings of this study were sufficiently encouraging to war- rant additional research on the evaluation of communications systems, clinical practices in audiology, and how normal hearing and hearing im- paired individuals process complex signals. APPENDICES APPENDIX A TABULAR SUMMARY OF AGE AND AUDIOMETRIC DATA FOR INDIVIDUAL SUBJECTS AND GROUPS The two tables which follow summarize the age and audiometric data for individual subjects and groups. Table A-1 summarizes the ages, ,two-frequency average hearing threshold levels, test ear discrimination scores, and the means and standard deviations for all normal hearing sub- jects (Group 1) and all sensorineurally impaired hearing subjects (Group 2). Table A-2 lists the pure tone thresholds and median thresholds as a function of ear and test frequency for the hearing impaired subjects. 129 130 Table A-1. A summary of ages, two-frequency average thresholds, test ear discrimination scores, and the means and standard deviations for all subjects. 2-Frequency % Age Average Threshold Discrimination Subject Yrs. R dB L dB Score Group 1 (Normal) - 1 22 0a 0 100 2 24 O 0 100 3 23 0 0 100 4 24 0 0 100 5 23 0 0 100 6 22 0 0 98 7 27 O 0 98 8 28 5 5 100 9 22 O O 100 10 22 O 0 98 11 23 O 0 100 12 23 0 0 100 F 23.58 0.42 0.42 99.5 S.D 1.98 1.44 1.44 0.9 Group 2. (Impaired) b 13 36 48 45b 90 14 49 25 20b 84 15 21 33b 28 90 16 35 35 37b 70 17 48 18 18b 70 18 20 48 40b 88 19 19 60b 50 60 20 18 35 45b 90 21 46 35b 25 74 22 37 48 50b 90 23 30 20b 15 88 24 32 43 73 68 3? 32.58 37.33 37.17 80.17 S.D. 11.37 12.61 16.87 10.97 3The right ear was the test ear for all subjects in Group 1. bTest ear thresholds for Group 2: 'i'= 33.50, S.D. = 12.13. -.. .uouoEoHvam ofiu mo muHEHH huHmcoucH um uncommou oz "mzm 131 o.mo m.Nm m.Nm m.Ne m.NN o.oN guest: you away o.me m.Ne o.oo m.No o.mm o.mm m.Ne m.Ne m.NN m.NN o.oN m.NN guano: oN ma so as mm mm on mN me on mm o «N om om mm oa ON me ma mN ma ma oN ma NN as ea on mm on me mm mm om om me oe NN mm mm mm mm mm mm mm me ma oN ma oN aN mm mm co co mm me om os as on on «N oN mmz mmz mmz mmz mm moa mN «N on me on on as me no oN no on mm me mm mm oe mm mm we oN ms mN om om no as ON n ma o m Na mmz maz mmz mmz mmz cos om on ma ma as on 02 no so me on me as ea oe ma mN mN mN ma mmz mm mN mN mm mm mN mN ma mN oN om ea no om an mm mm mo mm ow mm mm oN oN Na 4 m a m a m a a a m a m uuufiaam um 88 um 83 um 88 N: 83 u: com um RN .muoofinam wouHmaEH mcHumo: onu now Aumv hocmsv Iouw can AH .mv new mo =0Huocsw m mm mvHonmmusu :oHvoa paw Amvv mvHonmounu ocOu ousm .NI< oHan APPENDIX B TRANSCRIPT 0F SIX STIMULUS PASSAGES Practice Standard B Balboa named his discovery the South Sea because it lay directly south of where he started his march. It was not until after Magellan's voyage that the sea was called Pacific, the name we use today (Wilder g£_al,, 1954, p. 62). Practice Comparison B After several years spent in preparation, Pizarro set off on his great adventure. He landed safely on the coast of Peru, where he re- mained for some time 'sizing up' the situation (Wilder et_§1,, 1954, p. 66). Experimental Standard A Two months later the weary Spaniards stood looking in amazement upon the Aztec capital. The city was built on islands in the center of a large lake, and was connected with the mainland by three roads or causeways (Wilder et 31,, 1954, p. 64). Experimental Comparison A One of the early settlers, named John Rolfe, learned how to produce fine tobacco. Smoking was becoming popular in England, so the Jamestown colonists found it easy to sell all the tobacco that could be grown (Wilder e£_§1,, 1954, p. 87). Experimental Standard 9 The bold explorers who searched this land did not find the waterway they were seeking, but they accomplished something more important. They turned the attention of Europe away from Asia to the New World itself (Wilder gt 31., 1954, p. 53). Experimental Comparison 9 Every kind of disaster happened to the expedition - storms, sickness, death, mutiny, desertion. But at last the men who remained alive anchor- ed once more in a Spanish harbor (Wilder gt_gl,, 1954, p. 43). Source: Wilder, H. B., Ludhum, R. P. and Brown, H. M. This is America's Story. Boston: Houghton, Mifflin Company (1954). 132 APPENDIX C TABULAR DESCRIPTION OF THE SUBMASTER RECORDINGS, RECORDINGS PRODUCED BY THE COMPUTER SYSTEM, AND THE FINAL TEST TAPES The three tables which follow describe the makeup of the three sets of recordings generated from the master tape. Table C-l summarizes the nine submaster recordings. Table C-2 describes the interim recordings produced from the submaster recordings by the computer system. Table C—3 describes the final test tapes which were spliced from sections of the interim tapes produced by the computer system. 