‘ V 13 ‘1'” H31 3 ' ‘l. ‘ . v 1 3.1.13 111103“ ' . ' ~ 11W 111;: .. . ~ .- .fl-p 4‘ o ' ”—fi- - -o-sw _ . .. - :7: ,J.--‘—_-‘ , w ‘; 5-2," "I’m-<31. u’t— . . -.._. . , afwo‘,‘:‘{.t— . ~ .a.--‘.“ -4 ' " _. ’ , ‘ ,. u o 0-. .4... _.. ‘ . out" I . um. 7;,,‘.-.... ' u D f“ . . . _ - .31....5'p. - £3151..- - . . -352: w. Wu .4 '1 - .._ .. ~ - e. .. T... ‘J-F h. J. v”- ..-.¢ “ —- ¢ ”293-: x“. ’7‘“ ‘33-! ” ,. . ..Eg”.~_,.. u . '1..- E. é'flw £313-);1—vi‘: ”1“ _. - ' .-‘.Z.“ .p.-w-— ,.. .. ' , .. ~. .. ...-w—,"11:P:’."WE,3-F$ ”3"”; u ’ ’ - I, C:— v ‘ " 1 , wfifia “‘0’ \AJ v - ,_ I. 3" ...~ . "7",» M jing- Ad 5:4: :‘0 -‘. ,_n L... 7"1‘ "" 0’... _;:-:' . v 3 I"; "' “15:1 ~‘: ’ 3' .3 33~3 ' ' .1. 1 - .1 . 1'1}: -. 14111 1‘ 4.. iv ' 13‘ui "1111 ' 11111111 11111M ,1 1; . -. = ‘ '3 1.1::1-1. a: 21,13,512: - ”2h?" : 1 31111‘1m1’11‘1‘111111‘1 ‘1‘! 11111113131111“ I ;: Q1171" 11;: 51; 13.1“ r1. “1 3'1 31’111..‘§:13: 1133 '* “$1525 5:... 11', 5' 1.1.1.1113" {11111 - .. 1.: 33133131111111." 113111M11'1- -~.“' 1111' ' .10" 13:1 1 1211111111111: 1121-1 ”13' 1'1: 3* . ' " ' 3"" [ARI }3 33333133333333'13'33‘1'35 3333 {3233115733 ‘3‘3‘3113'3'31'3' .11 11111; 1,111.1: 1 '31" 12 111.1%), 1111; :~- 2 “3 ” 11' 1.1111 11:11:, 1111’ "'21" W1 1. 1m 1, ,1 1 ”'1 :1: 1,1113" .1: 1:11:11. 13‘ ”“11 1333313113" 1'": 2111111133-1~"' 1' "11:1 * ' 1'1 111.1 a. 1. "3'3"“ 3"3'31'3'33 3"" ' 3"" '3 "3" 3 373333 B" 33"“ '1.“ 's"l:' 111' 1411137"3'3"31"' “5:11,“ 33,333 3333*?“ $11133"; )1'3 3'3: 3'3 '5‘3133‘1'3 ‘ 33331 533"}? . .'3;‘-;1~113-52335:' H 33333:]; '1" 3313.311 33"1131111'3 33333 1. '3: 3, ffz’c‘ili. 33:3" 1;“;11013‘“ 3:11;;91 31:33 1; 1"" ... 11:; "3'13! '3""'""""'3"3\ 1113 "1'1" 3131-113“ 11:1. .33 $2 12 11.171211}; . .1 1; v ‘ : *1 WM :11 ' 1313.; : ‘ ' ; w . 1.1.3111. ' " 13333311191 ”'3 "" " 33333131331111] 13 11‘ 3333333333333 33333112333333??? '33: “313353.333" . {1313 113; 1:111 ' 311111., 13' 33311 '1" [1 31311 111113 1 11111; H 11:11 3134131133313 : 3:31'111111a"1'3231;3;‘gif =131113“*11 *‘wér ‘1 1 '1’1‘1‘“‘ 11111111311“M111m11 ' ' ""'"1" . ' M?" :13" 1’“,1,;.;“1..1::1: 133:. '53-'31: “3333'. 3'3"' M"' "' 3" 3'331' 333" ;"3'11""'1“" " "1111333 '3133331‘31‘wg33'3"""1'133t3"'- 33' 313.13" '3' 13:1. '11"; 1313'3333'1J11‘331 1116'11-4231133 :1 In "M 3'3 ""33 " 3'3 “’11 31, 411 :1113'2. 133 “111133 11.3. "1%;331‘1111131} 3.333133 33',“ 113;,1‘111; 11111133413” . "1‘1 1 1 '11: c. 2131:1111 311“" 1111 1111111"'3'11H111"' "1'" '3 ‘11:" "11 '1'11‘3'1 311:. 43113111 11311121111111 '1 77 THESIS 77777777777 7 7 7 77777 7777777 777 777 1411 This is to certify that the dissertation entitled THE EFFECTS OF CDMPEI‘ING MESSAGE SEMANI‘IC CONTENT OF 881 PERFORMANCE presented by Michael Stewart of the requirements for Ph.D. degree in Audiology & Speech Science 7 has been accepted towards fulfillment 7 7 WWD/m Major professor Ma 16 1983 Date y ’ MSU is an Affirmative Action/Equal Opportunity Institution 0-12771 HIE-Fat?“— w \ 7 but-.37". ? ' ;i 43.1 '. «ya-wash“ “ W11??? . ‘ .-. 3.5 a J .4 _'-.'I .2“... a 1% 7 ; " ’ _§ . ' ,fi . ‘1 ..- a. ‘o- ’3 -..,.=;- J. 6G. ‘ a"?‘ .‘3‘ . ,' ‘9’ .9 .- ‘ “.‘l ; s 1.1 , .',~ .7‘ ~ .' Day‘s“ - 13.... x 7 . .. ‘ . 4. CW'Y' k’ .1. .' '. ‘ X', {I ~ ‘.; I. ._ ”’5‘ .— “7 5.3.3L Q . .‘1 it .' .w‘ic a \ .i.v.., . «Io-WW w'*'- ‘ - MSU LIBRARIES -— \— RETURNING MATERIALS: Piace in book drop to remove this checkout from your record. FINES wili be charged if book is returned after the date stamped below. $7 3:864:33, Jim/x ' - " . i 1 .1111! w!“ J. THE EFFECTS OF C(IVIPE'I'ING MESSAGE SEMANTIC C(N'IENT ON SSI PERFORVIANCE by Nfichael Stewart Submitted to Michigan State University in partial fulfillment of the requirements fer the degree of D(X,‘IOR OF PHILOSOPHY 1983 H.9- H.254» ABSTRACT The Effect of Carpeting Message Semantic Content on SSI Performance Scores lIhirty—six normal-hearing adult subjects served in an experiment that examined the effects of carpeting message semantic caitent on SSI psychometric function slope values and 50% masking efficiency levels. To avoid the confounding variable of temporal discontinuity inherent in a single-talker carpeting message, the study utilized multi-talker stimuli and stimuli derived from the multi-talker materials. The effect of carpeting message semantic content was investigated by comparing SSI performance under six different carpeting message conditions. 'Im competing messages allowed examination of semantic content effect in a single-ground carpetitor while four others allowed examination of semantic content effect in foreground- background stimuli. 'Ihe six carpeting messages were as follows: (1) multi-talker, cmsisting of eight equally prominent talkers (MI' 8); (2) MI' 8 in background plus one perceptually praninent foregrotmd talker (MI' 8 + 1); (3) dynamic envelope noise derived from the MT 8 (DEN MT); (4) DEN MI‘ in background plus single-talker in foreground (DEN MI‘ + 1); (5) MI‘ 8 in background with modulated single-talker noise in foreground (MI‘ 8 + DEN 1); and (6) DEN MI‘ in background with DEN 1 in foreground (DENMI‘ + DEN l). A 25-item four-alternative forced-choice respaise paradigm was used to reduce test variability associated with small test size. The present findings indicate competing message semantic content had little, if any, effect on psychometric function slope value. However, carpeting message semantic content did serve to increase masking efficiency. Semantic cmtent in a single-ground stimulus caused 2.6 dB excess masking. Semantic content in foreground-background competitors caused excess masking of 1.0 dB when it occurred in the temporally variable foreground only, 2. 8 dB when it occurred in the temporally continuous background only, and 8.9 dB when it occurred in both grounds. ACKNGNIEIISEMENTS With deepest appreciation to Michael Chial for providing an ideal for professionalism, precision, and hard work; with @epest affection to my father for instilling in me the value of education; and to my wife for living with both me and my dissertation for 4 years . TABLE CHAPTER II. IlSTOFTABlES. . . . . LISTCFFIGURES . . . . LIST OF APPENDICES . . . BACKGROUND. . . . . . . Introduction. . . . . . ProblemStatement . . . ExperimentalQLestions. METHOD. . . . . . . . . Subjects........ Stimulus Materials . . . PrimaryMessage. . RecordingSynthetic SelectionofTarget Construction of SSI Re-recording the Target Synthetic Sentences OF C(l‘lTEN'IS Sentences Synthetic Test List Sentencesm... Generation of the Experimental Tapes . Carpeting Messages Recording of Carpeting Speech Message . . . RemiXing the Single-Talkerand Multi—-Ta1ker Pacaflbratiai of ' '5in Levels Analysis of Speech Material . . . . . ......... Page l6 19 21 21 21 22 24 24 26 26 28 29 32 33 35 37 CHAPTER METHOD (continued) III. LongTermSpectraSimulation............ Apparatus for Pbdulating the Speech Spectrum Generation of the DEN Carpeting Message . . . . . . . Recording Calibration Tones for DEN Stimuli . . . . . Speech) Generation of Combination Stimuli (DEN + Real ExperimentalApparatus........ ..... . ExperimentalProcedures .. ........... RESUDTS . . . . . . . . . . . . . . . . . . . . . Introduction................... Statistical Procedures . . . . . . ........ Reliability Procedures for SSI Data . . . . . Odd—Even Reliability .............. Test-Retest Reliability . . . . . . . . StandardErroroftheMean.... .. . AnalysisofSSIData ............ Description........ ........ . Reliability of Measurement . . . . . . . . . Odd-Even Reliability ............ . Bjmnj-al TESt o o o o o 000000000000 Test-Retest Reliability . . . . ...... Standard Error of the Mean . . . . ....... Page 43 43 45 47 48 49 50 55 55 58 58 59 59 59 59 60 74 74 83 84 CHAPTER ' Page RESULTS (continued) InferentialAnalysis..........V...... 92 Psycharetric Function Slope Values . . . . .7 . 92 Fifty Percent Masking Efficiency levels . . . . 101 IV. DISCUSSION.......................115 Introduction......................115 ExperimentalFindings..................ll6 Reliability . . . . . . . . . . . . . . . . . . . . 116 Odd-EvenandtheBinanialTest 116 Test-RetestReliability............ 116 StandardErroroftheMean .......... 118 Reliability Indices of Other SSI Studies . . . . . . 119 Psychometric Emotion Slope Values . . . . . . . . . 120 Fifty-Percent Masking Efficiency Levels . . . . . . 123 Single-Ground Semantic Content . . . . . . . . 124 Semantic Content in Foreground-Background ders . O C O O O O O O O O O O O O 129 The Effect of Changing Foreground Semantic GaiterIt 8mm 0 O C O O O O O O O O O O O 1'34 ClinicalImplications..................136 ImplicatiatsforFutureResearoh 138 V. ‘SII‘MRYANDOCNCLUSIONS.................141 Intmductim......................141 mg 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 142 CHAPTER SUNMARY(mntide) ExperimentalDesign..... Subjects ....... Stimuli........ syntheticSentences CompetingMessages Procedure ......... Findings Cmclusions ........ APPENDICES............ was 0 O O O O O O O O O O O Page , 142 , 142 , 142 , 142 , 143 . 144 145 148 . 150 - 196 Table 1. LIST OF 'I'ABIES Page Position of semantic cattent within the carpeting message (background vs . foreground). + = present instimulus, -=absentinstimulus, O=nostimulus . . . . 18 than, standard deviation, and range values of synthetic sentence inter-stimulus—intervals (expressedinseconds)................... 30 than, standard deviation, and range values of stimuli level re 1 KHz calibration tone for List D and three sentences frameachoftheeightlists . . . . . . . . . . . 31 Mean percent-correct scores , standardized deviations , and ranges of Group I subjects for competing messages MI‘8+DEN1andDENMI‘+1asaf1mctionofmessage— to-carpetitionratio(N=12) ............... 61 than percent-correct scores , standard deviation and ranges of Group II subjects for competing messages MI‘ 8 and DEN MI' as a function of message-to—competition ratio(N=12)....................... 62 than percent-correct scores , standard deviations , and ranges of Group III subjects for competing messagesMI‘8+landDENMI‘+lasafmcl-.ionof message-to-oarpetitionratio (N=12). . . . . . . . . . . . 63 than psychometric function slope values, standard deviations , and ranges of the three experimental groups for their respective carpeting message conditions . than psychometric function slope values were determined using linear regression equations oomputedforeachsubject 72 than estimated 50% masking efficiency, standard deviations , and ranges of the three experimental groups for their respective carpeting message conditions. These levels were estimated fran linear regression equations computed for each subject . . . . . . . 75 Table (Cont. ) 9. 10. 11. 12. 13. 14. 15. 16. Intra-list reliability correlation coefficients of three experimental groups for their respective carpeting message conditions as a function of message-to—competition ratio. . . . . . . . . . . . . "Average" intra-list reliability correlation coefficients of three experimental groups across message-to—carpetitiai ratios for their respective carpetingmessageccnditions............. Percent-correct of odd and even items for the six carpeting messages as a function of message-to- carpetition ratio. Percentages are rounded off to thenearestwholenumber............... Z values for the six competing message conditions as a function of message-to—competiticn ratio. Equivalency of odd and even item correct rate is assumed if Z value is less than -1.96 or greater than +1.96.. . . . . . . . . . . . . . . . . . . . . . Test-retest correlation coefficients (Pearson r) for six carpeting messages. Test-retest measurements were obtained at the message-to—carpetition ratio which produced the nearest to 50%—correct score for eachsubject..................... Test-retest correlation coefficients for the six carpeting message conditions if data are eliminated fran subjects whose secald scores were 24% greater than the first. Test-retest measurements were obtained at the message-to—carpetition ratio which produced the nearest to 50%-correct scores foreachsubject................... Standard error of the mean (SE'M) for six carpeting messages as a function of message-to-carpetition ratio (N = 12) . Also sham is mean SEM across message-to-carpetitionratios..... ... ... . . Che-way within-subject analysis of variance of psycharetric function slope values for competing messagesm'BandDENtfl‘ Page 77 82‘ 85 86 87 89 91 93 Table 17 . 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. (Cont.) (he-way between-subject analysis of variance of psycharetric function slope values for carpeting messagesMT8+DENlandDENMT+DEN1 . . . . . . Che-way between-subject analysis of variance of psychometric functiar slope values for carpeting messagesDENMT+landDENm+DENl . . . . . . (he-way within-subject analysis of variance of psycharetric function slope values for carpeting messagesMTB+landDENtfl'+DENl. . . . . . . . Ore-way within-subject analysis of variance of psycharetric functiar slope values for carpeting messagesMI'8+DENlandDENMI‘-i-l . . . . . . . . (he-way between-subject analysis of variance of psycharetric function slope values for carpeting messagesMI'8,M.['8+l, andMI‘8+DENl (GrOUPN=12)................... Ore-way between-subject analysis of variance of psychometric functiar slope values for competing messagesDENMI‘, DENMI‘+1, andDENtfl+DEN1 . . One-way within-subject analysis of variance of 50% masking efficiency levels for carpeting messagesMI'BandDENMI' ............. (he-way between-subject analysis of variance of 50% masking efficiency levels for carpeting messagesMI'8+DENlandDENMI‘+DENl . . . . . . One-way between-subject analysis of variance of 50% masking efficiency levels for carpeting messagesDENMI‘+landDENt/IP+DEN1 . . . . . . Che-way within-subject analysis of variance of 50% masking efficiency levels for carpeting messages m 8 + 1 arld m MI‘ + Dm l O I O O O O O O O O O O (he-way within-subject analysis of variance of 50% masking efficiency levels for carpeting messages I“ 8 + Dm l and DEN m. + 1 O C O O O O O O O O O O Page . 102 . 103 . 104 , 105 , 106 Table (Cont. ) Page 28. Que-way between-subject analysis of variance of 50% masking efficiency levels for carpeting messagesMI‘8,MT8+l,andMI‘8+DENl. . . . . . . . 107 29. (he-my between-subject analysis of variance of 50% masking efficiency levels for carpeting messages DENMI‘, DENMT+1, andDENMP+DENl . . . . . 110 30. Tukey's test of specific carparisons of mean 50% masking efficiency levels for carpeting messages MT 8' MI‘ 8 + l, arld m. 8 + Dm l O O O O O O O O O O O 0 ll]. 31. Tukey's test of specific comparisons of mean 50% masking efficiency levels for carpeting messages DENtfl',DENtfl‘+l,andDENMI‘+DEN1 ......... 112 LIST OF FIGURES Figure 1. Block diagram of equipment used to.remix-stimuli. . . . . 2. Block diagram of equiprent used to perform and record narrow/bandanalysis . 3. Block diagram of equiprent used to perform and record l/30ctavebandanalysis 4.1mg-termspectrumofMI‘8... ..... 5. Iong-termspectnlmofMT8+l 6. Iong-termspectramofsingle—talker. . . . . . . . . . . 7. MajorcomponentsofCD-l‘................ 8 . Block diagram of equipment used to generate DEN stimuli . 9. Experimentalapparatus 10. Block diagram of experimental events for SesSiar I . . . 11. than percent-correct scores for carpeting messages MT 8 + DEN l as a frmction of message-to—competition ratio. Each datum represents observations of 12 normal-hearing adult subjects tested monaurally. Vertical lines denote i- 1 standard deviatiar. . . . . . . 12 . than percent-correct scores for carpeting message DEN MT + l as a function of message-to-carpetition ratio. Each datum represents observations of 12 normal-hearing adult subjects tested monaurally. Vertical lines denote : l standarddeviation .. 