133 134 Table C-l. Summary of the nine submaster recordings. Submaster Degradation Degradation Cassette No. Contents Type Level 1 Comparison L-PF l Passages H—PF l A,B,C LR l 2 Comparison L-PF 2 Passages H-PF 2 A,B,C LR 2 3 Comparison L-PF 3 Passages H—PF 3 A,B,C LR 3 4 Comparison L-PF 4 Passages H-PF 4 A,B,C LR 4 5 Comparison L-PF 5 Passages H-PF 5 A,B,C LR 5 6 Comparison L-PF 6 Passages H-PF 6 A,B,C LR 6 7 Comparison L—PF 7 Passages H—PF 7 A,B,C LR 7 8 Standard L-PF 4 Passages H-PF 4 A,B,C LR 4 9 Labels None None (i.e., Trial 1, standard, item, one, etc.) 135 Table C-2. Crossbreak matrix showing the makeup of the recordings pro- duced by the computer system. Interim Three Pairing of Six Number of Tape Degradation Stimulus Random No. Types Passages Orders 1 L-PF PSB & PCBl-PCB7 6 2 H-PF PSB & PCBl-PCB7 6 3 LR PSB & PCBl-PCB7 6 4 L—PF ESA & ECAl-ECA7 6 5 L-PF ESC & ECCl-ECC7 6 6 H-PF ESA & ECAl-ECA7 6 7 H-PF ESC & ECCl-ECC7 6 8 LR ESA & ECAl-ECA7 6 9 LR ESC & ECCl-ECC7 6 aL-PF: low-pass filtering H-PF: high-pass filtering LR : linear rectification bPSB: practice standard passage B PCBl-PCB7: 7 degradation levels of practice comparison passage B ESA: experimental standard passage A ESC: experimental standard passage C ECAl-ECA7: 7 degradation levels of experimental comparison passage A ECCl-ECC7: 7 degradation levels of experimental comparison passage C 136 Table C-3. Crossbreak matrix: Presentation orders for degradation types, passages, and random orders for comparison degradation levels. Three Order of Seven Degradation Pairings of Six Comparison Subjects Trial Typesa Stimulus Passages Degradation Levels 1 8 2 1 LR PSB 8 PCBl-PCB7 2135764 (13 8 l4) 2 L-PF PSB 8 PCBl-PCB7 3246517 3 H—PF PSB 8 PCBl-PCB7 4126375 1 L-PF ESA 8 ECAl-ECA7 7631524 2 L-PF ESC 8 ECCl-ECC7 6274513 1 H-PF ESA 8 ECAl-ECA7 5374621 2 H-PF ESC 8 ECCl-ECC7 1426375 1 LR ESA 8 ECAl-ECA7 7613542 2 LR ESC 8 ECCl-ECC7 1743652 3 8 4 1 LR PSB 8 PCBl-PCB7 2653741 (15 8 l6) 2 H-PF PSB 8 PCBl-PCB7 6543721 3 L-PF PSB 8 PCBl-PCB7 2351764 1 L-PF ESC 8 ECCl-ECC7 5764231 2 L—PF BSA 8 ECAl-ECA7 4352761 1 LR ESC 8 ECCl-ECC7 4267153 2 LR ESA 8 ECAl-ECA7 7631524 1 H-PF ESC 8 ECCl-ECC7 7325416 2 H-PF ESA 8 ECAl-ECA7 7651342 5 8 6 l H-PF PSB 8 PCBl-PCB7 2453671 (17 8 l8) 2 LR PSB 8 PCBl-PCB7 2574316 3 L-PF PSB 8 PCBl-PCB7 1463275 1 H-PF ESA 8 ECAl-ECA7 2475163 2 H-PF ESC 8 ECCl-ECC7 4657123 1 L-PF ESA 8 ECAl-ECA7 3251647 2 L-PF ESC 8 ECCl-ECC7 1426375 1 LR ESA 8 ECAl-ECA7 2143657 2 LR ESC 8 ECCl-ECC7 4723156 7 8 8 l H-PF PSB 8 PCBl-PCB7 3726154 (19 8 20) 2 L-PF PSB 8 PCBl-PCB7 6245137 3 LR PSB 8 PCBl-PCB7 4316572 1 H-PF ESC 8 ECCl-ECC7 3527641 2 H-PF ESA 8 ECAl-ECA7 1547236 1 LR ESC 8 ECCl-ECC7 6512374 2 LR ESA 8 ECAl-ECA7 3152647 1 L-PF ESC 8 ECCl-ECC7 1435276 2 L-PF ESA 8 ECAl-ECA7 6574132 Table C-3 (cont'd). 137 Three Order of Seven Degradation Pairings of Six Comparison Subjects Trial Types Stimulus Passages Degradation Levels 9 8 10 l L-PF PSB 8 PCBl-PCB7 4271635 (21 8 22) 2 LR PSB 8 PCBl-PCB7 7326451 3 H-PF PSB 8 PCBl-PCB7 2134756 1 LR ESA 8 ECAl-ECA7 5463712 2 LR ESC 8 ECCl-ECC7 7123645 1 H-PF ESA 8 ECAl-ECA7 7352416 2 H-PF ESC 8 ECCl-ECC7 6274135 1 L-PF ESA 8 ECAl-ECA7 5176324 2 L-PF ESC 8 ECCl-ECC7 4716235 11 8 12 l L-PF PSB 8 PCBl-PCB7 3742156 (23 8 24) 2 H-PF PSB 8 PCBl-PCB7 3267451 3 LR PSB 8 PCBl-PCB7 2156437 1 LR ESC 8 ECCl-ECC7 1357462 2 LR ESA 8 ECAl-ECA7 5164732 1 L-PF ESC 8 ECCl-ECC7 3657241 2 L-PF ESA 8 ECAl-ECA7 5412637 1 H-PF ESC 8 ECCl-ECC7 6274315 2 H-PF ESA 8 ECAl-ECA7 4531726 aL-PF: low-pass filtering b H—PF: LR PSB: PCB ESA: ESC: ECAl ECCl high-pass filtering linear rectification practice standard passage B -PCB : 7 degradation levels of practice comparison passage B experimental standard passage A experimental standard passage C -ECA7: -ECC7: 7 degradation levels of 7 degradation levels of experimental comparison passage A experimental comparison passage C - .. PT: ~ .-.. >— . . ...‘8 v \el.‘ l -| »\~ ~— 44‘ ,; .;, , u.- I- _ - ,— ‘ a a“: '— r h . ", r .. " .‘.. I ...: "J a .1 . .- - I ..q . —- ‘ k. ' 1 n 5 r. -. - - -.. ,~ . 2‘ . 1- . _ .. . -1— .__ . 1 ‘ - 4 - ’ " W O ‘. N h P . 7 . . .. ".27, "--V'L' APPENDIX D GLADC: COMPUTER PROGRAM USED FOR ANALOG-TO-DIGITAL CONVERSIONS . q - , . .. , _ ‘JI.-'MIL. ‘.'L L "Jl tum“; H431: l m; ' ‘" '1‘ ' car; ~- .7 >'—» - --c ~" ” ¢ ‘ . '\ "'Uf' ‘2 ".‘z- 3 "‘13?" ‘r 7:.) ' 1-1-.1" ‘5 - ‘1.)“3‘: .7." .1”! .‘Z."':: ll" 7". ..‘... 31.7 , 1.3 . d 1,. ’ t x J ., ' .. -|, 'I #- .-: 4 1 .. "'T‘.,‘." e -2. u, -...C' ' ‘2: L. ’ “.. I.".‘ l b I -_.- 3 . .. . ' p .] V y ._ - - ' - ~.. I h. 1.. ' '\ . . r ’,_ o ".1:- 'H'TT“ o 1‘. u ' 7 L.‘ . .- .- "'"".' 9‘3] ILL: | ; I I ‘ C: 7" . \ '" ' '. E:' " 1" 1’1 T I\' T" :I‘ 3; o 0%; _ - . ,L. U ‘ ,1 ' r - ... ‘--’ - 1 . 4 ... . \ =”“." \ e -1 r; --‘ .- 15.-.: a. LI ,1“! .p_‘_ ,1 ‘ I_'IP- I“ :H ,. ... ‘ , .. _., - 1.--..-“ 1': 1' "w‘ 2". ’ i‘v . It Lv-«H ."l.. f. ~~ :f': ,. .. .. - .-- ... .. .. _ l - .mlj" ;_u ‘ “1.2-£2! 1' a '12.; 'Z’L'Jl 1“ ’ ‘- .~ v -o. ~ J- ...R ‘JJ ' .2 D " ' I -‘ ‘Qf‘ o uL' '1“ -..- '1 s3 . - 1" i 3 g‘ I ‘, ... ,' T? 1- ‘ ‘ I - \ -4l .. ., . ....- .V‘._I 3': a. 3 ‘ “‘l :2,— l - ..‘ 1" 1- .—" .2...” 5 ..._ v . L H .- ,, ..‘ I-_,I_-"_:\ ‘\.L r.-"-§o- .'.‘: a .... e 1 s “1;. an -r~ . .. F - ,.._ bun-2d 3‘ ‘ ' - I." " I'Cz" : 7'" '13-." 7‘ '- ll' .1 2:. ".18..-- u . Hr- L1 5’” "' ".'\ Ll-“ r‘s) . .. _ . “UV :3!" 1’" u; .. fl . , . . f’HJU -.'- pr'xL a.) f “ ”’1 fl_1 V 'T_I :T .91 §HIEH Paar {N :1 LF'TF. .v F52 151.1le 5' ~15? T I N 5;; F; - 1 ~ F‘ .' 5.074172% N": T IN 1: ONE ELEM) . HF! ! 1 I I I n”- ,. l 0;) 1 ... I -. r—i m «1' M 1'1": '11 H m '3‘ -.q >..-_ .;L \11‘ ' 1‘30 .. F- -1 -,....... ..N , . .. ,...'......-‘ . 1" ‘ i. ' " 1:“.J."Y"'7'3 1"H\*L‘ ‘ ...Y "11:? I‘T.‘..’|“LL.‘.’T"-' LH-Iq I "t a 1d ‘1 . - I \- , DI v » -\ t v - '3 1 - , . . vI_. 1" ‘ 'V‘ "I »' ‘ ~ ‘ Dc ’ ]\ h 138 ... 9'! 139 N e r. .. . -. .. . wnHW! flw; 1“.U- : "wot ‘. '- I. J' ("| s: I’: ‘ ' '7. ’ - " "‘ "f‘. L“ " ‘ ~— .. g. I .. . ~.. ‘.‘ I ' r ‘ '. . .". .. . . -‘ , .' “I ‘6‘ N AH - t ' \ - .. . . W, .\ . n I “‘ I _' C. ’ a. - . . .. . .. a -- v 1 -\ 0‘ 4 . . . . _ .3 i ' a h.- ’ ‘ ' l- . .. . l _. .. ‘ 4 ' ~ ~ . -1 A- ...... -. ‘5' . .. - ..., -.-, .- .1; oLM 13.-0.~,j"—'I. I. I" ‘--* cv‘l ML. . ‘.L'.‘.L'..:.‘.“.- 2-1‘ “2.3!“: ewes a1 ,xtnuw TwaHu$Hs¥Eerhur;.-J ”'3‘; "1 JFJO C ‘~ J ..g :0“ - - .- CLR 91 " 1.7:. " '1 D ._'.° " -‘HL - .. . .. ,1 1.1-111”- . ._ :“U'I ...‘n-‘tL -LJ - - v-o - 3"“.ltl I 1"...“LL” -m'a...: \ - .—'.‘ .— «I «10°: . m - . Q ~ will“ fit ~00 2.e:hafie ”1:"; ;"_’r n1..,3'r;—, 4 %0 {WIN 3 L.) \- i (3‘: I"; O 2:3. 1~J 3|: Fur-'1" ”a; I'f.. ,' o- ...° 9‘: 5?-“3 zillif'H‘EEZ ‘ . - r< I: L. '7 .. 109 watt .vgfi 9‘17“) °. F3 3 ' W '1" ’ str ‘07 *UFT? AC": :.\:."-‘;'.; ’s. ‘".",’ (5;; Ivuyr‘-‘ 1%.::v1 3F ‘f‘t'l‘i 5'; It ~ MGR ‘1." ".3 - °) '- C1573 7 3:. . «hr-FLO a," ":1 5 1'15: 1. 7'7: 5‘5 "‘11-? 1-314 '._ :"2 I 9 WHERE ‘1‘ P 13' ' '.~I 17. II}: I“ '1: ‘ “ ' ‘31". 1.. . ."' 140 “.11.! ... -- - . _ - ,. ... -. ‘ 1. , .. .. VJ f‘ ' 1.: ..m - a a a .u o. N, ..v . ._ Ln. 1. .U .I. ,0. VI .1” .. . \ In .> . u . u _ .Ial. . - -,, .‘.‘.7r‘+ 1 - 9'; d - N "I!“ In .\. .2 .vL ...u 3 H... . w... ..N. ..h c... . m, . :_ NU ...x .3. ...... v .... 9.; .L .a. ...n a .. ..3. n ..H J“ .U 21 r- .... u 4 .... ..u ..N. ....H _. _ N _ v + 2.. w. a w: .... N... r1. c uh .n A. 4» r i w»! ... u.“ H o L. - an r «.... v... .N .1.» J 2- Ha. r... .- ... 21.: u .u ..-... 1 ..7 .\.. . t. - _.,_ _ u , ... . l .1... ..... Q... T ...." ...... T 7 .-. .... C .... ...... .w ... .w :- n a .. .... if u. i. L l... -L a ....v ...... .11» .H.. .t. 1. v .. . ...... .3 ..., 4.. ... .. T r 7 T. m T 1 ..u. ... r- C T .1 Y M. 3 .... .3 r.” 3 2 .- r L ... .. ... .4 J. in m. 5.. ....t .5 v. 1.; a a .u ..u ”.... 2. .n ..... ...fl Q ... w - ... a. H... 1. in «D u. ..y. :H .. a Q.“ «U - .... u, u. U ..., n. C .-.. .H... ...-. 3 u, .H ...... ...o . . a”. .... .... F m ... ... ... .U .... .... .... Z... n. .-t ...: T. ... ..L L. ...» «...... 31:..- .1. «.2. ...!“ . _ w! W. M! .uA r :1— «...; .1 .H o .1. In em: ...- 3.x.“ .fi. :. _ s a v EU ,— 2- , C. "I ' 4; 1.7. II.) 1 .‘V? a 'N 1:: E T fan 141 «3L905.n9r “*~ n —79 ‘. u «m- J 3 C in": .4 1 to \. QTLQQ Z. :, ;,J' 7'; PM“? -,. v... - -. \ C m + '0' ~"' I :d.‘:~ .j 'll ' ‘-> M . .‘ , ,. , .. . .--. MLV ngJmewTHt:% “CW «*;:w::«w“~::n+2 4:0} ‘31": x': uifivfij-‘Jfifi AA: u"~::Hr: -— . r- 'u - ‘ r ... L: A -0ur1 _. z) ‘0 .~ C. '- SJ‘ v’ '1|__'yu “i l“. I V _‘. ' 5-x u L;ur3 El‘lL-p ”..‘II; ;,'T' '7'“: .2! ..n_u‘; Ln‘)..__.‘.;. .‘1 ... ...-v -Ic-lb . -‘ an. ‘ C :wwE H 1‘,._,. -. ,. . “1:6 RL”1\ ‘I . “I. LLRB 1L .... -,' , . ... -._ .3 LF *Jfl) LT lwr Igfl+, '1R V..- a .. , 1.14.-'r‘ ' ._ -.I ... I '5. ‘ ”IL“: 31’. v .4 . ‘ I. 3’." -,. ._ . . .. ... fluv flUouT9JJHTi 7"”; ."lal/ 'u 3“) Ell": 0"“; Ox: { z.. 2 s f I 3 ."7 __.-. , :3 mqh w'_mw~g~n- ‘1 - ~ 4: :HB ~:'h4 Eat “a -_- -_ , nu: '_~\* -- - _- . _ ., 'r :44 :u3 +' 'HVU {m4 ~ ...: "E: :7 ~~l ‘- l d _-, ,. -~ “L5: \_-N¢ . “I 3 8 M 2+ 1 Q (:1 E .l; l 'J {i l. a a... ,.. .p. .f‘ C 1| Id 1| :mu ‘ ---\ P ‘ v. -0 in : :v. — :’ "‘- I J ' .3!- .v ‘ n-‘n NOV ‘Lv‘ :r' 'HISJ .\’-,_._“:l .. -v. . "-‘l/ ”I - '3‘ '1! lf’ - i “ \ :ud U'“JN'1U .‘u.o:; t... - .-. --w . +DU i: ’-*1 -.fl, -; LP: “ .. vyI-H'i ,: ' A‘; ‘ zus '~.7. -, . ADL F’ - _..”" .A ATM rB;r_-fiv ln‘r ; --- . -.v " ".a . LII-" I-nl‘.‘ ‘- - - v - . - .- - v- .. . - - ; . . -- a l ... .. c ’0. I l I ~ .- , v ‘b \., D I ‘ . .. o \ ‘ i- . 7 0 ‘ . - . 1 V ' -a -- . I" ' Q j- " ' .L 1 - '-w' w . - .l l‘- -.. . III . . _ ...- , J . _. .\_| - V ‘1 . - ‘- -§ “a 1., . ‘IIH A ... - - On '»4 -. . I '\ ‘ 1. a ,- -. .. ..l . 1 \ .a . - d - ..— '1‘- u‘ ‘1 _, 3 l.‘ .. .— .. . c -‘ . . u. w .. -. I .. .~ .. v - .. ‘ a I. L I 6 , ~-- ~ L O ,_ . v - ' ' ~ .‘ \ . .\. .' '.. . q ._ . ... .. .. "a. - ' - _. .- , A \l .- ,,. — . . I.‘ n L- .\ § ".2 ' 3." . — v— I I .. ‘ - , . ‘ .‘ u. - '\ ‘ v. | u . ' ' A ' - ‘a 1‘. .i. - - .1 . u VF.’ ' . . .‘l .4 ... . - . u ‘l. v ‘ I ...L. 1A: .. . ' '3; . - ... ... ‘ ¢ - .. , . .,‘ \ n \" ;"u I . I .. .: _. s H‘ _ . . I \ l \ o . . 0"‘ \ 142 A. - 7. _. ' . O l. . a ; - . . K -. q . . . _. y‘ D I .. .. . .. . .-‘ - , . . _. - . ._. . . '\ ‘ . - v-‘r'.." :\'i\ 2‘ .. . . .. .' vm .. . u g v ..- . r. .. v k a - I - '\ . n. ha I , ': ' ‘..' .-- 9 .... . Y: 3., I. __ v - .. y- ' .’ 1.. 143 I.’ ‘11,.- \A ,u ‘ k... .— I; !' ...: ... ;: .‘I‘."'.‘ .. '-."L- -2 9 H... s u.‘ - .. .. .f"<‘l." ,<.,\|v: I- I— a —‘ U i .I': :40 3i ‘ I o ‘1 H';V‘:"l"\’:‘: . ‘ ,. .. .-- -. ‘~ 1 - ' ' .‘ k . l - I. . . - 73 _3' : '4ll " ‘d .. .. .. I .. . . ..I " 7‘ .. -1 "' " .. - . . a. "‘ "" 4 'r1 " ", r - . 1 .. . . . 7.. -. .. _ ( q' -- 1 I U "'1’ .. . I .. .. .‘ .. _ ,. , I I «I '-.‘I l I. ' '~ .. '1‘ s. n . ' ' '«I ‘ - - y '1 ‘J -- I I .. l .l .. 4 \| -. . g I I ' . 4' ' a I" .h “ ' 1. .. . , ..‘ . ..‘ . fl ' I " . ' - J 1" I. ' ‘ .