13. . than percent-correct scores for competing message MI' 8 as a functiar of message-to—carpetition ratio. Each datum represents observations of 12 normal-hearing adult subjects tested monaurally. Vertical lines denote 1- l standarddeviation.......... ....... .. Page 34 38 38 40 41 42 44 46 51 52 65 66 67 Figure (cont. ) Page 14. than percent-correct scores for carpeting message DEN MI' as a function of message-to—carpetition ratio. Each datum represents observations of 12 normal-hearing adult subjects tested monaurally. Vertical lines denote + l standarddeviation -68 15. than percent-correct scores for competing message MP 8 + 1 as a function of message-to-carpetition ratio. Each datum represents observations of 12 normal-hearing adult subjects tested monaurally. Vertical lines denote + l standarddeviation .—69 16. than percent-correct scores for competing message DEN MI‘ + DEN l as a function of message-to—carpetition ratio. Each datum represents observations of 12 normal-hearing adult subjects tested monaurally. Vertical lines denote ilstandarddeviation...................70 17. than percent-correct scores for six carpeting message conditions as a function of message-to-carpetitiar ratio. Each datum represents the observations of 12 normal-hearing adult subjects tested monaurally . . . . . . 71 18. than psychometric slope values for the six competing message conditions. Psychometric function slope values were determined for each competing message condition using linear regression equations carputed for each 19. than 50% masking efficiency levels for the six carpeting message conditions. Fifty percent masking efficiency levels were estimated for each carpeting message condition from linear regression equations carputedforeachsubject.................76 20. Intra—list reliability correlatiar coefficients for competingmessagesMI‘ 8 +DEN land DENMI‘+ 1 as a functial of message-to—carpetition ratio. Each datum is based on 12 srbjects. Significant correlations exceeded .497 at the .10 probability level . . . . . . . . 78 21. Intra-list reliability correlation coefficients for competing messages MI‘ 8 and DEN MI‘ as a function of message-to-ccmpetition ratio. Each datum is based on 12 subjects. Significant correlations exceeded .497atthe.10probabilitylevel . . . . . . . . . . . . . 79 Figure Page 22. 23. 24. Intra-list reliability correlation coefficients for carpeting messages’MI‘ 8 + l and DEN MT 4— DEN l as a function of message-to-carpetition ratio. Each datum is based on 12 subjects. Significant correlatiars exceeded .497 at the .10 probability level . . . . . . . . . 80 Test-retest correlation coefficients for six carpeting messages. Test-retest measurements were obtained at the message-to-carpetition ratio which produced the nearest to 50% correct score for each subject 88 Test-retest correlation coefficients for six competing messages if data are eliminated from subjects whose second scores were 24% greater than the first. Test— retest measurements were obtained at the message—to- carpetition ratio which produced the nearest to 50%- cor'rectsccreforeachsubject...............90 LIST (IF APPENDICES Appendix A.Subjectscreeningform............. B.SSItestlists................. C.Informedconsentform.............. D. Standardizedinstructions............ E.SSIpracticetest................ F.Deterrminationcf-tCRs.............. G.SSItestprotocol................ H. Percent-correct scores of the 12 normal-hearing subjects in each group as a function of carpeting message and message-to-carpetitcr ratio . . . . . Page 151 153 175 177 179 183 191 193 CHAP'IERI BACKGROUND INTRODUCTION The Synthetic Sentence Identificatiar Test (SSI) was devised to systematically examine the role of temporal processing in speech intelligibility (Speaks and Jerger, 1965) . Single unit speech stimuli, including nonsense syllables, phonetically balanced moncsyllables, and disyllabic words are of insufficient duration for use in studies involving the parameter of time. Calversational sentences are undesirable because: (1) real sentences may convey meaning or be identifiable by only one or two key words; and (2) message set equivalency is difficult to achieve due to the difficulty of con- trolling the factors of word familiarity, sentence length, word length, and syntactic structures (Speaks and Jerger, 1965). The construction of artificial sentences according to specific pre- determined rules avoids the problems inherent in single word or real sentence stimuli. Speaks and Jerger (1965) employed Miller's (1951) method of generating artifical (synthetic) sentences in which each successive word is selected solely on the basis of the carditional probabilities of word sequence without regard to overall sentence meaning. In other words, the selection of each new word is determined by the preceding word or words in one of three ways or orders of 2 approximation to real sentences . Synthetic sentences can be constructed by: (1) choosing successive words fram a camon pool (first-order sentences); (2) choosing successive words based on a preceding word (second—order sentences), and (3) choosing successive words based on the prewding pair of words (third-order sentences). All words are selected fram a pool of the 1000 most common words in the English language (Thorndike and Large, 1944) . Sentence length (allowable number of words and syllables) is pre-determined. 'Ihird- order sentences are characterized by greater syntactical constraints than either first-or secaid-order sentences and thus more closely approximate real sentences. The majority of research involving the Synthetic Sentence Identification Test (SSI) has focused on third- order sentence stimuli. The SSI test as described by Speaks and Jerger (1965) and Jerger (1970a) examines an individual's ability to identify a syntactically caistrained sentence I. given a ten-alternative closed message set. Identificatiar of synthetic sentential stirmlli in the ten-alternative forced-choice paradigm is a departure from traditional test tasks which require repetition of verbal stimuli. The paradigm allows for the determination of leaming or practice effects and minimizes experimenter error through the utilization of autaratic data collection procedures (Speaks and Jerger, 1965) . The experimental procedure is similar to that of traditional discrimination testing (Fletcher and Steinberg, 1929; Miller, 1947; Egan, 1948) in that intelligibility is determined as a function of signal intensity (i.e., psycharetric function). Speaks, Jerger, and Jerger: (1966) found the general shape of the resulting function to be sigmoidal, with performance scores rising steeply from 20% to 80% correct within a 6 dB range (Speaks, 1967a) . 'Ihis slope reflects the relative simplicity of the identification task in quiet and subsequently fails to differentiate between varying abilities to discriminate speech (Jerger, Speaks, and Trammell, 1968). A flatter psycharetric function slope would be required to differentiate among srbject groups. Several researchers have modified the SSI test procedures in an attempt to flatten the slope of the psycharetric function for the purpose of making a more clinically useful test instrument. Speaks, Jerger, and Jerger (1966) used low pass filtered synthetic sentences (FC = 500 Hz) as stimuli and found no appreciable effect er the shape or slope of the resulting psycharetric function. Speaks (1967b) erployed low-pass filtered stimuli (FC = 125 Hz) and high-pass filtered senterce stimuli (PC = 7000 Hz) . 'Ihe psycharetric function again remained unchanged in shape and slope. Other studies degraded the stimuli via broadband masking noise (Speaks and Karmen, 1967; Dirks and Wilsm, 1969) , also with no significant effect on the shape and slope of the psycharetric function. In these studies, the psycharetric function is observed to shift along the abscissa to a degree proportional to either the amount of filtering or the magnitude of the signal-to-noise ratio of the ipsilateral masking noise. Al- though both the filtering process and the addition of masking noise elevate the sound pressure level necessary for subjects to begin to be able to identify stimuli, once this point is reached, the psycho- metric function rises quickly. 4 The SSI task also has been degraded through introduction of a continuous speech carpeting message (Speaks, Karmen, and Beritez, 1967; Jerger, Speaks, and Trammell, 1968; Dirks and Wilson, 1969; Dirks and Bower, 1969; Garstecki and Mulack, 1974; Goldschmidt, 1979). Speaks, Karmen, and Benitez (1967) were first to discover that the steepness of the SSI psycharetric function could be considerably flattered by the introduction of a single-talker carpeting message. Inthis case, thesamespeakerwasusedforboththecarpetingand the primary (SSI) message. The flattening of the psychometric function was attributed to "random masking" (i.e. , masking carparents occurring randcrmly in time) and the disruptive features of carpeting speech. Whereas electronically—generated broadband noise was a very efficient masker in terms of shifting the psycharetric flmction almg the abscissa, the single-talker carpeting message was efficient in flattening the function along the ordinate . Speaks , Karrren, and Benitez- (1967) confirmed the stability of the SSI sentence task with a constant message-to-competition ratio (MIR) over a wide range of presentatiar levels. 'Ihus, test difficulty was increased with selection of various presertation levels without the problems of level-dependency inherent to the quiet presentation (no carpeting message) situation. Four studies (Dirks and Bower, 1969; Garstecki and Mulac, 1974; 'I‘ramrell and Speaks, 1970; Golet, 1979) investigated the possible effects of the disruptive features of serantic content (meaning) within the carpeting speech message. Dirks and Bower eliminated the effect of semantic content in a single-talker carpeting message by reproducing it backwards. No difference in SSI 5 performance was found between the competing message forward (CME‘) and carpeting message backward (CMB) conditions at t/CRs of -30 dB, -24 dB, -18 dB, and -12 dB while holding the primary message at a constant 40 dB SPL. Dirks and Bower concluded that the serantic content of carpeting message stimuli does not affect SSI performance. Three other studies found otherwise . Goldschmidt ( 19 79) used a single-talker carpeting message and observed decreased SSI performance witthdE‘ascarparedtoCt’lBatt’CRsof—lZdBand-M dB. 'Iheprimary message was delivered at a constant 40 dB SPL and the MIR was varied. Garstecki and Mulac (1974) used a multi-talker competing message and alsoformdpoorerCMFthanO‘stcoreswiththeprimaryandcarpeting messages both delivered at 40 dB SPL (0 dB VCR) . 'Iheir carpeting message consisted of a carbination of two separate! group discussions (3-5 speakers per group). The spectrum of the multi—talker competing message used by Garstecki and Mulac almost certainly differed from the single-talker spectrum used by Dirks and Bower (1969) and by Gold- schmidt (1979) . Trammell and Speaks, (1970) , who also used a single- talker carpeting message, found poorer OVIF than (MB scores at MCRs of -30, -20, and -10 dB. Other methodological differences which may account for the discrepancy in results include the use of different test stimuli, levels of subject sophisitication, response modes, and presentation modes . Dirks and Bower erployed monaural earphone presentations of ten sentences arranged in several randam orders with an ipsilateral carpeting message to traired listeners; Garstecki and Mulac used sound field presentation of mo different lists of synthetic sentences 6 to naive listeners; and Goldschmidt used monaural earphone presen~ tations of six different SSI lists (generated by Chial, 1978) with an ipsilateral carpeting message to naive listeners. Trammell and Speaks are not specific about their experimental calditions , noting only that the study paralleled those of Dirks and Bower in most respects. Dirks and Bower's listeners may have been overtrained in the primary message set, which may have obscured the serantic content effect. In contrast, Goldschmidt's subjects were trained in the SSI test paradigm but not with the primary message set. This approach is more suitable to the clinical situations for which the SSI test was designed. Another reason for conflicting results may be the high level of test variability due to the relatively small number of SSI test iters. 'Ihornton and Raffin (1978) questioned the use of small-sample tests such as the SSI, noting the increase in variability which occurs when a test is reduced from 25 to 10 iters. 'Ihey conclrmd that test variability is dependent a1 both the subject's true score and the number of items am the test. In support, Goldschmidt (1979) found normal subject SSI test ranges as large as 100% and standard deviatims nearly one-half the mean values under certain conditions. Speaks and Karmen (1967) also found large standard deviations on the SSI task in quiet and in some noise conditions with normal-hearing subjects . Although Trammell and Speaks (1970) do not report standard deviations, they did note considerable variability by which the (NB scores exceeded CMF scores. Dispersion data are conspicuously absent in all other SSI studies. If the standard deviations in other studies are also large, interpretation would be difficult and clinical application impossible. The presentation paradigm employed in most studies involving the SSI presents an additional problem. The forced-choice method is used with a set of ten alternatives. Unlike other forced-choice methods, the iters are successively drawn fram the set ultil each one has been presented. Thus, the 'apriori probability that a given item will be presented varies with the presentatiar nurber: for the first presertation, each item has the same probability ( .10); for the last presentation, one item as a probability of 1.0 and the others have a probability of .0. Variation in the ‘apriori probability of item presentation may interact with listener memory of prior presentations , as well as the listener's memory of past responses and the carfidence with which those resparses are made to increase variance in test scores . Another potential source of test variability which could create discrepancies among SSI studies is the amount of temporal discontinuity in the carpeting message. Miller (1947) noted that although the long- term spectrum of a single voice is nearly optimal for masking speech, the spectrum at any given moment may be incarplete (void of necessary frequencies) or contain gaps of silence (nonmasked periods). Miller reported that multi-talker (four to eight voices) carpeting messages are more efficient than single-talker competing in masking PB words at various message-to-carpetition ratios . 'Ihe multi-talker carpeting message produced a steeper function similar to that observed with white noise; Miller was also able to flatten the psychometric furctim by changing the amplitude envelope of a continuous masking noise. Interestingly, the caldition erploying the 80%-on 20%-off 8 masking noise (broadband) produced a psycharetric frmction very much like the single-talker function. This finding indicates that temporal discontinuity alone, apart fram any semantic content effects is partially responsible for flattening the PB psychometric function. Even though Miller used PB words rather than SSI material, his results are relevant to SSI studies. Recent research involving single-talker carpeting message effects on SSI performance supports Miller's (1947) notion that at any given moment the single-talker masking spectrum may be void of can— parents and thus an imperfect masker. Wofford (1977) found that the intelligibility of a single key word in a synthetic sentence could allow identificatiar of the entire sentence. Based on this finding, Martin and Mussel (1979) predicted subject SSI performance under two conditions: (1) presentation of synthetic sentences with speech noise am to the single-talker carpeting message and (2) presentation of synthetic sentences with only the single-talker as the competing message. In this study, a subject was given full credit (10%) for a sentence if one or more key words were identified. The assumption was that a subject could determine the correct sentence given one key word. 