- T' .. .1 u q t. ,‘I ‘ ‘. I _-\_) ....- ...,1 V- ..l .4" ’1 .1. ' l .L VI] : giflh. -2 ‘I 3 fl ~3- :1 I Q ‘\ w?" I -I 3. '3 L.I J) kvgfilx; 5 -h ‘l '7- ‘ I“ . A”. O 7 -... .‘ . L“ v L a "- L. ’ *.p Av : .‘(1 q' ; ( 1 L.) «n .-4 \ .Q -‘ a: r. -\ ~. - Q - Q - ..., ""JU " L'bl.;!;r'\1 :-'~f "ka - . . r- - ijJHL" 5 l 54 _" f. "4 1:. 1 a :6ij 144 1.1.: ‘:3|-'§U4:."""~‘Il. 1.“- '1! " '3' Fin-2E ' 5") _f. L .JI .. \.A . -. ~ ~ ‘ "" ".‘- ' ,-" ' I ‘. .'. L“ ijf: ST “0. JR (3+; 1% WU' ANN '2 H1 R1.,v MOW 4QI‘3*2 ‘00 :L-\: 2‘40 "t‘vk’: - I v {1: ...}(I -" . H; "-~~ - ~- :wwd "a QMqR Q3 " 'I“ " " LLAd W- "- " 1. "wv an - I :‘L- 1“.) '"-. CLR? 83 SNAH :3 _ :1 )\, J ‘ :Ji I w I] U [VII 1 .41 (”I J N I3! L1 1 LLGMI n D a 4. - 2313-11.- CLRB ~4 3rxs :7.»4 Flip .2 “‘1 Q. ~\- 9'! :39 :3 .rnxhéc tGU;LriflF-$7-4hfi’90kfl'wlis4mrn§3vFU , .-..-". T. ,-...- f '1. n-k‘l +MZD; q‘ “7' ‘-r"~ I; '_z .nzrrcT .' -I'J" -L— '"'“ d ‘ 'c . ‘ : ... ' '1'5.’ - - x -. .. v- . .-— 71L“; 1“! a y]. 31. _' : CLF fiflmzw QLJi w --L”¢‘ - rs: ” sing 4“.- II .» r: ’- u- “7.3;,4x‘ ...l__.a A.‘ III .. HI -- :- v'v- ' LJ‘T u' ' 'I“ ”I T J L w m 1 + I J 4LU (LIWDP*R ”DU ‘.+Z-RERV§‘T any ‘:.:1 W‘U *S-ZaFl SLE (1-53 39F F4 SUD :L': §4 «nu 13:7“.J) ’ Ci C l:| 'II II ‘ ,I I ll; If' '11 5L: 13C 33 515 Q4 DONEIF _54 LT #Wb U} l I'! _- -1.-,, h'519UWU 145 I' {-5 L j. ‘x l l..I v LA C": , 1“ I" q \- If! m.— ... ...." T. NI... .. . .... .2. «I. :- . - ..-. .. I- N|.I O RPLAY: .... -, .- L .. . . . v ‘ f- I |V~' - .. _ - ‘1 APPENDIX E COMPUTER PROGRAM USED TO CONVERT DIGITAL SOUND FILES TO ANALOG FORM AND PLAY THEM IN RANDOM ORDER .- - -. - . n?" :H. "4’- "’J‘..t..' ‘- ““3" 3' "I s -A If-I I- w -v!:; 7 :'.I'.1.- '. ‘rq z...‘ ‘ 1‘ .xl-‘n .b . IV... ,I\ 1 4. IV: -- -.-. _ .. ,-_.-- - . .,.‘ -.- . _. ’-'_F"‘L-" .. .“MI ‘»' ’" II‘"-e".- ., 3:;.(_':‘.-:.‘\- “11'." ‘1 . .3: -.-_ :‘-..-_ ..fi ..- -.. .. .. ..- ,l- . .. w”: 5:3... -.. . ‘41-- [tr-'45 .."._' . ‘1L:‘1-.3H~ : ""4 .r":'n-‘-:‘I .'. ...... -,., -. -. ‘- TWLFJ 5"” wE 1: - f:- .77."); I ELI}...qu ' ‘ -... -1' .. __, 'Q' '\ 1" ..... .. ._ -- . .. -- . --.. ._,- . 2 .. I _. . ..a ‘4'] f' - .IL NF. .m’k‘ .. \... 7 1.- .‘I 1 :' .. ... -... v ..- .- - --.--. .. ---..- . .. ' L. ' ‘f‘- *3 LL." L:.' ‘w .2" r..." A‘ "M ’ 3"..." ’_:g.. -.__; ...-...? “a r: ,3 .:; 7“ .3...) .. . -..: ~a L - - ' 1.. . ‘ --.“ _L,,.- .. .' u- ’ _, .. - -.- _ -. .91 '. ~-~.- '- *1 I- '7 "3L; 1". rlrn’ 1. ‘I "2 ._ - .. -. - -. 4. .. m I ... -... .. ‘4- ' .- -. L ,~. ‘ . ..~. . . . : ..L. It) 4 "Lu“:a‘ . - .... .. . . .- :-‘; .. rd- _ L '11: MHJ -N -:; _d 3:- 1.-.: -:.r-s-n~ "ITJ‘I‘. . T" .- - .—-.-- - - . v- -\ U ~- ~ . 4 o .- ... u. .. ... .‘. ng‘, ,. ..\ \.- . . - no ‘ .. .. ‘ .-.:_ '.."!F~""'-’.1‘-‘f“ :.:‘! I‘m: L?..’c--:*.r,-_:'T .2): 2 _-{:< ;'.I z-.) L," .vv- .... -v -'N -r— . - i- r- n. ' "s '\ ‘_ >‘F w .1 “fight 3 I,“ : a: '-H‘J.L' NC) : 13!: t. ..«N :3 7:04:21: - ...--. -..- . 1 - -.. ._ ..-.. . .. - V. ’ I ~ ‘ u‘ ‘ - L ._' .v. .- ‘ a .' -¢ «v. ~~«. ‘f ‘ ..H 1‘51. '3: ., z’ ' '_ '_ ~ #:1‘. ' ".91. :'!l.'.'-l1’_3‘.-...'uf H‘JJJ ‘Ir my '1 ‘:-‘ -. _'__ , .-W-‘ --. _ . . g-.- - . . I I I . . I.‘ -L19- ‘E‘ |l|:l'l b‘l‘l 7' :1“: L" II J‘ ‘ _-. _ - - ..-,. .. W-.. --. .. l ‘ c ' . - .l ‘ -. ' . .. “J. ‘.?:‘1”LI".‘-L:-‘.-. ' -.. f't’.” r": .-" mun - . . - .. , , - ,. . TWI; '-)'.‘L' r... ’_"_.‘T "“!.i‘.‘. L.’ "‘H‘b'r'“ f" '.-i‘ .-- -_,._ .— -....- 1,- . ‘1 .( .. Us): "I "' LL ; -I_‘ .L_,. ,J WA: _ I“. H .'".‘ .’\' 3““..N. ._.-. -.: ._,__ .-.... .:1. HI Lint- .:-.“!I .LM-..‘ 146 .— \ V‘ . a ’ ’ Q ' I .-| ' -. ~‘I '1'“ .. ... . 4 n I- "I ... - T'...- ,.I_. .. ..i : ... 3 -' I"; ‘3 I... 3.5 =uu 147 I t I j ["7 IOVI‘I“ 3 ‘3..- II I; —- 7: .I: ’7' . I - I I 5. ..- ~.. 1.. .. J - n- .. .. T. . .~ ... ,- .. .. l b— - - -I , . ... ‘. . . w , -. - . I . . .. v I ~ . u. z -. I: I‘ :- ~15] . . '. I... .‘J - l.. "I \ .. -1- 5.. '. H.’ I’ :‘2 ." .g I .\. N.” .. K I I I " fl do.- ’- t .< \ PI A. ~ _I Own” " .w‘luo ‘1’; ' .'... f" I~\.--...~ ..‘I‘u' :- I‘ll': -. p, I'- d L. . I I: I...” V. J :. ‘1' :.-’ ... .' ‘.::.- '. n. .. ”'4 . - ID ' ’....-.... ....' -..J .. -I... .‘.I .' . - .. .— .. - .. .- .. .. —‘ .. ... .. ... . n .- I '1 I< I I hp" . _ 1‘ .8 .— ' o I I ... - ... . In. .. ~ .... - -. so ‘ . ‘ -— . - - . . ~l . ._ .u. h“ . ... 1 |. -l 1 .. I— * - v ‘ l" - Q‘ a .I h I‘ .. .... .. , ,.— .. .. - .- .. .-. \- - » ._ I I - ~ . I I.. .-- ‘.."" -"!..‘L..... ‘2‘ ....-.va ..Y ‘ .. -— .— ‘- .1? , .— . .‘ v. . . ... ~. _ . l -. ...: . __ _ . .. . . ... \ I ‘ . I - ‘A , ... . -. I: i L L. ' n -, _ .4 ." H I... ." »_ ..1 ~ ... . .q ,- - - q - .. .. - ,—. .-.. - a-o I k. g- I 'I I ~ ‘ . " - .- ~.' - " L3 ""1 ." V? ‘7-.":.”‘.-"’...-. u . ...- . ..- a. I - ‘ I l a‘ .. . u; I \ .,. fl . Jar: ." . r: |...I~-.'..I.._‘ L... ..--.....- .. .....- -..—... .... . . —"I" .. V' .. - .lb-ou .- I... _ in a. ,.. .- v (I. I .. . .. .. hml- '1..- '1: I..'l-‘.\..‘|.~ 1‘ J1-" ‘.-\ 3.." 1". .. .. ._ .~ .— ‘ . ., . l ‘ .r. .. ._ . ..',_..—.-,~ ,—. ——. I- a! p. .. . D- .-< - I, . I. .... ‘1. I I-‘I—o ‘.I.—_— a I‘ - I .A._.-. - 'I‘ 1-.....4N- .- ..‘L‘ 9'.) . I... I.-. ...1\:I‘.1-.. .. ... - I. - I ' l---- ‘ .. n " A ' \ ‘ l .‘, \ ‘ \ .U I ..‘ é ' 'J -‘ -.‘ "I‘— - .~' - .- ‘ T ... - . a- I- ‘ .~ .. .- . - .- .. u _ a - '.. ..J .. ‘fi . ~— -. w - . . ~ ‘ ‘ c - ... ... ,. [9‘ D .. . ... 148 . .. . ‘..,‘._ ...: _ . ., M . ..., _ ~ .- -- 1,": "' ' .~ '.'.‘ ' I f'. -..V, .~ ‘, : .. ~. . ...-..--1.‘9.. '-J ‘ A. .I ..‘A1 ) . I a .‘ 3., . -1,~..n‘ ‘ - . . .c . , . 'Jl I \ .' I I 6 ‘I . I" . ,. . v ‘4". ‘ v“ L.‘ .... .... .‘ ‘bw - \ . 1..-: . ~fl ~ ' ... . ,. ._ .. K, A -. . ..w1r..!\ ‘,.U\'.g — .‘ ...‘ -a - ‘h4 _. , _- .-" 3... , ) -.. .. | . \..I ..‘v' ,. 9 -’\~‘ .r « ,_ . — -- . . _‘p -4—- -1 o f‘ . Np! ... _ ‘3' : ‘3 -' ’ I _- .. , .. . ' —, ‘ J ‘ I' . u’ \‘ -‘ N. .‘ .. ' a. _ .'/ ‘ ‘ .... 1‘- ‘1 .. ‘ AJJIHV ‘ - .. . ... I'llv , - w... .- .. .- _V . ... .5. \‘l . .- Q‘ n- M - - v- . 1 n 4 ' \ . .- a .a ... _ ,4- ~¢ . . .... . ~n ’ .. r u w J. ... ... 1 .. _ .. .. ,. 4 r. b‘ .- . a." .- .." ... .. .- -. .— — . .. . ... - -1 I— ~ ~ -. - - . .. . . I" ‘ . a ."I .4 ‘ . . 'vl .. . .. |‘ \I .4 - .. Hr- , - . . ,4 . _ ... V- “ ~‘ ¢.‘ .. .. a I .‘g a " _. .. .... ‘ J . , _ . . — - \ I -. . ‘ ' . .. .5 .. ' _. .. . .-. V '- , " . .. .a \\ n .,I“'l"’ V. . . . u . ,. .. . C. ”u. .. '9" 5.. 4 _ ‘ .. .- I ..IV " .4 ...... ~- . V ‘--. h . ~ u m- ., . -.. . .. . -\ - r. .. 1.. .. u, .- ..l---‘ ' ’~ ~. - '- -a L.- _‘ . - H, . . . - h. .., . . .. . J. ' .. n r .... . .. A- . .- _ - u .y C \A: - _. -...» - n- - - '1. .<.. .- -- .. .. .. .__ .- .— ... - .. - _ - vs .. - - .. . A I L I < Db|u - ‘I q - "c. n. a... .u_ _ . -.,- -5 v. 5 . * ... w - . -. 1 .-‘ . h, .. LJ x .. , V , ‘ A J ' o .- . ‘ _ . .- . f v :‘f: .; ' I ~ .§ 5 - . ‘. - , .- .‘ Ab ’ " " ‘. - t T... .I v .. .. ‘ . :‘zd. : g .1: - . . ._ 5 ‘J 1‘. .‘ T‘. 3' ~ .- .5 ~ ... ‘ ‘1 '~ DI - .-. ‘ . x p - — .- . . . .. a ‘ - .- V r “ “ . '1‘» r- - . ... i ... _ I 1 ‘ d ‘. - -‘ ..- . ,.. n. - - . ‘ \ .. o ‘5 m‘ . - . '9 - ‘~ : '. o .. - , ., o — l- ‘ - - 5.. o 5. .. , . 2 J. ‘ o m n. \ Z ._ . ' 74 _ . ,. . .- - - : .‘ ...- , . " 1' (‘ 'i - -. . q 1 m H. L 1’; I '\ -_-‘-.J- ._ , , __I 1 ' h ‘ _' :' -.. ‘ .. '."..'5l.-' 2 *4" ‘3 I. >r ‘n. '-— 4 A .. .. .. .5-.......‘ ... -. 1‘ 'v. 5. .. .. .- —. .. — ‘- I . v- ‘ .-.- . - .- r . . _. u ' l . .. . - ‘4‘. 7' -.1‘14 -3 .T‘L,‘ ‘ - - . -u‘. ~ I. a. - 3 .. "-L,..,. - -u'l.‘ .. V.- > .. 5. -. ‘1". . _...- . L; . v-1. a; .-m w- .— , ' .. V“ n .. _ ‘ . .. - .. _ .-. . . _ - - . H . . 5 .5 _ -_ '_' .' :wi‘! f._ _..* - _- ".5. _ _ :. - .‘.‘~"‘ .2 ... 5.2,. .- .u -' I .. ... .. ... .- .. -. 5 ~ “ I - . L .- ‘ ..‘t‘ - ...—... I 1‘ 5'. ....4 o :— - ......_ - .. .. ..-.. .‘. . 7 . , ‘ 5 .. 3.. . ;”‘:n ;H"H~T' xuwu? -“v 4L :T«;' .. .. x .. I . r' .. -.-.-. ‘ ....... - ... - ‘ _ ... -..-.- ... .. . .. . . - . x . .. . .- . .. . L... :2“ 4..'! -.- "'7. ...3' ‘.-..-.. . "'.-.‘~".“'... .‘-..~'.“ .- - . ‘ ‘- II I -. ..5. . ..- .. ... .5 . .--~ - ..-. .. .. . .. . ‘ . I 1 I . a . , 4 ..‘.. .. .I' . '- - ’ u L. r'.’&’ 0...: .‘ "J -.\., - _' ‘4 ‘I .. I... ‘4 'h' "j .... .. , . . .. _Jl' .. ,:.‘,T.-_. .:J ‘. .. ‘ .-,.-. .- I - 4 p. L C 1- I ' '-- a“ I- h. L‘ O ' ‘. ‘ . a- '7- f ~ , -. ‘ .. “... . ‘C O:— - l .. -. .. .. .. . ".-.... -.5.‘ ‘ .' - ' "I" ..l ' N'h- a~ 0‘! q' ' .. . .. .. 5.. ' s— . .. 1 ‘7- . :' «r a, _. ..J». .. - .. -. . .. - -.. . I x -‘I ‘5 .. -.. ' ‘ . -5. .(.‘T' ' -' '-v...5‘ » . m -\ 1 a V- .1 | -. <‘ . ~’ ,-~- -..-.- ‘. ,.V' _‘,.. -.--.. ,.. ,... ...-....v--. . .. - . . a. - - ' c. -~. . ... -. L. .. + _ ""4 .. "3.. "-._ ..l"" --f ...-‘ r T ".3,“ — ~> u' .. .. ..‘V, _ _ . .. ‘. .. ...“ ; / ,. . A .- _ - .‘_\.) $1.4. 'l v-4' '3' . 2 .. ..‘ _ '. ,- . . ._ ...u‘) .. 7 ' -1 ..\‘__ ._ .. .5. .. .- ’. ... -. .~. 5... -5 - m - ‘ "vr‘ ,. — \ - "‘- - — ‘ u ' 1..- "..-._ "5"! -- ..4‘.‘.-f" > ‘2.» ,... .. 5 .. _ , .- 5‘ ¢' v' u, .. .._:... L.“ .:... ..r. .. .‘ . , . ' \ «. s . - ,\ . .. .. 7 ‘2. *3: "H N -..“ .__ 3 «A "... r I .3 - '1 .‘ ._ 1" I '7 ..-}.-J.‘ .. -... . -7“ .. -.- . 5 .,- ‘ ..- 1.‘ “hr. . ~1“:‘%~ 5. r- ‘ ;.:-.;:.'; “‘2‘": 7:5: ': ..., ...,... , -51.. ‘ .,._, -- . . 1- ' I -- . ' '5 'r __ -. _- .. - ‘ r ‘I ...- "L. :. 5.x: .-- . ‘L' - 3' .- .. .4 .-.- _ - - .. . ....-- . -. I I . ‘1‘- ,.'n ‘ "Iv..- n) .. ‘g -.— Iah IV I .... _ ,.L - . .. 'v-L. ....» —‘(_ ‘ . -... ,1- .- .-.. .. .. . ‘r- .‘ . ,. - ' . ‘ ... I ._ t L ‘- ’ I - Mn; ,. 5_ A. ._ . ’. ..‘ v .. 1' .. . . . _ 5.. . , - .. . -. 7.7" .‘I‘\o ' .-. .. .. . .‘ . .. ... ’I. »...L ..-.._.... .---) -.. fi‘.‘ 7‘7..." ." ,_ z ‘ -* v "H- 22.11". -11. . AL. - ‘- .r.‘. ‘ .. U: ’ , ....—- .. .. I .‘M |".‘|l ,‘ .. .- . .‘ I. ' , I UL} .' . 5.5.- - 5.. . f -.., L14“ ’ :,\l!l" ,.-—. v ‘ -:' ‘ -91 , :-“_'»' _- -v -- -‘ ‘\‘ .. _ .-’ -" .'. If. S".. .. v. - - ..‘» { .. ”T: . ..- 5 . .- - .. -.. ...”. .. -. ;‘ »-.- .'r---1_-.,\H! - _ ._ -. . .. .. 5‘ -. \ s... - . .I , - 150 n 1 ‘ I " " . .‘ . ‘ n ‘ A. A A I ... a l ‘ .“ -h.’ ,... ~ - ‘.I__ . .. 1.- . .. , .. .. m _ \I. lf" —. ..v ‘ . . .3 W H - v- I A- ' V - ‘ . xi '3 '. - . ,. 4 , ‘_}. ‘, ".. ., . ' " J" " " -'-‘ ' " AI. ‘ -. . .. . . . -_ \.1 "i .. ' .. .. ‘ ~ n ._ .‘ ... ... -‘I"' ‘Iu...l.' ,- ..‘ ~ .. - . . | ,. -.. . ~ : “ . "'— ~. g V-‘- .- ' - v y A. , a . _ ‘ ‘ ~- . 4 . 1 ' .- ' ' 4' .. - . .. .‘ . .. ‘ . - .. ... ‘ _ .. .. .. ., - -0 an ... ‘2 - n . .. ... .. u. \ ‘ .. V ‘1 -— _ .. _ ‘ — . — ‘ . . v... u ,’ ‘ ~ 3 . "t. _ _ .4' V v— - v- u «. m,»- 1 “1*! - 17 to — w u 0 -' .. ‘ .. .