'Ihe mean P-.I function for predicted SSI performance was shifted and steepened for the noise condition (terporally continuous carpeting message) relative to the no-noise condition (terporally discontinuous carpeting message). In other words, the noise condition was more efficient in masking because it contained no terporal gaps. Several other studies investigating the intelligibility of speech using a variety of verbal materials and experimental procedures have relevance with respect to the issues of serantic content and 9 terporal discontinuity which are raised in the SSI studies (Carhart, Tillman, and Greetis, 1968; Carhart, Tillman, and Greetis, 1969; Hogan and Hanley, 1963; Pollack and Pickett, 1958; Carhart, Johnson, and Goodman, 1975; Young, Parker, and Carhart, 1975). Carhart, Tillman and Greetis (1968) found that modulated noise and camected sentences produced nearly identical masking effects whether de- termined by spmdee threshold shifts or by intelligibility scores for monosyllabic words . Interestingly, they discovered that carbining these two maskers produced an increase in masking efficiency for spondees and shifted monosyllabic word intelligibility functions to a degree not attributable to a simple summation of acoustical power (4.8 dB shift for the spardees and 7.5 dB for the moncsyllabic stimuli). In other words, masking was presert in excess of that which was phySically presert. The researchers suggested that "serantic influences,“ manifested during the simultaneous presentation of the two stimuli may have caused the excess masking. The term "perceptual masking" was later applied to the observed increased masking efficiency secondary to serantic content in the composite backgmmd of sentences and modulated noise (Carhart, Tillman, and Greetis, 1969). Perceptual masking is the preferred term in that it denotes mly the existence of a phenarenon, whereas the term serantic content implies an under- lying mechanism responsible for the phenomenon. Carhart, Tillman, and Greetis (1969) investigated whether speech intelligibility (as measured by the threshold for spaldee words) was more adversely affected by stimuli containing a meaningful speech message than by meaningless modulated noise with carparable dynamic envelope characteristics. ‘Ihe researchers noted that modulation of a 10 masker, whether the normal modulation of the speech envelope or artificial modulation produced by interrupting white noise, should provide the listener with acoustic "windows" capable of reducing masking efficiency. Indeed, the masking efficiency of both the single-talker carpeting message and white noise modulated four times per secard to a depth of 10 dB with a 50% duty cycle (N2) were reduced approximately 4 dB carpared to the unmodulated white noise (N1) masking condition. However, they also discovered that a carposite backgrourd of carpeting speech (single-talker) and modulated white noise ((112 ) was a more efficient masker than either white noise modulated four times per second by 10 dB with a 75% duty cycle (N3) or unmodulated white noise (N1) . The N3 cmdition in the Carhart experiment was selected to approximate the dynamic envelope characteristics of CN2 while reraining free of any serantic cartent. Carparisar of N3 and CM2 performance allowed direct assessment of the impact of perceptual masking. Carhart and his colleagues found that masking by the acoustically similar but meaningless stimuli (that is, N3) was approidmately 3 dB less efficient than masking in the CNZ condition. The CN2 cardition was also superior in masking efficiency to the N1 (white noise) condition by 2.2 dB even though the CN3 stimuli cartained acoustic "windows" or spaces of silence not present in the N1 stimuli. These comparisons suggest that the perceptual masking carparent (serantic content) of the CN3 stimuli increases masking efficiency more than acoustic windows decreaSe masking efficiency. Carhart, Tillman, and Greetis. (1969) also investigated perceptual masking in the binaural situation by examining nine types 11 of background noise across three experimental conditions (homophasis, antiphasis, and time-delay presentation). The most interesting finding with respect to this discusSion is the comparison of single- talker plus unmodulated noise carposite stimuli and the two-talker stimuli. The latter was found to be a more efficient masker across all three presentation conditions. Thus, the two-talker complex must contribute more perceptual masking than it does masking efficiency (due to temporal gaps). These findings support the contention that within the single- talker competing stimuli, the two factors of semantic content and tempcua1.discontinuity influence speeCh intelligibility scores in contradictory rather than complementary ways. That is, semantic content serves to increase perceptual masking efficiency, while temporal discontinuity creates a perceptual advantage which serves to decrease maSking efficiency. Several other speech intelligibility studies add support to the findings reported in the Carhart studies (Miller, 1974; Polladk and PiCkett, 1958; Hogan and Hanley, 1963; Carhart, Johnson, and Goodman, 1975; Speaks, Wigginton, and Germono, 1971; YOung, Parker, and.Carhart, 1975). Polladk and Pickett (1958) and Miller (1947) both report an increase in masking efficiency when a second single-talker competing message is added to an existing single-talker competing message. According to Carhart, Tillman, and Greetis (1968), part of this increase in masking can be attributed to the reduction of the acoustic "window” effect; however, a substantial increase is due to perceptual masking (5.5 to 7.5 dB). The manifestation of perceptual masking is also apparent in a study by Hogan and Hanley (1953). They averaged 12 intelligibility scores (pharetically balanced monosyllables) for competing voice nurber over several conditions of rate (carpressed and expanded) and singal-to-noise ratio and found increased masking efficiency with voice nurber increase (from two to four to eight voices) . Recently, Carhart, Johnson, and Goodman (1975) carpared masked spondee thresholds obtained in speech spectrum noise modulated by seven different talker carbinations, (l, 2, 3, 16, 32, 64, and 128 voices speaking cmtinrous discourse), with those obtained in the respective associated talker carbinations. The long term spectra of all stimuli (14) were equalized. This procedure (coupled with the modulatiOn) produced noise maskers which were nearly identical physically to their respective associated talker stimuli, yet were void of serantic content. The findings indicate perceptual masking occurred in all seven carbinations of talkers as carpared to their respective modulated noise derivatives . The talker situations produced excess masking on the order of 6.2 dB for the are-talker situation, 7.2 dB for the two-talker condition, and 9.8 dB for the three-talker condition. Masking efficiency decreased thereafter with excess masking stabilizing at 3 dB with the 64 talkers. Results of the study clearly indicate that, given identical masking spectra, speech stimuli are more efficient masker than modulated noise. The speech modulated noise has also been used with SSI stimulus. Speaks, Wigginton, and Germono (1971) carpared SSI threshold levels (tER corresponding to 50% correct) obtained using a single- talker carpeting message and white noise modulated by the single-talker as the carpeting messages. The modulated white noise had approximately 13 22% more area under the envelope function than the carpeting speech mesSage (i.e., more physical masking ability). SSI thresholds were -29 dB t’CR for the carpeting speech and —36 dB MCR for the noise condition. The results indicate speech to be a more effective masker than the speech modulated white noise. These results are oarmensurate with the studies of ‘ Carhart and associates noted above. It is important to point out that neither Carhart, Johnson, and Goodman (1975) nor Speaks, Wigginton, and Germono (1971) determined the entire psycharetric function. Therefore, it is unknown whether the perceptual masking ability of ‘ speech stimuli affects the slope of the psycharetric function relative to a physically identical noise stimuli. It is evidert that the thresholds (50% correct) of various verbal material (monosyllabic, spondees, synthetic sentences) are more adversely shifted by the carpeting speech stimuli; but no data are available at the 25% or 75% correct response levels. It could be that the perceptual masking ability of speeCh stimuli causes a negative shift in performance and also changes the steepness of the slope. In other words, speech stimuli may shift and steepen or flatten the psycharetric function relative to an acoustically similar noise caudition. In sumary, the SSI psycharetric function can be flattered by introducing an ipsilateral single-talker competing message. Most of the flattering effect is probably related to the dynamic amplitude envelope and variable spectrum of the single-talker carpeting message (i.e.,~ gaps of silence or nomasking, or low-level speech components). The factor of serantic content may produce perceptual masking which increaSes~ masking efficiency relative to a physically identical noise l4 candition void of semantic content. The increased masking efficiency apparently serves to shift the entire psycharetric function. It is not kncrm what effect (if any) serantic content has on the slope of the psycharetric function. Recent studies have contributed to an understanding of the impact of terporal discantinuity of a carpeting message upan SSI performance, but the issue of the effect of serantic content is less well resolved. It is likely that terporal dis- continuity effect, differences in the spectra of carpeting messages, and test method have contributed to variance in performance, thus obscuring effects attributable solely to serantic content. Also, research has not examined the effect of serantic content within a foreground-background carpeting message on SSI performance . It is possible that acoustic stimuli having speech in the background and noise in the foreground or noise in the backgrormd and speech in the foreground more adversely affect SSI performance than a stimulus with noise in both grounds. In other words, it may be that a fore- gromd-background stimulus with serantic cantent in at least one ground exhibits more perceptual masking ability than a foreground- background stimulus void of semantic content in both grounds. Further- more, the expected semantic cartent effect may be even greater when serantic content occurs in both grounds of a foreground-background carpeting message. In order to accurately measure the difference in perceptual masking ability of the respective stimuli it would be necessary to make the stimuli physically identical to each other with the only difference being the presence or absence of serantic content.- Use of multi-talker speech or physically identical noise spectra as the 15 background stimuli across all competitive conditions eliminates temporal discontinuity and allows isolation of the variable of semantic content. The background-foreground paradigm provides a certain amotmt of face validity because many real world listening environments contain both background and foreground stimuli . One reason Carhart and his associates were able to successfully investigate the perceptual masking ability inherent in a speech signal was their specification of noise stimuli that had temporal and spectral characteristics similar to the associated speech stimuli. 'Ihe spectral characteristics of the broadband noise approximated tlat of the long term spectrum of the associated speech signal. The temporal characteristics of the speech and noise stimuli were made similar by using the dynamic envelope of the speech to modulate the approximated speech spectrum noise. Although backward running speech also exhibits spectral and amplitude characteristics similar to its foreward running counterpart and its use as a carpeting message probably reduces per- ceptual masking secondary to semantic content, it can be argued that such stimuli still contribute semantic content because quasi- intelligible speech carponents and human voice quality remain imbedded in the stimuli. Also, the resultant inflectional patterns are novel and could create attentional effects. 'Ihus speech modulated noise is a more desirable experimental stimulus . Speech modulated noise has been used in several other studies (Schroeder, 1968; Horii, House, and Hughes, 1971; Speaks, Wiggington, and Germono, 1971; Katz and Berry, 1971; Berry and Nerbcnne, 1971; Chial, 1974), and is especially applicable in the proposed study because (1) it contains no semantic content or verbal attentional l6 factors; (2) it can be consistently synchronized with its associated input speech signal (Chial, 1973); (3) it can be carbined with speech stimuli to produce a foreground-background condition; and (4) the physical spectra, temporal and amplitude cl'aracteristics of the speech modulated noise can be made identical to those of its associated input speech signals. ‘Ihe last attribute of the modulated noise stimuli allows equalization of the physical masking spectra of a particular noise stimuIUS and its associated speech signal input at any given point in time. PRfBIEM STATEMENT This Study sought to examine the effects of carpeting message serantic content on normal-hearing subject SSI performance while eliminating carpeting message terrporal discontinuity as a variable. For the purposes of this study, serrantic content is defined as that characteristic of a carpeting speech message which contributes varying degrees of meaning. It was expected that serrantic content would cause perceptual masking through interference with auditory procesSing of primary stimuli (synthetic sentences). Perceptual maskingisdefinedasanexcessinmaskingthatcccursduetotheeffect of semantic content. 'Ihe impact of semantic content was investigated in this study by carparing SSI performance scores for the following carpeting messages: (1) multi-talker carpeting message consisting of eight equally intense talkers (MI 8); (2) multi-talker carpeting message consisting of eight background talkers plus one perceptually prominent foreground talker (MT 8 + l); (3) dynamic envelope noise carpeting message derived from the MI‘ 8 carpeting message (DEN MI'); l7 (4) DEN MT in background plus single-talker in foreground (DEN MT + l); (5) MI‘ 8 in background with single-talker modulated noise in fore- ground (MI‘ 8 +DEN l); and (6) DENMl‘inbadrgrcrmdwith DEN l in foreground (DEN MI' + DEN l). The six stimuli allow examination of semantic content effect (perceptual masking) in single-ground and foreground-background carpeting messages . Table 1 reveals the position of serantic content within the respective carpeting messages. The most desirable method of assessing SSI performance under the various carpeting message conditions would be to present all six carpeting messages to the same group of subjects. This controls for several confounding variables and allows data analyses using an efficient and powerful one-way analysis of variance .. on repeated measures. However, this experimental paradigm was not possible in the present study. Preliminary testing which involved presentation of synthetic sentences in the presence of six different carpeting messages at four presentation levels for each carpeting message (6 x 4 = 24 experimental conditions), indicated a learning effect in regards to the synthetic sentence test items . That is , several subjects were able to correctly identify target synthetic sentences with little or, in sare cases , no auditory information. In most instances, the learning effect manifested itself saretime after the 12th experimental condition had been presented. This situation obviously restricted the number of presentations that could be given to a particular subject. 