-, .. — - 1 . N -. .. .- ‘AT 1'! ‘0‘ ..14 1" ‘1 - — 1 0 - ~- ~ I I. I \ < .4 - 1‘ — —. .. _ , . . ‘ .. _ - - .... ., '~. "1" I'V'I‘ "— ~"' it 5..."! ,.. . - .. H .. _ u. -... u ,- —. - v. .. -v I -.. - ‘. — n. ._ .- ._ m . h - . - . .- .. ., A J. -“.. . .. . .. ' '1“) .. .. ...-. -- - -- =‘ :n- a ‘I i L - - .. - I, '1 . ,. ... H ;- o- ‘ ~n- ._ ‘ I .. a a! ' '1 Q. ' , 4 - .-v \ a“ I- " -. - .' .’ ... -1 1' .. . .. -- x~‘ 1 n. u 4;.— . . ._ v . 5.. . *3 I 0 - ,— ..._,- ..- -. v _.... - I. - - . . _ , ~ .a' -- a ' ~., -. .--....- .' h .. '- ‘ ll -. . ." r - . .. .. . 1f , .\“..‘ :. .-.- .” - ._ ,..A...,.... .. -.... - - \..i‘- fil u ... 1 4. \ . .. ’ . "a. —‘ -— n ‘ - -_ ~- - A— 1 - ' - ..- 4 ‘ ' \ ~- ' I . L ’ [PM .‘ ‘ , . — -...-.. a .v‘. 1 -' l .. u _ .. . I _ . -ta 4"- 'r --. .-_. -. . - .. \ II ‘-.4L ‘. \l - 4 ... «n» _‘ — ... '..| . . ... f. u " '. -- .. ,. ~ _. ' n -. m .- . < .- t .. v . " . _ " n k y ' ‘ . a. ‘ .'. -,--2- "WM .-1' m". 'x- ’A - , . ‘ ,... , . r.‘ I ' \,t- ._ O ‘l 4 1 “_- t. ‘. ‘1 - ‘ntfi‘. . _ .- . - A.. "‘ FR... - J ‘- .\‘l I l - -- - . -. ‘ ,..._ .. .., ...- + : ._ - . _ ._ . .. .. V . ‘-.. .. ‘ 1,. \ q .:,__ ... —. h Ll ‘ 4 a- . ‘ *1 .l .-.- .. _-.. ... .. - -.. u .... __ <‘ .._ ‘.. —‘ ‘I‘Ir 5- I . I ';. 1 ‘ ‘ _ \ ...- '. - ~ -" "".....- A? ‘-.".. -.' -7. . -.’ -‘l‘ ,_ .- - ... -. . ‘- g - V ~ v- - u ‘ \J... Q: J . . .... ...... .4- -- . .. .‘ r gran... , fl ’ ' - -. .. - .. ‘ ... _ , .‘ .. I - “ 12 ' “I‘fi ‘ . ~I . I. ~ I— < 3 l ‘ . 'II . .g '. ‘ - . .. _ - -. . - 'v ‘ ‘ .I -§ ~~ ‘ - .:_ . ' ... — - ‘ . ‘ .-. ' . ,_ ‘ v n .3 W4 ‘. ‘- .. " ‘ aw- - s. n . 1 .u. . . ..J-.... . .. .. . . ,.. . . .- . h. 4 . , :‘ ,. n. ‘- I t ‘ -‘ I " ‘i' L- I" -. ... - ..o' - - 1'. 3" f . .... ., , . . — .. .-. .- ¢ ,.. l3: .I'I ... - .. — - -. n- . . ._ t.’ .‘ V... .. . - .. - --\|‘_ ..a-. . .!.,-‘ - n. - .. x. a 4 - . . A _. ._ . .. .. .. .. .. A. . _ .. “,.. ~ .. .. .. \ 1 'l - ) . . é ’1 vi... .) . ..C' - .‘T' I . I .. ' -- .- -_ .. . . .- - ... \ 7" ' \ ~ | ' -a. .. .- ‘ fi 'q I ' --. ‘- \ — ... . , . ,..4 ‘ a- n | .v' - . L. .. J . I :_. D 151 ...-.\. ,. . ...- s. n. n” v. .. ‘ -_ . ”4.1.0.. “.v'l; ""‘U r! .‘\."" '- ..'._'." i ~-“..i|: -- ..-..- 1:: 3, run." I ‘v... I ‘- Voufi“: ' x1“ 1 A .. H *'.'l.' T'T"|‘;;I'. v-w- ' o "' I ‘ U III I ’ I 1 ‘|.4\ I 'v- ‘- ~- . c r .., . _g.“' .‘gq’fiuT o .. . v I ‘I'L-AJ :- ..l L.i\ Wm; 1" .‘..£:'-' " -I' 4 .-..kz- ,1 ..\... or“ ..."! ‘1?!" - .. '_'\”. :.H'.' 7‘:- 1417:": ,.. .‘ ’ -.. ht“— “WW - -4 ~" . :- -. . ---~.' MT DI ....I.. {Ll-I , L”: 'l ——\‘r- < - ' n . L. 0- {-‘u ... I... \ r- “.l‘. .1. -< . ‘ - . . - 7‘ , I— I - - .. u’ - ... ’ l. I- _- i— n .‘l v. “ a ” ‘ .. . . I -.-' 4- ,. :3 ’ r‘n .. A .1. ..t l r“ V>" »~‘ _, ... _, '3' 23‘} i? L :" .'. ‘» “I ‘h-H‘x .:"i .V... ‘2’) n\-' #3:: .h‘ " f‘.‘.4' - K ‘ 3' ~‘I _-\ -—-—- ... I g' .-. ' . ~ L. .‘-* , nt— I1.- "no. .';-t 0‘ u s: :u mm 7‘ : ‘ - :u: T , .1 .I‘. 2‘ I , . n , -. .. Q. n ~' .1 v ... , .4. "i ‘ . r- 3 ~ . ~- ~ ' 7* .-' L :3 .— :a‘ .— 3'21‘1 D P: p-~- "\ ~. - ““1 4 ' \ -lu~ufl H’. - - fi‘.’ ... l I L L. :1.” .43.“ 3 ' n. n '— u It- . - ~ . ..‘ .‘ .y .. "2.. W. ff - : ; , _ , 1;. ‘ t I --VI -l. ‘ 1 l A- “5 M; NH .. ‘- .. _‘ I", ,‘ ') ' ,v A- . _ .. ., . ‘ ' , 4 .. _ .. l.\ .. ": v T: ' I": t" 155 |-~ d ' _ : - ..w"‘I" I; , '.‘4‘ '.. “XL” -- _. -. AhV-a ” ...,. . ,4..- .. I.“ I _." ."'"!""“~.‘..." A ‘J . _ f. , '1, _. 2' - -r":‘ .-.. w a. " "4.". .-:~ . . . .. ._ . L".‘-"'.. t *' ‘. glut; ~: ‘—‘ to ‘~- ' ‘ I -—\—- .- . r -- "‘ -'v"“ V,. "' ‘61.:7'3" 3...-f-.';T" I u. ,- "v- .., \ . -~- ~;_Wn" zhHNNEL - . ‘.-~....... , ,.., ". ..LJN’ 1?. Lu 2‘ ..'. , .-.- .. .. i .- "‘ -n-z - ""3. t. 2‘. ..‘fi .“' .‘ HI I. “ T' :‘ ;;.. \. ‘1.“ ‘...."LJ LI A} ...} ‘. - .' q‘ o“ v- '- 0"\-H'Z.«";f'?L.-_Z. 4:!” .qi MEL—3.? :7 .-' r“: - -- -. ngLr' ‘JLNV '“J . . --. , ... ., \.‘.,:wr. .-.-3... «I. .. n- .- ... . ;I;',.'4i\. , 1L!” * {AJ- ."tJ \ ‘,..\-, . .-.. ...... ' ...-‘5. ' ”J3 r1.) J.-’ . \ v-~- "4'“ «k0 .U 13:44' «MQL 31 + gr :N . >.. ~A-x . -|V '-~ —‘ " UL TU - "L '1” -'. . I .n v " “" ‘ When:- I“) «Hm‘C- ‘_‘ ....- ”....-- ., -‘ 02-4 :L! ‘__s;‘.r'rj_ II ‘4 ,.....” _; .. ’ -"u‘fl-H‘NJC ". " ‘- .- .- ~ - .. ~v ... ‘ o. I .—Q .‘ , .< ’- o . ., b \ -« u. \- ' \-' “ _- ~— 156 -r . - .. a: L ”C x, i”‘fl’»“H' 1~*'n;- f ‘NJL k_ ,l‘ A _. - , ‘5. ... '~‘-. b: V... .41.!) “I". . ‘ .ni .‘-’L-A. I- ‘lu'h Q‘uJ, :ul \ .. -- . -. . ..._. . “‘W'} u')""“ “‘x O] f“ j J W - ‘ . - H..- h ‘. v- . ..1 .-1- \- .‘..7'..1 :It~ ‘ i. T»..,._-’;_ , . ~. .I 1:13.“: h -\ - ,.., - y... ..'-I- "L a . . . .‘ .. -- -. . 3'». .~ «HI; F - n ."‘3’ ; AN” 5 . v... - .. v - .l . .'., ' 'v-~- -\ .. l '- : T m4,- ?4W .... .--” w . ’ ‘4 a.” . . , ‘ .- -.. ‘u.\:. r .-H".‘ .-~;v.-. *1 I" " -1 -’~'r' :13.- - - .. ., ”...‘ , . 1.. .. '1‘ " Llll '1'»... 3r: a..~'_ -v ‘;NFJ4 . ...- v. ,. .. __‘ < H. -—, . .. .- 3‘ “mm J; * 2 1““ «r*'q *ué': P;1:: l - '0 "; u-‘.- '.|. ‘ L '- ". .. . , - .. .. _ - -. . ...- .q-J’ 11“»0" 'j"“ ' ...! _ ...: 7...: -n , ~ , .- - ... ' ‘ flu-.~ "L2: 5" - --L .- ...-.; .-'_." .. ’1- t -l- 1,}: .. 3:1,: ..-". «"II -...-R" .'.'| '1 1'1-‘.-.. .7,‘ ... .. .. .. . .. -‘.. . ,. , . ,. . , .. ‘ V _ nuu 90: V. 4 ’1' - '- s..-v.~_'_ ’"‘| n. .‘ .- INV. R” .._ .. . ._ .. A- . -. ‘. ‘L! ‘_ i } . . ..-. . .. W-‘ o . _I : ' .o”.....' ‘ ,.."L: -... - . - - . -, ‘.‘ ‘ .-. i. .‘ -_ 51'. I - ; .,~ ;- ILr-ac‘llvz.-- - " z“ 1.:91‘ . 1") I ‘- ‘ r- . ‘ -- ..-. .-' . .- _. _ 1. » ”\2‘. W“- 'i',"-' ‘ " 1'...” "" ' " TL". .". -27- .. l“. 1“» , ..L. I , ,. di‘ ‘ . . - LI I, 1’ u. l “.1, ' 1 \- r "w vrwr'v ..‘~l 1‘ ' an‘.-‘...f 2~4\i|' A - "$.40 .. ...—.--.-. _. “ E'“PLC «v- - ' .‘ . > 'E A” 119“ - .u "\ ’ '-".‘ _ .... - v '0', ’ 51:- fl , ‘\: .."' f!".* ‘. -._..\. - ' l‘ . w-r' .. . ...... ..‘ Q'Jw «I, ‘. .. . .. u ‘- _ - .\ ,.._ .. .. ‘1" I‘ll ;.~‘v ;v -‘.|Z.":J: -. .p.1" ,y: 7'“H ' 157 w _ "‘“rAr I i"¢"l =11: “7‘ | ‘ - M; :1-1' . 1' r" —r- «44.1 T. 5"“ 7'\ ‘4: L l'.) |._ "x E . \ " - HU—O L! -< .:w M. 1 _ . ...,. _H-HJ - .. V :s’r .-4 DI\0 U P' .- " ‘ - ‘ - - n - 'lJ n. ., u _ -,> ‘ \ ... 1- . , VI». .. .- .- . » ~ . -t. , .‘. I .. ‘- I ‘ y' ‘1': .... It .- v \'-. ... . 1' -\ v . m ... nu, ‘ ..‘v .m ... Y _ Y ,. ‘ ’ I - mu _. .‘ - s H.- um 158 v 5 '4‘ o .. .- . .. . .. -, '~ ..\ M . 1‘; ”I ‘f . ‘- - 0|.- 1 . ..‘ {‘1' '10 1 . ... ... - .. . ... . .. .. -.p . 0‘: n ." - .. . ‘- a - ’A' a .: ~-..'. '- ,_..;‘.- ~. a ‘> ... a...‘ a x.,. . ~- ,,. " : ‘47: '.|' ' ’l ... 1-. 1 m.- 'a' an ”if; _. ‘ ,... . ‘ _H +2.4. .. .*.. §:~Ov ‘ val it '0'! JV. l:- ... .... .. ‘l .- ' a .l ».~.v— Vt. ' ... .. l 'O 3. -!~ ' “ \_ -. , ,- .9... ,. . 4; ~ .4 O I ~ L< t .A ‘- . . ~‘- '" . .. ' . Q1: . -- . .l I ... .- .. . A . . .~ ' ':.J~ ‘ L1“. vi .a—. --—- v . >- . 3‘! ._ U 1 ’9 .- ... .- ... -. - . I ' VI ... 1 1.. m.- .4 1 1' ':' ..g': . -.., - I 0 - n " ‘J ~‘a: . ‘ V. f‘ .A! - -‘ ..1 Fly! '— to. .. ,\ ‘— I |.. " L .w‘. ‘v.. . «q .— - -~ . , ,. I .. . ~ "‘1‘...‘ F... s - \I .. a .. 1. ‘. . . .4 .. , n ,3 .. L. ' ‘ -¢ 1 u" . . _. - “q :1, ... .. .. W“? ‘. "‘ n" I . ‘ . . .o 3.: :-- H; ' 1“ ' 1PM” 7‘. : I. l' .. ‘ . -. , .. - ‘ - an - ..-N v .. . : _1 159 A 1: r. ‘APPEEUIEX F GSCALE: COMPUTER PROGRAM USED TO MAKE ADJUSTMENTS IN PLAYBACK LEVEL _. _ - N,...,\‘ H , 5,-1 ".: UQLHLE.HHC 19-JUL-79 ram: ' TLF ~L4‘E ~-“C»¥~; \\‘if~ +? .‘riECV?:'-‘??3}+Uui'\ 4F.ET[0- .§'5P£?T7 ' h: - z 5“ ; o" .Lo~ -KE:U:‘ aSTRULT c w R0:1 :5 HHGLWENT [I .NACPD SHZF‘Z AS? 52 FOR “3 {figm SHIFT: :HHCRO 3WDU c’ i! L I." i .7 ' F l- :1! L’ I: 7': "3' gym ?2’FO :" 3; ['3 L' :3! i5 3:1 I @901 STE 5801: SH F*u l-f HON w0.R3 HOV FL-fi3 SHIF’E DHCU €HFFT2 B 13 :7 [I ‘wo— SHLth J l :-7~"d :HmF‘h SHIFT: 160 ,— . _ r n ‘- .. .. .. ‘ ' U A. . ' ~. ) .. .. .. . ’ . v A s . - ...‘ ‘ - . . '4 ‘ x. ‘ . '1 ‘ " - 'fl .: D 1 3 .‘. , V . s - .. .— r ‘2. . . - .., - . , '. I \.' ~ ~ - . . ~ , ‘ . - - ..2 ’ - .4 .hn -' . in _CI -5. 3' L“ L“ H '7‘: 1-1 .‘11 If: '. ...- .-,-~ L'l‘ - .- . .. l .- ._ . .5? .... .3 [a -\ .2- .. ,- .1. I- 2 3 ‘.I 1‘} U] r;\ L— t”- '= U i' :- 5 ‘ .. ,- .; 5" r I: I t" {I "‘ L ."' “ I? “1 U: r. U; r.‘ P's £le .... --‘ . . .— —‘- .- ‘) ~v‘ \ ,' 'r '3 ‘5. A .- I.) T" -. v-"a !.. .-.- f 1”: ‘_ r" ...; ..'_ ...| ‘1 . .- TN. 9 .4- TC! V‘ \ i 'J ’-t.‘ '11 ,s‘.. .— .. 1“" a- Q 01". ‘\ ‘ v . 2* I . H'- : 1 —~ 0‘ H . I," 5‘. . ' .‘3 1. 5 a P 1-. -.. I ‘ l ‘ _( u |‘ '-- . L. L.‘ . F' c :L‘ I“. .. ‘5 - E'x -. w | n K ' . m ‘ I -- . w 1 I. ‘ .— .‘ n P‘ u u '1 ‘ z.- ,.‘ " , ' ‘ l-v .. q .. ~ J v . \_.' 2 C‘ 5 U .. u. l.‘ 161 U ‘i 7.. 162 '1‘ a I. do 7.. n ... . L w... J. . .. pH. rag. rul. .H... ..U wi- n“ a“ a“. P. a... a . E _ c 3. m... L... ...... .U. ht. ..R W... ..F. .U _. P .U — .... .... ..fi . R ”U ...‘w R ..I.. Z. V q . ..U Q Iv SI. .r f... O ...!- - P~ I ...—Om ..hl. V _lr_ :_ 9 ...—I. ... v «I. f. ... I ”H C a: u. “U D F. .-.1 3,. “v C .F. .1 U U .... P. .L 0. U. a. - . 1w. .. ... C 1. an” H. n. «V ..J a... C ... H... ..V .... .0. 1., w. .... .- : c H . a-.. J : .1 ..... 1.. ...- .... ...x ...H 1:. ...- ....- 2‘. ...w H. ...... .-.- :x. ...» m... S, S.“ .....W .U. ...- ... ...t ;\ .-. ..H ...4 C ...- ..... Q .n. .H. .-n .... .... .-. ...- T m. .1 a... ...: ..H 171...? .5 ...... L -L .-.; 3 7.x. L I. w: 2.. ..H -L L 3 :. 2.. L L r3 n9 L i. r... C; Sr. L L C.” .I. L 3 2-.. R .... L ..D L L r3 .5 ..i F .9. r ..-. .0 H. U. « ... .-... .\.. -V ..! .5. 5 Q U; .1 C. 2.7.1:. my _. 5 5 Q U T- 3 n... .v T 3 :5 C H1 .... ...»... U T- .... 3 1|. U ...... n.. h. v :5 .3 . . ...h I. ..4 ...A 3 ..... ._ a, Q. ...n. .....u . \H w...“ 3.. ..1 v. ..H 5. a. Q... -_ \m J... .5 .H \H ..R .H. ...n -... .m R to G ...... .L Q. I. A. _H D. «m 2. ..D ..h .-. .Ju t. to A. no O9 OP .- ...s. 0‘. . . I. m . 3 d ..-. 4.. 7, 3 9 n .. ...-. ...... 0.. ...n. 1:. ...o. D ,..: ..C ...L a... .... U... .5 «.... ...C .3 .... ...L a... .h. “I. a...» .U ..U u. . n. BUS 163 --.- ---,.n ..- .( :- -2 ‘-u;: Uaud;:,Wwb .Q’JLu*'. §5L g: aw: In | s a o. o ‘ ‘-‘ ans QUED 1‘ * " " s‘ .1 I D ." .1 ?_'5’ .59 *CoCEBS E';- 1.; Q :(_ :9 1 \ IO!- <“4 UULN \2L *1 .1 w I‘- ‘. L. '\ '.' Qtl'“ “'2'.” i: Q r '\ _-_; .'~ “.3 .- .SF! 363 ;§ mi» #7777?;a~ fiDv WL”777 :(l RVS PC a: new $1“ 777~ao «07 #:77777aRL -.» < --. " 6‘. r:} I. L! c - \ ' F ~. " " JuPF :«fiL: D°E'Q 3389 £803 BBQ? 3606 0355 USD4 0393 GED: LIL“ [9 .1. .“4RLE: 03« L' til". L Dab: DBU3 ”BUQ JEDi QBFfi " “.I "G rhd/ E959 DEM? DBb‘O iENdI -‘QF M. , . .LMFLJ' LNEHM' .FINW~ HH?H¢#YEE~¢ENEUF’L r' 1 ~ ' 3 .