'Ihe impact of carpeting message semantic content was investigated using an experimental design which included several one-way within- subjects and one-way between-subjects analyses of variance in the data 18 Table 1. Position of Semantic Content within the competing message (background vs. foreground) + = present in stimuli, - = absent in stimuli, 0 = no stimuli. Background Foreground Semantic Content Semantic Content MI‘8+1 + + MP8 + 0 DENMI‘ - 0 DENMI‘+1 — + MTB+DENl + - DENMT-i-DENI - '- l9 analysis. Three different groups were used to reduce the number of test presentations. Each group was exposed to two different carpeting messages. Psycharetric function slope values and 50% masking efficiency levels were determined from individual subject data. These dependent variables allowed direct carpariscn of the six different carpeting messages on SSI performance. A modified SSI response paradigm was used in this study to reduce variance arising from a small nurber of test items. The response paradigm is similar to that used by Greenberg (1975) . The ten—item message set was discarded in favor of a 25-item four- alternative forced-choice message set. According to Thornton and Raffin (1978) , the standard deviation for using a tel-word versus a 25-word list is reduced from 16% to 10% at the 50% correct response level. If the theoretical prediction is correct, use of a 50-item test produces an additional reduction in variance of 3% . This inprovement did not appear to justify the increased time and probable subject fatigue that would accarpany a 50-item test. Therefore, the subjects task was to correctly identify the stimulus sentence from a set of four alternatives in each 0f the 25 test presentations . The following experimental questions were asked: 1. Do mean monaural SSI psycharetric function slopes obtained with normal-hearing listeners differ as a function of the presence versus absence of serantic content in a single- ground ipsilateral carpeting message? 2. Do mean monaural SSI psycharetric function slopes obtained with normal-hearing listeners differ as a function of differences in a foreground-backgromd ipsilateral carpeting message containing serantic cartent in a. background only; b. foreground only , and 10. ll. 12. 20 c . backgrormrd and foreground? Do mean monaural SSI psycharetric function slopes differ as a function of serantic content positim and type in a foregrormd—background ipsilateral carpeting message? Do mean monaural SSI psycharetric function slopes obtained with normal-hearing listeners differ as a function of differences in foreground status in an ipsilateral carpeting message containing a speech (semantic content) background? Do mean monaural SSI psycharetric function slopes obtained with normal-hearing listeners differ as a function of differences in foreground status in an ipsilateral competing message containing a noise background? Do mean monaural SSI 50% masking efficiency levels obtained with normal-hearing listerers differ as a functiar of the presence versus absence of semantic content in a single- ground ipsilateral carpeting message? Do mean monaural SSI 50% efficiency levels obtained with normal-hearing listeners differ as a function of differences in a foreground-background ipsilateral carpeting message containing serantic cartent in a. background arly; b . foreground arly, and c . background and foreground? Do mean monaural SSI 50% masking efficiency levels differ as a function of smeantic content position and type in a foreground- background ipsilateral carpeting message? Do mean monaural SSI 50% masking efficiency levels obtained with normal-hearing listeners differ as a function of differences in foreground status in an ipsilateral competing message containing a speech (serantic content) background? Do mean monaural SSI 50% masking efficiency levels obtained with normal-hearing listeners differ as a function of differences in foreground status in an ipsilateral carpeting message containing a noise background? How consistent are subject resparses within an experimental condition? Are there differences in consistency across experimental carditions? How consistent are subject test-retest SSI performance scores (% correct) for each carpeting message condition? CHAPTER II METHOD SUBJECTS Thirty—six normal hearing srbjects were selected according to availability from a university student populatiar. Each subject was literate, had normal or corrected vision, and was naive with respect to SSI test material. Normal hearing was defined as pure tare hearing threshold levels (HTLs) at or better than 15 dB (ANSI, 83.6, 1969) for audiaretric test frequencies ranging from 250 to 6000 Hz. Spondee hearing threshold levels (obtained using the method proposed by Martin- Stauffer, 1975) were 15 dB or better. Subject speech discrimination scores were 90% or better on the Northwestern University Auditory Test No. 6 presented at a sensation level (SL) of 35 dB (re: the spondee threshold, Tillman and Carhart, 1966) . Tympanograms were Type A (Jerger, 19 70b) with acoustic reflexes present at normal sound pressure levels (70-100 dB according to Jepson, 1963) at 500, 1000, 2000 and 4000 Hz. All subjects had normal acoustic reflex decay at test frequencies 500 and 1000 Hz. See Appendix A for audiological screening form. STIMULUS MTEMAIS The study erploys two classes of auditory stimruli: primary message and carpeting message. The primary messages are third-order synthetic sentences. The competing messages include (1) real speech, 22 (2) speech modulated noise, and (3) speech modulated noise carbined with real speech. The following sections will describe the ‘ experimental use and discuss the derivation of both classes of stimuli. Primary Message Third-order synthetic sentences used in previous studies were found to be inadequate for use in the present experiment. The procedure used in generating these sentential stimuli may have inroduced experimenter bias in the structure of the synthetic sentences because the first two words for each sentence were chosen by the experimenter. Also, generally high variances associated with the most widely known version (Jerger's) suggest possible benefit fram increasing the number of test items within a given set and more vigorous control of linguistic, pharetic, and acoustic features of stimuli. Third-order synthetic sentences were generated using the following procedure. The Thorndike and Iorge (1944) Pool of the 1000 most common words were divided into 16 different lists. Each list was assigred an identification number. The experimenter randamly drew one word from each of mo randomly selected lists. The resulting pair formed the "seed" for a synthetic sentence. Three hundred such word- pairs were generated. These, in turn, were divided into thirty sets of ten word pairs each. Five differert sets were randomly assigned to each of six graduate students who volunteered to assist in the generation of sentertial stimuli. Each volunteer was assigned an idertification number (1-6) and a copy of the 16 lists derived from the Thorndike and Iorge pool. 23 Individuals were seated at separate work stations and instructed in the generation process. Each individual was asked to draw one Set of word-pairs fram their pool of five Sets.) For each of the 10 word- pairs on this initial set, the individual randomly selected one of the 16 carnon word lists and then chose a "third word" from the list that could possibly follow the original word—pair in a meaningful sentence. After each person chose 10 "third words", the first word of each three word sequence was masked and that set of words was exchanged with another individual. The secard individual again randomly selected 10 of the 16 canton word lists, chose an appropriate "third word" for each cf the ten word—pairs, masked the first word of the three word sequence for all ter items, and then randamly exchanged sets. This process continued until all 30 sets had ten synthetic sentences carprised of at least nine words each. The original word—pair used as a "seed" was then eliminated fram each synthetic sentence. This was done to minimize possible experimenter bias effects. 'me 300 synthetic sentences were then recorded on note cards and examined to verify that (1) each sentence cartained seven words, each of which must have come fram the Thorndike and Iorge pool. and (2) no sentence contained repetition of words. Synthetic sentences meeting these criteria (N = 300) were then reviewed by a panel of three judges. Each judge independently applied the following criteria to each sentence. (3) Each synthetic sertence must be between eight and ten syllables. (4) No obvious carplete sentence can east within a particular sequence of words. 24 (5) No synthetic sentence can contain a sequence of words that is sexually suggestive or scatological in tone. A total of 104 sentences were judged as acceptable by all three judges. These 104 synthetic sentences were then recorded according to the following procedure . Recording Synthetic Sentences The 104 synthetic sentences were recorded by an adult male speaker of General American Dialect. The speaker stood in a sound- treated chamber (I.A.C. Nbdel 402) and read the selected synthetic sentences into a boam-mounted microphone (Electrovoice RE-16) at a mouth-microphone distance of approximately nine inches . The speaker monitored his voice level via earphones and a rerotely mounted VU meter. The signal was routed to an audio mixer (Teac Model 2) the output of which Was passed to an electronic switch controlled by discrete logicond timing equiprent (COuJbourn) . The logic and timing system activated a set of cue lamps which informed the talker of recording status (ready, rehearse, record), controlled inter-stimulus interval (7.5 seconds) and activated the electronic switch during record intervals ((5 secards). The output of the switch Was routed to one input ctannel of a multitrack tape recorder (Teac 40-4) . Selection of Target Synthetic Sentences Ore goal of stimulus generation was to make sentences as similar to eaCh other as possible. Toward this end, each of the 104 recorded senterces was graphically recorded (B & K Type 2305) to facilitate measurement of amplitude and duration. The graphic level recorder was adjusted to a paper speed of 10 mmr/sec and a pen speed of 500 mmr/sec. 25 Sentence duration was estimated by measuring the distance between the points at which the tracing initially and terminally (departed from baseline. The range of sentence durations was from 2.1 to 3.4 seconds. The mean was 2.7 secands and the standard deviation was 0.3 seconds. Sentences were eliminated from further cansideration if measured duration differed from the mean by more than 0.1 seconds. Forty sentences met this criterion. Sentence amplitude also was measured. Here the goal was to assess the nurber of amplitude peaks (presumably related to vowel segrents) occurring relative to a reference level. The reference level was determined by adjusting a l KHz calibration tare to produce a convenient display (40 dB) an the graphic level recorder. The l KHz calibration tone had been previously recorded at a level which corresponded to fre- quent peaks of the target synthetic sentences (i.e., 0 on the VU meter). An average decibel value was deterrmined for each stimulus having between five and eight amplitude peaks above the reference line by summing the peak decibel values above the reference level and then dividing by the number of peaks above the reference level in that particular stimulus. Mean decibel, standard deviation, and range values were then carputed as being 1.6 dB, 0.5 dB, and 0.6 - 2.5 dB, respectively. Sentences were eliminated whose average decibel values were not within are standard deviation (0.5 dB).of the mean decibel value. A total of 27 sentences satisfied this criterion. TWO of these were arbitrarily discarded; the remaining 25 sentences were designated as experimental target sentences. The 77 sentences that met the first five criteria, but did not meet the final amplitude or duration criteria were retained for use as distractor sentences . 26 Construction of SSI Test list In order to maintain equal representatian of target sentences across the four alternatives (a, b, c, or d) for each 25 item test list it was necessary to systeratically assign target sentences to a particular alphabetical listing for each of the 25 iters within a test list. This was accarplished by randomly drawing coins labeled a, b, c, or d, in a consecutive manner. Four target sentences were designated for four successive test items according to the order that was drawn. Three distractor sentences were then carbined with each target sentence for each of the four test items according to the following criteria: (1) no two synthetic sentences in a given set of four can begin with the Same word or phonere, and (2) no mo synthetic sentences in a given set of four can end with the same word or phonere. If any of the three distractor sentences chosen for a specific test item caused the above criteria not to be met, it was returned to the pool and an alternative distractor sentence was chosen. This procedUre was repeated until one target sentence and three distractor sentences were chosen for eaCh of the 25 iters in all eight SSI test lists. See Appendix B for SSI test lists. Re-Recording the Target Synthetic Sentences Follanring canstructicn of the eight SSI test lists, it was possible to generate the eight experimental tapes. Each experimental tape replicates the order of target sentences contained in its corresponding test list. BefOre the eight experimental cassette tapes could be produced, it was necessary to record a submaster 27 reel-to—reel' tape that contained anly the 25 target synthetic sentences and a 1000 KHz calibration tone. This was accarplished using the following procedure. The stimuli fram the master SSI tape (104 synthetic sentences) were reproduced by one reel-to—reel tape player (Teac 2300-8) and routed to an audio mixer equipped with earphones with auditory monitoring of the stimuli. The output of the audio mixer was connected to a second reel-to-reel tape player (Teac 40-4) that was equipped with pause cantrol to record only selected stimuli . The first tape player was started and allowed to reproduce the entire 104 synthetic sentences, which were canstantly being monnitored auditorally. The pause control of the secand tape player was disengaged at the commencerent of an interStimulus interVal occurring in front of a target synthetic sentence and re-engaged immediately after the target synthetic sentence had been recor®d. This procedure was repeated Lmntil all 25 target synthetic sentences and their respective interstimulus intervals were recorded. The 25 target synthetic sentences an the newly recorded sub- master tape were then reproduced in turn and delivered to an audio mixer equipped with earphanes for auditory monitoring. Each time a synthetic sentence occurred on the tape, the tape recorder was stopped and a cue switch engaged. The tape was hand—turned to locate the beginning and end of a synthetic sentence. The sentence was then cut out of the subraster tape and inserted between two pieces of 70 inch leader tape. This procedure was repeated until all 25 target sentences were removed from the suaraster tape and connected by 28 leader tape. Seventy inch leader tape was used to insure against print-through during storage. Following insertion of the sentences between leader tape, all sentences were routed to a graphic level recorder (B & K 2305) and the amplitude measurerents discussed above were repeated. The 25 target synthetic sentences were found to have a mean decibel value of +2.38 dB re: the l KHz calibration tone with standard deviation and range values of 0.5 dB and 1.5 - 3.4 dB respectively. The calibration tone was subsequently re-recorded at a level 2 dB above the initial level. A recheck of amplitude revealed mean, standard deviation, and range values of +0.14 dB, 0.47 dB, and -0.5 - 1.0 dB respectively. Generation of the Experimental Tapes Eight different cassette tapes, each containing a specific order of synthetic sentences separated by interstimulus intervals of approximately 7. 5 seconds, were generated for experimental use according to the following procedure. The 25 target sentences and the 1000 Hz calibration tone were reproduced an a reel-to—reel tape player (Teac 40-4) routed through an audio miner and selectively delivered to a cassette tape player (JVC KD-lS) . The 1000 Hz calibratian tone was adjusted to 0 V0 an both the audio miner and cassette VU meters. The calibration tone was recorded at the begirming of each of the eight cassette tapes. The order in which the synthetic sentences were recorded on each cassette tape was dictated by the corresponding test list (a - h). Each synthetic senternce in a particular list was located an the reel-to—reel tape 29 recorder. The recorder was then rewound to the beginning of the 70 inch leader tape that preceded the target sentence. The tape was then hand-turned until the approximate leader tape timing mark was directly under the playback head to create the desired interstimulus interval (approximately 7.5 seconds). Both tape players were started simultaneously and allowed to run through the 7.5 second interstimulus interval and duration of the selected synthetic sentence (i.e. , approxi- mately 10 seconds). The cassette tape recorder was shut off immediately following the cessation of auditory stimuli (i.e. , end of target synthetic sentence). Each of the eight lists was then recorded an the graphic level recorder and the interstimulus intervals between the sentences were measured. One of the eight experimental tapes was selected (randomly) and graphic level measurerents of interstimulus interval durations and amplitude were made for each of the 25 target sentences. In addition, three sentences were randomly selected from the set of 25 targets. Graplnic level records were made of each of these three sentences as they occurred on each of the eight experimental tapes. Results of the interstimulus interval and amplitude measurerents are shown in Tables 2 and 3 respectively. These results suggest a high degree of success in attempts to produce consistent stimuli. Carpeting Messages The carpeting messages used in this study are real speech, speech modulated noise, and carbination stimuli (real speech plus speech modulated noise). The real smoh and speech modulated noise stimuli were devised to examine the amount of perceptual masking effect secondary to the semantic content inherent in the real speech while 30 Table 2. Mean, standard deviation, and range valms.,‘0f Synthetic sentence interstimulus intervals (expressed in seconds). .A B C D E F G H NEanlISl (in seconds) 7.96 7.84 7.7 7.8 7.8 7.7 7.7 7.7 Standard Deviation 0.15 0.12 0.14 0.43 0.12 0.13 0.13 0.15 Range 7.7- 7.5— 7.6- 7.6- 7.5— 7.5— 7.5- 7.4- 8 4 8.1 9.8 9.8 8.0 8.0 8.0 8.1 31 Table 3. Mean, standard deviation, and range values of stimuli level re 1 kHz calibration tone for List D and three sentences fromneaCh of eight lists. LIST D SENTENCE A. SENTENCE B SENTENCE C .Mean dB value (re: 1 kHz calibration tone) 0.56 0.85 0.58 0.48 Standard Deviation 0.5 0.36 0.21 0.16 Range -o.2— 0.5- 0.33- 0.37- 1.4 1.37 1.0 0.88 32 controlling for amplitude, spectral, and terporal factors. The carbinatian stimuli allows examinatian of the effect of serantic content position (foreground versus background) within a speech/ noise stimuli. Each of the carpeting messages will be discussed in the following sections . Recording of Carpeting Speech Messages TWo of the six experimental carpeting message situations consisted solely of speech material. These were the multi-talker stimulus (MI‘ 8) and the background multi-talker plus a perceptually prarrinent foreground talker stimulus (MI‘ 8 + l) . It was assumed that these two conditions would contain more serantic content and thus cause more perceptual masking than any of the other experimental canditions with the MI‘ 8 condition causing slightly more perceptual masking than the MI‘ 8 + l canditian. The MT 8 stimulus was, generated by Chial (1978) . Eight Gerneral American Dialect talkers (4 ferales, 4 males) read a passage from a news magazine for approximately five minutes. All talkers were free from disorders of speech and hearing. Each talker stood alone in an audiaretric test chamber and spoke into a beam-mounted microphone (Electrovoice RE—l6) , which was coupled through a miner (Teac Model 2) to a tape recorder (Nakamichi 700 II). The microphone- to~mouth distance was nine inches. Individual talker tapes were initially mined anto one of two tapes according to talker gender (second generation). The two tapes were then combined to produce a single 8-talker tape (third generation) .. Finally, the eight talker tape was re-recorded using splicing techniques to produce a fourth—generation tape of approximately 15 minutes in duration. Care 33 was taken during the dubbing procedures to maintain equivalent vowel peak levels on all talkers. Generation oftheMI' 8+ lstimuluswas accarplishedbymixing a single-talker tape (Chial, 1978) onto the existing MT tape. An adult male with a General American Dialect recorded the single-talker carpeting message in a "radio broadcast" style. The talker stood alone in a sound-treated chamber and read a passage published in the Reader's Digest into a dynamic microphone (Electrovoice RE-l6) which was connected to a cassette tape recorder (Nakamichi 700 II) . The taller monitored overall voice level via a remote VU meter and earphones. The single-talker competing message and the multi-talker carpeting message were remixed according to the procedure described in the following section. Remindng the Single Talker and Multi—Talker Stimuli Figure 1 presents a block diagram of the equipment used to remix the two submaster speech signals (multi-talker 8 and single- talker) to three experimental speech stimuli (multi-talker 8 alone, single-talker alone, and a carbination of multi-talker 8 and single-talker, i.e., MI' 8 + l) . Che recorder (Marantz Nbdel 5020) reproduced the single talker material and another (Nakamichi 700 II) reproduced the mined and dubbed MI‘ 8 signal. These signals were routed to separate input channels of an audio miner (Teac Model 2) equigped with a VU-metering system (Teac MB-20) . The preamplifier clnannels of the miner were adjusted to low-pass the signals at a cutoff frequency of 5 kHz, primarily to reduce residual tape noise. The miner was adjusted to produce different outputs at three of its EH2 one» “gum xenon Qu owns naming mo camera 2003 .H madame A ... .2320 A See __ .855 83 A _ .2520 ’ 5.2 Egssxez NIxe 55:80 xmim>m>> «:2 55:22 35 four output ports. . .One was a "pass-through" version of the single-talker (alone) signal. Another was a "pass-through" version of the multi-talker (alone) signal. The third was a carbination of the two. For this carbined signal, the average level of the MI‘ 8 carponent was reduced to approximately 10 dB below the average level of the single-talker carponent where both were monitored via VU meter. Each of the miner outputs was then routed to a separate channel of a third tape recorder, (Teac 40—4) . The gain controls of the playback machines, the miner, and the final recorder were adjusted to produce frequent peaks of -l W. Recalibration of Signal levels After the experimental stimuli were remined, but without changing any gain control settings, a level calibration tone (1000 kHz) was recorded an a separate tape segrent on each of the three program channels. The level of this tone was monitored via VU meter and was the same for each channel. It was anticipated (and later verified) that the amplitude peaks of the three stimuli would differ in level relative to the calibration tone. Thus, it was necessary to develop a procedure whereby the level of each calibration tane could be adjusted relative to the level of the speech signal associated with that tone. This was accarplished as follows. Each stimulus was reproduced in turn by the Teac recorder. The output of the recorder was routed through the mixer (Teac Model 2) to a frequency analyzer (B & K 2107) operated as a fastrms linear (20-20 kHz) voltmeter. The output of the voltmeter was graphically recorded (B & K 2305) . The level 36 recorder was adjusted to a pen speed of 500 mm/second and a paper speed of 10 urn/second.) For each of the three remined speech signals , the respective calibration tone and approximately five minutes of the signal was graphically recorded. Fifty successive amplitude peaks were identified at each of two locations (epochs) on the graphic record judged to be representative of the entire record for that stimulus. An amplitude peak was defined as a local maximum which exceeded the immediate adjacent signal level by 2 dB or more. A maximum amplitude was identified above which 10% (N = 5) of the peaks fell. This was done for both measurerent epochs, and an average level was carputed in decibels relative to the level of the cali- bration tone. These decibel values constituted a discrepancy in level between the calibration tone and the "lO%—peak" level of the stimulus. For the MI‘ 8 signal the mean discrepancy was +4.5 dB, for the single- talker signal +3.5 dB, and for the MI‘ 8 + 1 signal, 3.0 dB. Calibration signal levels were then modified as follows. The three previously recorded calibration tones were reproduced by a recorder (teac 40-4) and monitored on the miner VU meters. The reproduced calibration tone levels were set to -5 V0 on the miner. The tape recorder was stopped and a sine wave oscillator signal was routed through the miner to each of the three output channels. The level of this signal was adjusted independently for each channel to equal -5 W, plus the respective level discrepancy associated with the particular channel. These adjusted levels were then recorded on a new tape segrent which was subsequently spliced to the remined program tape . 37 The "10% peak" measurement procedure was repeated to verify the effect of the correction process. For each of the three speech stimuli (MP 8 alone, single-talker alone, and MT 8 + l) the discrepancy between the level of the adjusted calibration signal and the speech stimulus was less than 1 dB. Finally, stimuli were reproduced by the Teac reel-to-reel tape player, passed through the audio miner for level control, and routed to a cassette recorder (Marantz 5020). The MT 8 and MI‘ 8 + l stimulus tapes were used during the main experiment while the single-talker tapewasusedingeneratingtheDENMI'8+lexperimentaltape. Analysis of the Speecln Materials Before generation of the DEN stimuli, each speech stirmllus (MI' 8, MT‘ 8 + l, and single-talker) rmderwent a spectral analysis procedure to define its long-term spectrum. This spectral information was later used in defining the spectral characteristics of the associated DEN stimruli. Figures 2 and 3 describe the instrumentation used in the narrow-band and 1/3 octave spectral analysis procedures respectively. For the narrow-band analysis, each stimulus was reproduced from a different channel of a tape recorder (Teac 40-4) . The output of the tape player was adjusted until a 1000 Hz calibration toneproducedareadingofOVU. The signalwas thenroutedtoan audio miner (Tieac Nbdel 2), which was also adjusted to yield a reading of 0 VU. The weighting network of the miner was not engaged. Next, the stimulus was passed to a narrow-band analyzer (B & K 2031) . The input sensitivity of the analyzer was adjusted to produce a convenient 38 5:05 >.x v. e m manhood econ @2300 m) pom Escudo B own: ”Engaged mo amputee £0on .m wudafim 520.1 >.x v. w m ..mcm $680 $2 at... a: x e m A as...» T ease maniacs osmosiog Booms pom shampoo 8 own: £55an mo Emhomwo £0on .N 85on A 2< A $.52 .32 x e m same case 39 overall level deflection and to an arbitrary reference level which also produced a convenient deflection. This was done to allow measurement of the entire dynamic range of the stimulus. A linear average of the 256 samples was obtained at 200 m/sec averaging interval, weighted by a hanning window. The total time elapsed to average 256 samples was approximately 45 seconds. It had earlier been determined that a 45 second averaging time was needed to obtain a steady-state spectrum (one unaffected by duration or number of samples) . The average long-team spectra for each stimulus was successively displayed on the analyzer screen and later transferred to an X—Y plotter (B & K 2308) for permanent record. Following the narrow-band analysis procedure, the stimuli were analyzed using a 1/3 octave band analyzer (B & K 2131) . The DEN stimuli were generated using the spectral information obtained with the 1/3 octave band analyzer. Each stimulus was reproduced using a 4-channel tape player (Teac 40-4) , which routed the signal to an audio mixer (Teac Nbdel 2) and then to the 1/3 octave band analyzer. The tape player and audio miner were adjusted to produce a 0 VU reading when the 1 kHz calibration tone was introduced. The 1/3 octave band analyzer was then activated and its sensitivity adjusted until the 1 kHz calibration tone equalled 90 dB (relative). A 64 second averaging interval was selected to approximate the 45 second averaging interval in the narrow-band analysis. The stimuli were then passed in turn to the analyzer and the spectra of each was successively displayed on the analyzer screen. The spectra were later transferred to an X-Y plotter (B & K 2308) for permanent record (see Figures 4, 5, and 6). 40 m 92 mo gooam EmerocQH .v apnoea ANIV >02m30mmu w v N 2.3: can one mwp mse ago J u d d 4 d d a 1 u 1 1 . d d d a u d‘ in 4 d a q d u or cm on ow on (so eAueraJ) AlISNBlNI 41 a + m a: mo egomom lameness .m 8an are 55:85 a v .o. it: com com map . mm m. 5 I ‘ - d a q u d J u 4 e d u q q u a q q - or cm on 0v cm (89 eAmereJ) AlISNBlNI uofimelmaocam mo Epomam Sarasota .w 93on ANIV >02m30wmu 42 op ON on ov on (an alumni») AlISNBlNI 43 Long Term Spectra Simulation After spectral analysis of the speech materials, it was possible to generate spectra very similar to the long-term spectra of the particular speech stimulus. This was accarplished by generating white noise (Coulbourn S-81-02) and routing it through an audio miner (Teac Model 2) , which delivered it to the input port of a multifilter (General Radio 1925). The multifilter was preset to shape the input noise spectra in accordance with spectral specifications that had been obtained by the previously described analysis procedure. Apparatus for Modulating the Speech Spectrum Noise After the spectral characteristics of each speech stirmllus had been identified and simulated, it was possible to modulate the specific speech spectra by the desired speedn stimulus. This was aconmplished using an instrument designed by (hial (1973) . This device has the capability of accepting a broad-band noise input signal (in this case, specified speech noise) and multiplying it by the amplitude envelope of a second input signal (speech). The amplitude envelope is extracted by means of full-wave rectification and subsequent filtering (low-pass at 20 Hz, see Figure 7). According to Horri, House, and Hughes (1971) most of the amplitude variation in a speech waveform occurs at fre- quencies below 20 Hz. The output of the device (designated CD-l) is the analog product of the instantaneous amplitude of the signal presented to the noise port and the instantaneous amplitude of the waveform envelope extracted from the input speech signal. Changes in the amplitude of the specifically shaped speech spectra were dictated by changes in the amplitude of the corresponding speech input signal across time. A 30 mn/sec lag-time (in this case, 44 Too mo 35:09.00 no.9: .n 0553 \ w _ son .6528 v 225 _ SdSo J outcome SdrSo _r ..m r 1 defined $2.252! Biz. r :> \ wremue_ _m as: A $5.