- ¢ $ MQU *‘1;f;5F-a 3 30 NE :0» * . ~HRiHr$P3t¢ZNHUF-l 164 m . . .. .. . .. 3 " '2' I‘ ‘-‘ ‘ v.“ ‘ - ‘ — "‘. 5‘ ”A-.. .33.. ”L... . I 11""4 .A. ‘1 Jul... ~.. “ .~‘ .. ‘ LU» - , . . ~¢~ - _ v ‘wo --. * ‘._' T ; , . -‘ '7'. LN. 14"" 2‘ ..v . ,.. . untFLw '_ V .-‘ ‘ "‘ yDI. - "--\, .--.‘u. J.‘ .- , «r "‘: a q _. L c ~ -. ‘— "1- 5:“! -.'1_ _‘..,.. "1"!" ,.-. -;-. ”,.., I “Ii... O F'JC. : 'Fl 7C: '-|‘ “.—’ ’ .... .. ".5“. -... 4 u v Wu. . - '— . V-» VHiL “P-’~ w .- .- w . r..-'r'.'\‘ 1* '5‘r -— _fl ny- a . - IR .‘ ... '..:' '- 1 - :1 ”\'_..F § 'vt'Jt:' I it: . ._ - L ‘4‘ ”.NEI . " .L w. 9 3' ‘l "4 - .J . I... ,._.. I. y“.>'\:_. “W1": ... . ., :3... . .2...“ 2‘, -_.“. .. C... 3' ”a. ‘.- ' 5""? ‘ i 1. ‘1‘ .‘ .~ _'\-I'~c"" - hl— A H.) --— . ---v. an ‘ . .v .. I ’ . > (a- . .' :4. " ._ “.14“? I '\ ¢\'I\ _”U w..- ‘ ‘ "‘ 7 ‘ ‘ " . ."- T' "‘ ‘ '~ 1‘ .. _=.. . L 4%-?! f” H: .'.- . :Lv‘W'I- E : . . .. q». - — I . J t 1‘;_ _ 3') .-.- .... ., .. .. ,.- ""“'-'T .4- ». -... 2'1 ‘ . 3 ~ .. .“ . ':::..:4.. . . first-» .2; : ‘ ’9' u. -_ 9 .. ...)! '~l‘ “:5 Ll‘ ..IHL -... .. . . ‘4 v {- ' 1" U -— . — _,I L' “".':' 9 '\‘.TV‘ ..._ .‘_.,._‘. ,— , "f- ‘ 4 . . A .~L.,.. L r 9‘4.” : 23:1 ;."J I .' - . c4 ' . E -‘ .:7‘37.‘ ‘~“" .7" ‘. J -\ ‘. "L. :'-r ’04». 4.1. 7‘1.KA| ’37- JL'UJ .... - L‘.‘ ' ,‘ 9: - . , ... ... . ‘:: ;:- D ~ .. :~ 2.1.- ... .. ' -.. ,... . :j . III. ‘ 's '- r-v - ..4:L,-.-:. . - .- .- .. '. .~ . Arr-l. 1.1. ' .' v'a ' ... ; 'r‘ ' '. "4H- ) ‘fik.-l\h:" \31. I .. .. ._ .. ":1: I . ““2 [Z f'" '3 ‘ 1 .\-,' " ““1. "I a": -- ”+32. 3.... .‘HJ' .. ...]. .. - . . \ -- In: .:"1‘ ... ".“h' -..- ._ z . ‘ . - , . - \ . I § lo 1 , .‘J .1 21‘ A . A" 5" . I" . _J 'si‘Lfi- '|$:-l....“ 1.... ..--.- - ._ . ‘ l‘ . ’ ; ': r LL: .- wth‘ L! J ..-..--. ...-r ‘3; :O b 3“" .’ --' : In." ' “an" ~‘..":'-..n‘« ‘I ”..--'L‘ . n} ‘- 2".— '3 ”(‘0’ T' .3 ' . n I... O v4-9”. u .. . . ... .. .... -A... .‘ ~ - ..- Z "J L r "4 L15“. L! x.- - - .. - _. .-_- :: - -r. -. . 'm- r L! J - ..--.-” - . .:‘-» ... ", :17 '2 ‘. T ol".."".['\.”\} :1“, U 5"“; CE 3‘ b A. 165 1:-JUL-7a l' ;I: R“- 4:. IF' ..‘4 T 3- .3 (J i“ .. K H__ '31 ... V‘ x" I. L ("i E Q {I l ... -: I A . ..xl C.” U a - MI. ..w H. ru 1 W. --. 3.. C. . .... .w. .u, 4.. W... n... ..U u L LI : ...... .-. v ..S , . ,. M. .... ...... :_ .. 4 1L 3.... f .. u... . 5 _ 5.. h k 5‘ H c. .... w. E , . 0 ". ¢._. F. ..J :3! ‘3 9'. 3 4 “it? ~ ..3 \i ‘z I L- «O L. «D. .\ "\' 4 R T I; 5 ;‘i."( r3; -. ". ‘4 ~ I HUMBLA v 3, :- F. - . .l v ...o 1. .r- E v... ... -m U I... H. .u 0 r .... H. .x. .... -.. ...... ...... ..n .-.. at ...J t. 4v .wv J. ... ... --. ..- N. _ .mr ..n ...... E F‘ '41? ...»- ...... 2. “K m... n. ... M. T ...- R .U L U L n. ... .... 1... ...u U U .-. .V m... ... ..- ..- . . m.“ .1... r ... qD ..U .-.- M... ...m m... ..D .U n... -.. «I C ..- 19.. ... .U $1., .... Mu » 3.. ¢L 5 p :u. -.L. o ., ~ .....I. , ..1 C... Q. \H. n: M- «I. .L. 9:... . . Wu. C. M. ....— n n... ‘l' 74.31” 1| H m *m'¥ 'H i-1 ,i. I fig x}! 166 19--.*UL--.7S ~r -cq.¢o.#eu:e:%,2€.fixrsLt .‘l ‘- ov fi’rwvofi- 1'55 I. .LZ'. .-..--i ... rah'b4 .4 I 1.. I ...- , I .9.” ...?z. -> .J 4‘ ‘ .. .4 34""JF33E5' ,- i3. '5 ,- MK 73!...\ I". IV .-1 ‘ .:é. H ‘C 'J HDMBLK : A ., ~. .. -. .. ’1. .- \ . .- :u IL) :7; 15:0“: ‘0- .> '< .\ ‘u\ -.- ._ . .- ‘ u ' g.-- ' ‘42)"? .:vax ,"A' r I .. 4‘. :'- .“x z.-::. *w n! -‘ ----.- 'n ,, ..., '"r-H'fiv ¢1_~ 7..“ 'r...eN'~-L‘...‘ :.' l I ~:.- ., _ " .\;‘V’ ‘7; .\ L_.4o n , _I__Ir' , ,vr- *;... '— an. .m- . x; \l APPENDIX G FREQUENCY RESPONSE MEASUREMENTS ON EQUIPMENT USED TO PREPARE AND PRESENT SPEECH STIMULI The frequency responses of filtered speech stimuli are often subject not only to intentional filtering, but to unintentional filtering as well. Unintentional filtering may be imposed by all instruments used in stimulus preparation and presentation. The purpose here is to describe the apparatus and procedures used to obtain response curves on (1) the Ampex AG-SOO recorder, (2) the Nakamichi 700 II recorder, (3) a computer system, and (4) the Grason—Stadler 162 speech audiometer with TDH-49 ear- phones mounted with standard cushions (MX 4l/AR). For measurement pur- poses, the upper frequency limit was arbitrarily set at 10 kHz for all systems. Tape Recorders Figure G-l describes the apparatus used to measure the frequency re- sponse of each of the two tape recorders. The input to the tape recorders was produced by a Wavetek 185 sine generator and was monitored by a Ballantine 5500 B frequency counter. The output of the recorders was mon- itored byaiBruel and Kjaer 2607 measuring amplifier. Initially, the sine generator was adjusted to emit a 1 kHz signal and the Nakamichi recorder was adjusted to zero VU and 580 mv output. Subsequently, readings were made from the measuring amplifier in dB relative to an arbitrary reference. A similar procedure was used with the Ampex recorder, but the Ampex re— corder had a 1.5 v output when adjusted for zero VU. The response curves obtained in this manner are shown in Figure G—2. 167 168 Sine Tape Generator I Recorder ...... I I I I _;_1 Frequency Counter Measuring ' Amplifier Figure G-l. Apparatus used for measuring the frequency response of the tape recorders. 169 .sz tmuuoumu HH ecu «goasmxmz «so can A