“. mmrcbmm r assume 85.33 5.2. \ 45 DEN following speech stimuli) secondary to the filtering process has been identified (Horri, House,and Hughes, 1971; Clnial, 1973) but is not relevant in the present study. Generation of DEN Stimuli Carpeting Message The goals of the study also require generation of stimuli which retain the amplitude, terporal, and spectral characteristics of specified speech stimuli (Nfl' 8, single-talker, and MT 8 + 1), but were void of serantic content. This was accarplished by modulating a specified speech spectrum noise by the associated speech signal (either MT 8, single-talker, or MI‘ 8 + l) . Each speech modulated noise stimrulus (DEN) exhibits acoustical characteristics of speech spectrum noise that has amplitude fluctuations and terporal patterns identical to its real speech stimulus counterpart. Subjectively, the DEN MI‘ sounds like steady state speech noise with slight fluctuations in loudness; the DEN 1 sounds like speech noise that has fluctuations in loudnness and noticeable pauses between consecutive noise segrents (i.e., words); and the DEN MT + DEN l is characterized by steady state speech noise in the backgrournd with the DEN l in the foreground (-10 dB foreground/background ratio) . Figure 8 depicts the instrurrentation used in generating the DEN stimuli. Three different speech stimuli (single-talker, MT 8, and MI' 8 + l) were reproduced by a tape player (Teac 40-4) and routed in turn through an audio mixer (Teac Model 2) to an amplifier (Coulbourn S-82-24) and finally to the speech input channel of the CD-l. The noise input channel of the CD-l received the associated simulated long-term spectrum produced by the white noise generator (Coulbourn S—81-02) and multi-filter. The input preamplifiers of the CD—l were 46 3.89.50 no.0: 0:22, Each-:00 eds—Em Zmo sunshade 8 com: ”egg .u—o amputee #003 .m 953m ' ‘ £22 . Noumea: e r s r— , /\ mflozi toxmoaw AlllJK‘ comlv 383580 :3on v. 5950 , .Edz _ e > — Saw to 9: to 9: _ 5:09:00 _ .omaémo 5:09:00 _ _ came _ a at). coo... 47 adjusted to levels previously determined to produce the greatest dynamic range (about 45 dB) in the input signal. The amplitude modulated output signal was recorded by means of a cassette recorder (Marantz Model 5020) . The two input signals and the output signal were monitored with an amplifier-loudspeaker system and an oscilloscope .' Recording Calibration Tones for DEN Stimuli Following the initial recording of the DEN stimulus, each stimulus was reproduced by a cassette tape player (Marantz Nbdel 5020) through an audio mixer and routed to a multi-track recorder . (Teac 40-4) . The level of each DEN stimulus was Set to produce frequent peaks of approximately 0 V0. A 1 kHz calibration tone, 60 seconds in duration, was recorded in front of each DEN stimulus on the reel-to—reel tape. The calibration tone was generated by a wavetex oscillator (Model 182) and routed througln the audio mixer before being passed to the Teac tape player. The level of each calibration tone was adjusted to approximate the VU deflection of their associated DEN stimulus(0 VU) . All three DEN stimuli and their associated calibration tones were then passed in turn from the tape player through the audio miner and a frequency analyzer operating as an RIVB voltmeter (B & K 2107) to a ‘ graphic level recorder (B & K 2305) for visual inspection and carputation of the respective levels. An adjustment in calibration tone levels was made for each stimulus using the "10% peak" method discussed previously. Finally, the stimuli and their associated calibration tones were passed from the reel-to—reel recorder through the audio mixer 48 (calibration tone set at 0 VU) and recorded on a cassette tape recorder (Marantz 5020). The DEN MI‘ and the DEN MI' + l tapes mere used as experimental tapes, while the DEN 1 tape was used in generation of the MI‘ 8 + DEN 1 enperimentnl tape. Generation of Carbination Stimuli (DEN + Real Speech) Another goal of this study is to compare perceptual masking effects of two temporally continuous stimuli opposing each other in background versus foreground serantic content position. The DEN MI‘ + 1 andMI'8+DEletimuli allowthis corparison. TheDENMl‘+l stimulus contains serantic content in the foreground (+10 dB S/N) , but not in the background. The MT 8 + DEN l exhibits serantic content in the background (-10 dB S/N) , but not in the foreground. The two stimuli are identical in amplitude, terporal, and spectral character- istics and each contains serantic content. Thus, it is reasoned that any difference in perceptual masking efficiency must be related to the positioning of serantic content within the background-foreground dichotomy. Subjectively, the DEN m + l stimulus is characterized by a single-talker being heard approximately 10 dB above a fairly steady- state speech noise. The MI‘ 8 + DEN l is characterized by multi—talker speech babble in the background and a louder (+10 dB) yet fluctuating speech noise in the foreground. The audibility of speech is greater intheDENMI'+lstimulus,butmorecontinrxnusintheMI‘8+DENl stirmllus. Generation of the carbination stimuli involved a rim-step procedure. First, the MT 8 and DEN l stimuli were reproduced by two 49 different cassette recorders (JVC KID-15 and Marantz 5020 respectively). These signals were routed to separate channels of an audiomixer (Teac Model 2) with the calibration tone of the DEN 1 adjusted to 0 W and the MT 8 calibration tone adjusted to -10 W. The stimuli were mined and delivered to one channel of a reel-to—reel recorder (Teac 40-4) . Second, the newly generated carbination stimulus (MI‘ 8 + DEN l) was reproduced by the reel-to-reel recorder, routed through the audio- miner, and delivered to the JVC cassette recorder. Generation of the DEN MI‘ + 1 followed the same procedure with the single-talker stimulus being adjusted to +10 dB re the DEN MP stimulus. After both carbination stimuli were recorded on cassette tapes, they were passed to the graphic level recorder to determine stimulus levels. Using the "10% peak" procedure discussed previously, it was determined that both stimulu were approximately +2.5 dB re. their respective calibration tones . An adj ustment in calibration tone level was made for each stimuli and subsequently verified. EDCPERII'MENIAL APPARATUS Primary message SSI stimuli were reproduced on one tape player (JVC 103-15) and the carpeting messages were reproduced onanother tape player (JVC KD-15) . The primary and carpeting messages were routed. to separate channels of a speech audiometer (Grason-Stadler Model 162) . The primary signal was also connected to logic devices which activated a cue lamp each time a target sentence was presented. A voltmeter connected to the output of the audioreter monitored output signal level delivered to a test earphone (TDH-49 with MX-4l/AR cushion) 50 mounted with a TC-89 headband. A dummy earphone and cushion was placed over the non-test ear. A block diagram of the experimental apparatus is shown in Figure 9. EXPERIMENTAL PKIZEDURES Following the informed consent (see Appendix C) , screening procedures, and group assignment subjects were seated in a sound- booth and read standardized instructions (see Appendix D) by the experimenter. The experimental events are sequenced in Figure 10. The initial portion of the experimental session was devoted to practice. Five non-experimental synthetic sentences were presented to the test ear at a 40 dB sensation level (SL) in quiet to familiarize the subject with the four-alternative forwd—choice paradigm. Eighteen other non-experimental synthetic sentences were presented at 40 dB SL with the carpeting stimuli adjusted to obtain a -2 dB signal-to-noise ratio to provide the subjects with nine listening experiences with each of two carpeting message conditions according to group assignment. Each of the three subject groups received one of the following pairs of competing message conditions: (1) DEN MI' and MI‘ 8 (2) MI'8+landDENMI'+DENl (3) MI‘8+DENlandDENMT+l The subjects task was to circle the correct item on the response sheet. Subjects had to score 100% on the practice test to participate in the main experiment. The practice test is displayed in Appendix E. Following the practice session, each subject received eight experimental conditions and two test-retest conditions during a one hour 51 aEmn 0:0 Bill .omommz nonsense m amassing .m momma 500m ocaow :25 2950 I _ 83o: — Boo... oammwoz 9:39:00 _ IE2 . _ o>e fio:<.zomoaw 862m .comeo - 5.2 333.2 caste. 533.2 52 H scammom MOM mango flamingo mo Suwanee £0on .3 mamas Sago 9: ~89 3:55 A smooom a mm; .OZ EoEcemma. __ v— 90th _ :26ng r EmE:oaxm it A) :3”. mo> tease «£620 } —-n C 0 Q 23 0‘) 02 £1 — L. e .230 J— 2820ng 225a t _ mmammmmi E 8:08.... _ __ Eomcoo BEBE. 23mm r 9.22500 5:5 mozomi 53 testing session. Each of the eight experimental conditions consisted of a 25—item test list presented at one of four MCRs for one of two competing messages. Test-retest data were collected at the MCR value which yielded the nearest to 50% correct score. The MCR values were -l4, -12, -10, and -8 dB for the MI‘ 8 and MI‘ 8 + l carpeting messages; -18, ~16, -l4, and -12 dB for the DEN MI' competing message; -20, -18, -16, and -14 dB for the MI‘ 8 + DEN l carpeting message; and -22, -20, -18, and -16 dB for the DEN MI‘ + l and DEN MT + DEN 1 carpeting messages. MCR values for the competing messages were determined in a pilot study (see Appendix F for pilot study). The synthetic sentences were presented at 40 dB SL and the SPL of the carpeting message was varied toobtainthedesiredMCR. Acuelampalertedthe subjecttothe presentation of each target sentence. The subjects task was to listen to each presentation and circle the correct item on the answer sheet. The orders of MCR and list presentation for a particular competing message were both randomized for each subject. The competing messages were presented in a counter-balanced order. Appendix G reveals the test protocol. During preliminary testing, it was noted that for sore carpeting message conditions a particular subject may have scored (1) entirely below 60%, (2) entirely above 40%, or (3) entirely between 44% and 56% across the MIR values. This phenarenon was probably related to the small 2 dB step size between adjacent IVER values coupled with the steep psychometric function slope, which together are extrerely sensitive to variations in subject performance. The problems this situation created were (1) artifactual slopes and (2) inability to carpute 50% masking efficiency levels. These problers were Solved by 54 implementing a rule that stated a given subject must score above 60% and below 40% for each Carpeting message condition. If this rule was not met with four IVER values, additional MIR values were presented. Performance at all IVER values were used in carputing psycharetric function slope and 50% masking efficiency levels. CHAPTER III RESULTS INTKlDUCI‘ICN This study was conducted to determine the effect of ipsilateral carpeting message serantic content on normal-hearing listener SSI performance. Thirty-six normal-hearing subjects were chosen from a university population and equally divided into three experimental groups. SSI performance was evaluated using two different carpeting messages for each of three experimental groups. The two carpeting messages for each experimental group were designed to investigate a particular aspect of serantic content effect. Group I subjects received carpeting messages designated MI‘ 8 + DEN l and DEN MI‘ + 1. These were similar in terporal, amplitude, and spectral chacter- istics, but differed in the position and type of semantic content within a foreground-background carpeting message . Group II subjects heardncarpeting messages MI' 8 and DEN MI‘. These carpetitors allowed examination of the effect of the presence versus absence of semantic content in a single-ground carpeting message while holding constant terporal, amplitude, and spectral characteristics. Group III subjects wereexposedtocarpetingmessagesMI'8+landDENMI‘+DEN1. Again, the carpetitors were similar in terporal, amplitude, and spectral characteristics ., but differed in the presence versus absence of semantic content in both grounds of a foreground-background carpeting 56 message. That is, the MI‘ 8 + l carpeting message had serantic content in botln foreground and background while the DEN MI' + DEN l carpeting message had no semantic content in either foreground or background. Each subject received an audiological screening to ensure bilaterally normal hearing, middle ear function, and acoustic reflenes . Tie initial part of the experimental session was devoted to practice. Each srbject listened to synthetic sentences presented in quiet and in the presence of their respective carpeting messages to familiarize them with the experimental task. Each subject then listened to 25- item test list presented at four erRs for both carpeting message con- ditions, (i.e., a total of eight experimental conditions). Psycharetric functions were plotted for each subject. The dependent variables of psycharetric function slope valres and 50% masking efficiency levels were derived from individual subject psycharetric functions for each carpeting message via the method of least squares. The 50% masking efficiency levels were interpolated from individual psyclnoretric functions and represent the message-to- carpetition ratio at which 50% correct performance should be achieved. Test—retest reliability information was collected for each carpeting message at the IVER value which produced the closest to 50% correct score. Retest data were gathered imrrediately following the experimental procedure. The following experimental questions were asked: 1 . Do mean monaural SSI psycharetric function slopes obtained with normal-hearing listeners differ as a function of the presence versus absence of serantic content in a single- ground ipsilateral carpeting message? 2. Do mean monaural SSI psycharetric function slopes obtained with normal-hearing listeners differ as a function of differences in a foreground-background ipsilateral carpeting 57 carpeting message containing serantic content in a . background only; b. foreground only, and c . background and foreground? Do mean monaural SSI psychometric function slopes differ as a function of semantic content position and type in a foreground-background ipsilateral carpeting message? Do mean monaural SSI psychometric function slopes obtained with normal-hearing listeners differ as a function of differences in foreground status in an ipsilateral com- peting message containing a speecln (semantic content) background? Do mean monaural SSI psycharetric function slopes obtained with nonral-hearing listeners differ as a function of differences in foreground status in an ipsilateral carpeting message containing a noise background? Do mean monaural SSI 50% masking efficiency levels obtaired with normal-hearing listeners differ as a function of the presence versus absence of serantic content in a single- ground ipsilateral carpeting message? Do mean monaural SSI 50% efficiency levels obtained with normal-hearing listeers differ as a function of differences in a foreground-background ipsilateral carpeting message containing semantic content in a. background only; b. foreground only , and c. background and foreground? Do mean monaural SSI 50% masking efficiency levels differ as a function of serantic content position and type in a foreground-background ipsilateral carpeting message? Do mean monaural SSI 50% masking efficiency levels obtained with normal-hearing listeners differ as a function of differences in foreground status in an ipsilateral corn- peting message containing a speech (serantic content) background? 58 10. Do mean monaural SSI 50% masking efficiency levels obtained with normal-hearing listeers differ as a function of differences in foreground status in an ipsilateral com- peting message containing a noise background? 11. How consistent are subject responses within an experimental condition? Are tiere differences in consistency across experimental conditions? 12. How consistent are subject test-retest SSI performance scores (% correct) for each carpeting message condition? DATA REDUCTION Subject performance was indexed by percent-correct scores at each IVER for each carpeting message (see Appendix H). Subjects who did not score either below 44% or above 56% using the four experi- . mental NCRs were tested at additional NCRs until this criterion was met. Psycharetric function slope valres and 50% masking efficiency levels were derived for each subject from their respective psycho- metric functions using a linear regression equation. These derived dependent variables allowed direct carparisan of the effect of the various carpeting messages on SSI performance. STATISTICAL PKX'IEDURES Group means , standard deviations , and ranges were carputed for each experimental condition (6 carpeting message conditions x 4 MCRs = 24 experimental conditions) . Mean psycharetric functions were plotted for each of the six carpeting message conditions. Group means, standard deviations , and ranges for I the derived dependent variables of psycharetric function slope values and 50% masking efficiency levels were then determined for each carpeting message condition. Reliability Procedures for 881 Data Reliability of measurerent was assessed through three statistical procedures . Consistency of response within adrministrations of the 59 SSI test was indexed by odd-even reliability carputations ; con- sistency across administrations was reflected by test-retest reliability computations; and consistency within experimental cornditions was indened via the standard error of the mean. Odd—Even 'mliability. The intra-list odd versus even reliability data were obtained by determining the percent-correct of odd-even test iters for each list at eadn IVER for all carpeting message conditions. These data estimate reliability within each test administration. Also, tlne Spearman-Brown formula was used to correct the odd-even coefficient for the lesser number of iters used in calculating the coefficient. Tast-Retest Reliability. 1 Test-retest data were obtained at the MIR value which yiel$d the closest to 50% correct score for each subject across both carpeting message conditions. These data permit i assess- ment of test stability (reliability) across successive administrations. Standard Error of the Mean. T’re standard error of tie mean indicates the "normalized" dispersion of a group of scores and is defined as: SD J N SEM= The resulting value estimates the amount of precision or degree of accuracy with which the SSI measurements can be made. A small sample standard deviation will yield a small standard error of the mean; in turn, a small standard error of the mean reflects good precision, minimal variability, and (presurably) good reliability. Analysis of SSI Data Analysis of both dependent variables (psycharetric function 60 slope values and 50% masking efficiency levels) for eacln of the three groups followed a one-way within-subject analysis of variance (ANOVA) with repeated measures on the two carpeting messages (Linton and Gallo, pp. 166-174, 1975) . Analysis of the respective dependent variables across subject groups was accarplished using one-way between-subject ANOVA's (Linton and Gallo, pp. 138-144, 1975) . Strength of association measures were carputed for ANOVA' s exhibiting significant F-ratios at the .05 level. The strength of association measures, ETA squared (n2 ) for the within-subject ANOVA's and one squared (002) for the between-subject ANOVA's, estimate the proportion of variance in the dependent variables accounted for by the independent variables. Tukey's test of specific carparisons was ‘ used to assess the significance of mean differences for ANOVA's with three levels of the independent variable (Linton and Gallo, pp. 316- 319, 1975) . Description Tables 4, 5, and 6 surnarize mean percent-correct, standard deviations , and ranges of performance as a function of DER across paired carpeting messages for subjects in groups I, II, and III respectively. Subject performance fell with each decrease in MCR for all competing messages except the DEN MI“ carpeting message. Mean percent-correct scores for this carpeting message were equivalent at tie -16 dB and -18 dB MCRs (approximately 35%) . A reduction in per- forrrance apparently did not occur when the MCR was decreased from -16 dB to -18 dB because performance at the -16 dB MIR already reflected the lowest point on the psycharetric function. Also, performance at the least regative MCR value (-12 dB) for the DEN MI‘ carpeting message, 61 Table 4. Mean percent-correct scores, standard deviations, and ranges of Group I subjects for carpeting messages MI‘ 8 + DEN l and DEN MI‘ + 1 as a function of message-to—competition ratio (N = 12) . PERCENT CDRRECT WC MESSAGE MEAN S.D. RANGE MI‘ 8 + DEN 1 IVER - 14 78.3 12.1 36 - 18 47.6 12.5 40 - 20 35.6 7.5 32 DEN MT + 1 MCR - 16 75.3 11.4 40 - 18 62.3 15.7 44 - 20 43 13.3 32 - 22 33 10.9 36 62 Table 5. Dean percent correct scores, standard deviations, and ranges of Grorp II subjects for carpeting messages MI‘ 8 and DEN Ml" as a function of message-to-carpetition ratio (N = 12) . PERCENT‘OORRECT CCMPEI‘ING MESSAGE MEAN S.D. RANGE MT 8 MGR - 8 80.6. 9.9 28 - 10 58.6 14.3 40 - 12 47.6 11.9 36 - 14 28 5.9 20 DEN MT MIR - 12 63.3 8.4 36 - 14 52 10.6 36 - 16 35 7.2 24 - 10 35 8.2 24 63 Table 6. Mean percent correct scores, standard deviations, and ranges of Group III subjects for carpeting messages MI‘ 8 + l and DEN MT + l as a function of message-to—carpetition ratio (N = 12) . PERCENT CORRECI‘ CCMPETING MESSAGE MEAN S.D. RANGE MT 8 + 1 D’CR - 8 80 7.6 20 - 10 61.6 13.2 40 - 12 44.6 11.7 32 - 14 33.6 10.5 36 DENMI'+DENl DER - 16 82 10.8 28 - 18 70 5.7 20 - 20 49.3 9.9 32 - 22 41.3 10.6 32 64 yielded a mean score significantly less than corresponding NCRs of the other carpeting messages. 'Jhese findings suggest the MCRs selected for the DEN MT competing message should have ranged from -16 dB to -10 dB to include a more carplete portion of the psycho- metric function. Figures 11 through 16 separately illustrate psycharetric functions for the six carpeting message conditions. Each datum represents mean percent—correct scores across 12 subjects. Standard deviations ' (i 1 standard deviation) also are displayed. The figures show the linear portion of the psycharetric fimction was bracketed for each of the various carpeting messages. The relations among the six carpeting messages are compared in Figure 17. Although the psycharetric function slopes of the Six carpeting message psycho- metric functions appear similar, there is a noticeable difference in masking efficiency. More efficient maskers are shifted horizontally toward the right-hand side of the figure. The most efficient maskers aretheMI'BandMI‘ 8 +lcampetingmessages. '111eDENMcarpeting message exhibits an intermediate degree of masking efficiency while theME'8+DENl, DENMI‘+1, andlIINMr+DENlcarpetingmessages are less efficient. Table 7 and Figure 18 summarize psycharetric function slope means , standard deviations , and ranges for carpeting messages within subject groups. Because of the configuration of the mean IIN MP psycharetric function, individual srbject psycharetric function slopes were carputed using three data points (-16, —l4, and -12 dB) from the linear portion of the psycharetric function. ‘Jhe steepness of the psycharetric function slope indicates how Mean Percent-Correct Scores (°/o) 65 100 I I I T 90- ° 80- 70— 60%— 50*- 4o-— 30- 0 l l l I -20 -18 -16 -14 Message-to-Competition Ratio (d 8) Figure 11. Mean percent-correct scores for competing message MT 8 + DEN l as a function of message-to- crmpetiticn ratio. Each datum represents observations of 12 normal-hearing adult subjects tested monaurally. vertical lines denote :_1 standard deviation. Mean Percent-Correct Scores (%) 66 100 I l l r 90' - 80 - > 70 - 0 60 - O 50 - 40 r- 30 - 6 20 - 10 - 0 1 L 1 1 -22 -20 -18 ~16 Message-to-Competition Ratio (d 8) Figure 12. Mean percent-correct scores for competi ng message DEN MT + l as a function of message-to- ccmpetition ratio. Each datum represents observations of 12 normal-hearing adult subjects tested monaurally. Vertical lines denote : 1 standard deviation. -Correct Scores (%) Mean Percent 67 100 I I I m 90 — ' i 80 - . P 70 - o 60 - 50 - 40 - 304- 20 - 10 - 0 l l l 1 -14 -12 -10 -8 Message-to-Competition Ratio (d 8) Figure 13. Mean percent-correct scores for competing message M: 8 as a function of message—to-ccmpetiticn ratio. Each datum represents observations of 12 normal- hearing adult subjects tested monaurally. Vertical lines denote : 1 standard deviation. 68 100' I I I T 90— 70— 60- 4o- 30- Mean Percent-Correct Scores (%) 0| 0 I 0 L I J l ~18 -16 .14 .12 Message-to-Competition Ratio (dB) Figure 14 . Mean percent-correct scores for campeting message DEN MI‘ as a Emotion of message-to—ccmpetition ratio. Each datum represents observations of 12 normal-hearing adult subjects tested monaurally. Vertical lines denote j; 1 standard deviation. Mean Percent-Correct Scores (d B) 69 100 , , 90 - 80 _ 70- 60- 50— 4o- 20+ 10- 0 l l l L -14 -12 -10 -8 Message-to-Competition Ratio (d 8) Figure 15 . Mean percent-correct scores for competing message MT 8 + 1 as a function of message-to- competition ratio. Each datum represents observations of 12 normal-hearing adult subjects tested monaurally. Vertical lines denote : 1 standard deviation. Mean Percent-Correct Scores (%) 7O 100 I I I I 90 - 9 80 - E 70 - ° 60 - o 50 - i? 40 - < o 30 - ° 20 t 10 - 0 J 1 1 1 ~22 ~20 ~18 ~16 Message-to-Competition Ratio (d 8) Figure 16. Mean percent-correct scores for competing message DEN MT + DEN l as a function of message-to- competition ratio. Each datum represents observations of 12 normal-hearing adult subjects tested monaurally. Vertical lines denote :.1 standard deviation. 71 100 I I I I I I I I 90 - 80 F 60- 50- 30- Mean Percent-Correct Scores ((1 B) 20 " O- MT8 O- DEN MT .-MT8+1 A—DENMT-I-DEN1 O-MT8+ DEN1 A— DENMT+1 0 l l l L l l l l I l ~22 ~20 ~18 ~16 ~14 ~12 ~10 ~8 Message-to Competition Ratio (d B) Figure 17. Mean percent-correct scores for six carpeting message conditions as a function pf message-to-competition ratio. Each datum represents the observations of 12 normal-hearing adult subjects tested monaurally. 72 Table 7. Mean psychometric function slope values, standard deviations, and ranges for the six carpeting message conditions. Mean psycharetric function slope values were determined using linear regression equations carputed for each subject. PSYCHCMETRIC FUNCTION SLOPE VALUES (%/dB) (INPETIM; MESSAGE MEAN S.D. RANCE MI‘ 8 + DEN 1 7.0 1.8 6.0 DEN MT + 1 7.3 2.9 6.0 DEN MI' 7.4 2.4 8.0 IEN MI' + DEN 1 7.3 2.2 7.0 MI' 8 + 1 7.8 1.8 5.8 MI' 8 8.5 2.1 7.2 Mean Slopes (%IdB) 73 MT8 + DEN 1 DEN MT + 1 DEN MT + DEN 1 DEN MT MT8 + 1 MT8 Competing Message Condition Figure 18. Mean psycharetric slope values for the six competing message conditions. ‘ Psycharetric function slope values were determined for each competing message condition using linear regression equations computed for each subject. 74 quickly performance improves as the MCR becares more favorable . The Mr 8 carpeting message exhibits the steepest slope value (8.5%/dB) followed by the MI‘ 8 + l carpeting message (7.8%/dB) and DEN MT com- peting message (7.4%/dB) . Carpeting messages DEN MT + l and DEN MI' + DEN 1 had identical psychometric function slopes of 7. 3%/dB. The flattest slope value occurred for the MI‘ 8 + DEN l carpeting message (7.0%/dB) . Overall, the psychometric function slope values for the six carpeting messages appear similar, ranging only l.5%/dB from the flattest slope (MT 8 + IEN l) to the steepest slope (MT 8). Table 8 and Figure 19 present mean 50% masking efficiency levels for carpeting messages according to subject group. mese values re- present the interpolated IVER at which a 50% performance level would be achieved. Masking efficiency varies appreciably among the can- petitors, ranging 9.1 dB from the least efficient to the most efficient masker. 'Ihe MI‘ 8 and MI' 8 + l competitors are the most efficient maskers with 50% masking efficiency levels of -ll.4 and -ll.6 dB respectively. The JEN MI' carpetitor is the third most efficient, masker (-14.0 dB) whereas theMI‘ 8+DEN l, DENm+ l, andDENMT+ DEN l carpeting messages are less efficient with levels of —l7.7, -19.5, and -2o.5 6.8 respectively. Reliability of Measurement Odd-Even Reliability. Table 9 and Figmes 20, 21, and 22 summarize intra-list reliability results for the four MCRs of each of the six carpeting messages. Significant (.10 level) correlations exceeded .497. These coefficients reflect the correlation between the odd-even items in SSI test lists across subjects as a function of IVER and carpeting message conditions. The response to the first test 75 Table 8. Mean estimated 50% masking efficiency, standard deviations, and ranges for the six carpeting message conditions. These levels were estimated from linear regression equations carputed for each subject. 50% MASKIME EFFICIENCY (dB) C(NPE'I‘ING MESSAGE MEAN S.D. RANGE MI‘ 8 -11.4 0.5 2.1 MII 8 + 1 -11.6 0.92 1.8 DEN MI‘ ~14.0 0.87 3.3 MI‘ 8 + DEN 1 -17.7 0.65 2.3 IEN MI‘ + 1 -19.5 1.2 4.1 DEN MI‘ + DEN 1 -20.5 0.99 4.1 Mean 50% Masking Efficiency Level (dB) 76 MT8 MT8 + 1 DEN MT MT8 + DEN ‘l DEN MT + 1 DEN MT + DEN 1 Competing Message Condition Figure 19. Mean 50% masking efficiency levels for the six carpeting message conditions. Fifty percent masking efficiency levels were estimated for each carpeting message condition fran linear regression equations carputed for each subject. 77 Table 9. Intra—list reliability correlation coefficients of three experimental groups for their respective competing message calditions as a function of message-to—competition ratio. GROUP C(IVIPETING MESSACE NESSAGE-‘IO—CINPETITION RATIO (dB) I MI'8+DEN1 -14 -16 -18 -20 (N — 12) r +.61 +.7l* +.75* +.85* DENMI'+1 -16 -18 -20 -22 r +. 77* +. 52* +. 79* +. 32 II MT 8 - 8 - 10 - 12 - 14 (N = 12) r +.7Z* +.28 +.IO +.I7 DEN MI‘ - 12 - 14 - 16 - 18 r +. 77* 0 +. 35 +. 61* III DENMT+DEN1 -16 -18 —20 -22 (N = 12) r +.95* +.29 .29 +.38 *Significant beyond P on _<_ .10 level. 78 432 323808 S. mfi pm Re. pmommoxm mcoflmamuuoo unmoflwcmfim .3835 NH so 083 ms.” Sumo comm .038 coflufiummpnconouimmmmmme mo cofluocsm m mm H + 92 zmo can a Ema + m a: mommmmme magmasoo now mucmfioammooo coflumamuuoo wueafinmflamu umfiaimuucH .om ousmflm 62 03mm. co_~_.anoo.o..oammmo_2 vd Nd 0.0 Nd v.0 md 8:650 mammmms. oczmano S. 3. cm. mm. 3. 9. mp. cm. a A + _ . A . ill llfilfill IIIII III-Ill ill. I'll .I I'll I F F I r1 r1 r1 r _ r b . _ _ r _ F+._._22mn_ szo+mt2 0.? ed . (”8111810511300 uoueieuoo 79 .Ho>0a >uflaflnmeoua ca. 02» um hmv. owommoxw mcofiumamuuoo ucmoaw recmflm .muowflndm NH :0 00mg we :Bumo comm .ofiumu 2.0330950 rowimammmme mo 830:3 m mm B: an. pom m 52 mmammmme 93.5950 qu mucwqoflmmmoo coflumamuuoo Nuwafinmwamu umfiHImHucH .HN madcap Am 3 28¢ cozzmqeoo.2.mammmmz vd Nd Tll IIIIIIIIE.I|.III lllll llllll 1 _ _ _ _ od Nd «.0 od md :2 zmo . . 3.2 8:650 mammmms. 0:23:50 0.? (J) siuegomeoo uoumeuoo 80 .3833 NH :0 comma we Esomo comm . Hoe/0H xuflaflnmnoua ca. ocu um emu. oopoooxo mcoHomHouuoo ucm0HMHcmHm .ofiomu soflfiooasboiouiommmmme mo :ofluoca m mm H zmc + B: zma ocm a + m a: .mommmmoe 0:309:00 ecu mucofloflmwooo coflumaouuoo Nuflaflnmfiaou omflarmnocH .NN ousmam Am 3 ozmm cozzmanoééommmoz or. 9. ON. NN. w. or. NF. 3. fl _ q q _ a _ _ fill llllllllulla'lllII-lllllll IIL T 1 In _ _ _ _ _ _ _ so No 3 .3 to ad md wzmo +._._22wo F+m._._2 coEocoo oammmms. aczanoo o... (J) stuegomeoo uouereuoo 81 item of each list was anitted when tabulating the responses for each list to provide an equal number of odd and even items (12) . Omitting the first item and splitting the list in half reduced the mmrber size front 25 to 12. Therefore, the Spearman-Bravn formrula (Kerlinger, pg. 438, 1964) was used to correct for the lesser number of items (half—list) used in calculating the coefficients. Odd-even correlation coefficients for each of the NCRs of individual carpeting messages were transformed to 2 scores, summed, and divided by the nurrber of correlation coefficients (4) . 'lhe average Z score for each carpeting message was then re-transformed to a correlation coefficient to yield "average" correlation coefficients for each carpeting message (see Table 10). 'me highest intra-list correlations occurred for the MI' 8 + DEN l carpeting message condition (Group I). Correlations as each MCR for this carpeting message exceeded the significance criterion. Three of four correlations for the DEN MT + l carpeting message (also Group I) also achieved significance. The MT 8 carpeting message produced poor intra-list reliability, except at the easiest MCR (-8 dB). 'Jhe DEN MI‘ carpeting message had a good reliability for the hardest and easiest IVCRS, (-18 and -12 dB respectively) but poor reliability for inter— mediate IVCRs. Reliability for the MI' 8 + l carpeting message was good cnlyatthehardesthCR (-14 dB) whileDENMI‘+m:Nlcarpeting messages exhibited good reliability only at the easiest MIIR (-16 dB). Intra-list reliability for the various carpeting messages did not display any particular pattern as a function of message to carpetition ratio. However, a pattern of reliability was established as a functiar of group assignment. Subjects in Group I had good reliability across 82 Table 10. "Average" intra-list reliability correlation coefficients of three experimental groups across message-to—carpetition ratios for their respective carpeting message conditions. GROUP CCMPETING MESSAGE "AVERAGE" CORRELATICI‘I COEFFICIENTS (r)* I MI' 8 + DEN 1 + .74** (N=12) DENMI'+1 +.64** II MI' 8 + .31 (N = 12) DEN MI‘ + .52** III MI‘ 8 + 1 + .37 (N = 12) DEN MI' + DEN 1 + .62** *r to z transformations interpolated from Fisher' 3 z transformation (Bruing and Kintz, pp. 250-251, 1977). **Significant beyond Per 5 .10 level. 83 all message to carpetition ratios for both carpeting messages , except -22 dB for the carpeting mesSage DEN MI‘ +1. The odd-even correlation coefficients calculated are sensitive to performances at the level of the individual subject, but not at the group level. 'Ihe correlation coefficients do not take advantage of the "averaging out" effect that occurs at the group level. A case in point is the odd—even correlation coefficients obtained at -16 dB IVER for the DEN MI' carpeting message. The odd-even correlation coefficient for this particular experiment is +.35. , 'lhis relatively low correlation coefficient suggests a significant discrepancy between odd and even item performance. Yet, tabulation of scores across subject group yielded odd and even incorrect responses of 82 and 81, respectively. These results reflect almost perfect agreement between the number of odd incorrect responses and the nurber of even incorrect responses across the srioject group (n = 12) . Binamial "lest. 'Ihe binomial test was employed as an alternative method of assessing intra-list reliability. 'lhe binamial test assesses whether there is a significant difference between observed frequencies and expected frequencies. In this study odd and even incorrect responses each have expected frequencies of 50%. That is, 50% of the incorrect items should be odd and 50% should be even. The null hypothesis, then, is incorrect even items = incorrect odd items = 0.5. This hypothesis may be tested using the formula 2 125.219 ATP—Q 84 where X is the number of odd (or even) items incorrect, N is the total number of incorrect items, I? is specified by the null as 0.5, and Q = l - AP (also 0.5) . The resulting Z value is approximately normally distributed with 0 mean and unit variance. Using a two-tailed test and a significance criterion of 0.5, the null hypothesis is rejected if the Z value is less than -l.96 or more than +1.96. Tables 11 and 12 summarize the percentage of odd-even incorrect items and calculated Z scores for each experimental condition, respectively. In no instance was the binomial test statistically significant. Thus, although odd-even item reliability was not high, odd-even error patterns did not differ significantly. Test-Ihtest Reliability. Test-retest correlation coefficients were computed for the six carpeting message conditions at the MSR value which produced the nearest to 50% correct score for individual subjects (see Table 13 and Figure 23) . Significant (.10 level) correlations exceeded .497. The test-retest correlation coefficients ranged from -.33 (DEN MI‘) to +.84 (MI‘ 8 + 1). If data from the subjects whose second score was at least 24% greater than the first score (possible learning effect) are eliminated, correlation co- efficients becare more positive, ranging from -.17 to +.87 (see Table 14 and Figure 24). In either case, correlation coefficients are quite variable across the carpeting message conditions. ' Standard Error of thans. 'Ihe standard error of mean values as a function of MIR for the six competing messages are sham in Table 15. than standard error of mean valLes across carpeting messages were similar, ranging fram 2.6 for the DEN MI‘ stimulus to 3.8 for the DEN MI‘ + 1 stimulus. This indicates that the interval containing the "true" 85 Table 11. Percent—correct of odd and even items for the six carpeting messages as a function of message-to—carpetition ratio. Percentages are rounded off to the nearest wtnle mmber. MESSAGE-MENTION RATIO (dB) MT8+DENl -14 —16 -18 -20 % Odd 52 48 46 49 % Even 48 52 54 51 DENMI‘+1 -16 -18 -20 -22 % Odd 52 SI 52 48 % Even 48 49 48 52 MT 8 -‘ 8 - ‘10 - ‘12 - l4 % Odd 55 50 50 49 % Even 45 50 50 51 IIN MI' - 12 - 14 - 16 - 18 % Odd 49 58 50 50 % Even 51 42 50 50 MI‘8+l ‘-'8 -”10 -‘-‘12 -'-14 % Odd 56 51 49 53 % Even 44 49 51 47 UZNMT+DENl -16 ‘—"18 '-'-‘20 “-“22 % Odd 59 48 50 49 % Even 41 52 50 51 86 Table 12. Z values for the six competing message conditions as a function of message-to—carpetition ratio. Equivalency of odd and even item correct rate is assumed if Z value is less than - 1.96 or greater than + 1.96. MESSAGE-MENTION RATIO (dB) MI‘8+DENl -14 -16 -18 -20 Z value .35 .37 1.15 .22 DENMI‘+1 -16 -18 -20 -22 Z value .34 .19 .47 .44 MP 8 - 8 — 10 - 12 - 14 Z valLe .90 ..09 .15 .29 DEN MI‘ - 12 - 14 - 16 - 18 Z value .21 1.32 .08 .21 MP 8 + l - 8 - 10 - 12 - 14 Z value .85 .19 .16 .73 DENM+DENl -16 -18 -'20 -'-22 Z value 1.43 -.32 .08 .05 87 Table 13. Test-retest correlation coefficients (Pearsm r) for six carpeting messages. Test-retest measurements were obtained at the message-tc-carpetition ratio which produced the nearest to 50%-correct score for each subject. CCMPEI‘INGMESSACE 'I‘ES'ILRE'IEST CORRELATICN CCEFFICIEN'I‘S (r) MI'8+DEN1 +.36. (N=12) DENMP+1 +.36 (N=12) MP8 -.05 (N=12) IINMI‘ -.33 (N=12) MT8+1 +.84* (N=12) [JENNH'+DEN1 +.35 (N= 12) *Significant beyond P or _<_ .10 level. Correlation Coefficients (r) 88 1~0 I I I I T I MT8+DEN1P DENMT+1- MT8- DENMTPL MT8+1~~ DEN MT + DEN1 r- Competing Message Condition Figure 23. Test-retest correlation coefficients for six carpeting messages. Test-retest measurerents were obtained at the message-to-carpetition ratio which produced the nearest to 50%-correct score for each subject. 89 Table 14. Test-retest correlation coefficients for the six carpeting message conditions if data are eliminated from subjects whose second scores were 24% greater than the first. Test-retest measurements were obtained at the message-to—carpetition ratio which produced the nearest to 50%-correct score for each subject. CCMPETING msszxcm 'IES'I‘-REIEST CORRELATION COEFFICIENTS (r) MI' 8 + HEN l + .36 (N = 12) DEN MI‘ + 1 + .70* (N = 10) MI' 8 + .29 (N = 10) DEN Ml‘ - .17 (N = 8) MT 8 + l + .87* (N = 11) DEN MI‘ + DEN 1 + .50* (N = 11) *Significant beyond P or _<_ .10 level. Correlation Coefficients (r) 90 10 i I l T I f + I 10 I _ MT8+~DEN1 a DENMT+1 ANTS - DENMT - MT8+1- DENMT+DEN1 - Competing Message Condition Figure 24. Test-retest correlation coefficients for six competing messages if data are eliminated from subjects whose second scores were 24% greater than the first. Test-retest measurements were obtained at the message-to-ccmpetition ratio which produced the nearest to 50%-correct score for each subject. 91 Table 15. Standard error of the mean (SEN) for six carpeting messages as a function of message-to—carpetition ratio (N = 12) . Also sham is mean SEM across message-to—carpetition ratios for each carpeting message. CCMPETII‘E MESSACE MESSAGE-TO-CCMPEHTTICN RATIO (dB) MEAN SEM MT8+DENl -14 -16 -18 -20 3.0 SEN 3.6 2.6 3.8 2.3 DENMI'+1 -16 -18 -20 -22 3.6 SEN 3.4 4.7 4.9 3.3 MI' 8 - 8 - 10 - 12 - 14 3.2 SEN 3.0 4.3 3.5 1.8 [EN MT - 12 - 14 - 16 - 18 2.6 SEM 2.5 3.2 2.2 2.5 MP8+1 -‘8 -10 -12 -14 3.2 SEN 2.3 4.0 3.5 3.2 DENMI‘+DEN1 -16 -18 -20 -22 2.8 \l U) o o (A) o N SEN 3.2 1. 92 mean score is similar in size across carpetitors. In other words, precision of measurerent appears to be fairly equivalent among the six carpeting messages. Inferential Analysis The statistical analysis included several ANOVA's of psycho- metric function slope Values and 50% masking efficiency levels. Significant F-ratios exceeded critical values at the .05 level. Psycharetric Function Slope Values. Tables 16 through 22 summarize the one-way within-subject and one-way between-subject ANOVA' s of psychometric function slopes for the various carpeting message conditions . ANOVA of the psycharetric function slope for the MP 8 versus DEN MT carpeting message carparison (Table 16) in- vestigates serantic cartent in a single-gromd carpeting message. Tables 17, 18, and 19 contain ANOVA results of three carpeting message pairs designed to examine possible differences in psycharetric function slopes as a function of differences in a foreground- background carpeting message containing serantic content in (1) back- ! ground only, (2) foreground only, and (3) background and foreground. Table 20 presents ANOVA results of competing messages MT 8 + DEN l and [EN MP + 1. This carparison examines the effect of the position and type of semantic conteit in a foreground-backgrormd carpeting message . 'Ihe ANOVA results of psycharetric function slopes of carpeting messagesMI'8,t/rl‘8-I-landMI'8-i-DENlaresl‘aminTable21. 'Ihis carparism was made to determine the effect on psychometric function slope of changing the foreground status in a carpeting message con- tainingaspeechbackground. AnANOVAofDENMI‘, DENMI‘+1and 93 Table 16. (he-way within-subject analysis of variance of psycho- metric flmction slope values for carpeting messages MI‘ 8 and DEN MI‘. = 2 SOURCE (11“ SS MS Fobs Fcrt .05 Ti Carpetitor (A) 1 17 17 15.04** 4.84* 0.14 Subjects (S) 11 89.5 A x S 11 12.5 1.13 Total 23 119 *Fram F distribution table (Winer, pp. 804-869, 1971) **Significant beyond For _<_ .05 level. 94 Table 17. Ole-way between-subject analysis of variance of psycharetric function slope values for carpeting messages MI‘ 8 + DEN 1 and DEN MI‘ + [EN 1. SOURCE (31“ SS MS Fobs Fort = .05 Carpetitor l . .6 .6 .14 4.30* Error 23 91.7 4.16 Total 24 87.6 *Fram F distribution table (Winer, pp. 804-869, 1971) 95 Table 18. Onedway’between-subject analysis of variance of psycharetric function slope values for carpeting messages DEN MI' = 1 and DEN MI‘ + DEN 1. SOUICE dF SS MS Fobs Fort = .05 Carpetitor l 6 6 2.5 4.30* Error 23 56 2.4 Total 24 62 *From F distributim table (Winer, pp. 804-869, 1971) 96 Table 19 . Che-way within-subject analysis of variance of psycharetric functial slope values for carpeting messages MI' 8 + l and DEN MI' + DEN 1. SOURCE dF SS MS FObS Fort - .05 Competitor 1 1.5 1.5 0.88 4.84* Subjects(S) 11 71.3 A x S 11 18.8 1.7 Total 23 91.6 *Fram F distribution table (Winer, pp. 804-869, 1971) 97 Table 20. Ole-way within-subject analysis of variance of psycharetric function slope values for carpeting messages MT 8 + DEN 1 and DEN MI‘ + 1. SOURCE (31" SS MS Fws Fort " .05 Carpetitor 1 0.5 0.5 0.18 4.84* Subject (S) 11 99.7 A x S 11 29.1 2.65 'Ibtal 23 129.3 *From F distribution table (Winer, pp. 804-869, 1971) 98 Table 21. Chedway'betweenrsubject analysis of variance of psydhometric function slope values for competing messages MH‘8, Nfl‘8 +,1' and.Mfl'8 + DEN 1 (Group N = 12). SOURCE dF SS MS FObs Fort‘— .05 competitor 2 13 6.5 1.8 4.14* Error 33 120 3.6 Total 35 133 10.1 *FromrF‘distribution table (Winer, pp. 804—869, 1971) 99 Table 22. (he-way between-subject analysis of variance of psycho- metric function slope values for carpeting messages DEN MI‘, DEN MI‘ + 1, and DEN MD + DEN 1. SOURCE (11" SS MS Fobs Fcrl: = .05 Carpetitor 2 0.07 0.035 0.005 4.14* Error 33 210.6 6.4 Total 35 210.67 6.435 *Fram F distribution table (Winer, pp. 804-869, 1971) 100 DEN tfl‘ + DEN l stimuli psychometric function slope values (Table 22) was performed to determine the effect a1 psycharetric function slopes of changing the foreground status in a noise background carpeting message. None of these analyses revealed significant differences in mean psychometric function slopes, except for the MI‘ 8 versus DEN MI‘ carparison. ‘Ihus , the serantic contert factor in the various experimental conditions generally had no effect upon SSI slopes. Two consideratiars qualify the significance of the single-ground carpeting message serantic content effect on psycharetric function slope values. First, an ETA-squared value of .14 indicates that only a small portion of the variance in slope can be attributed to the serantic content factor. SeOOnd , slope valres for the DEN MI‘ competing message were carputed ar the basis of three NERs while slope values for the MP 8 carpeting message were carputed using all four NCRs. Different carputation procedures may have created an artifactual disparity between the psycharetric function slopes of the two carpeting messages, especially since the slope 0f the MI‘ 8 carpeting message rises more Sharply from the third tCR (-10 dB) to the fourth MIR (-8 dB) than any other carpeting message. A carparison of psycharetric function slope values carputed on the basis of three MIRs for both carpeting messages was not possible because several subjects had not reached the 60% correct level by the third MIR (-10 dB) for the MI‘ 8 carpeting message. Psycharetric function slope values computed using mean scores at three corresponding NCRs (i.e., linear portion of psycharetric function) revealed similar slopes of 7%/dB and 7.5/dB for the DEN MI‘ and MI‘ 8 carpeting messages, 101 respectively. Fifty Percent Masking Efficiency Ievels. Tables 23 through 28 summarize results of the ANOVA's of 50% masking efficiency levels for the various carpeting message comparisons. Several of the ANOVA's produced sigrnificant results; each will be discussed separately. The MP 8 versus DEN MI‘ carpeting message carparisan exarmines the effect of single-ground carpeting message serantic cantent on 50% masking efficiency levels. Table 23 contains the enemy within- subject ANOVA for these carpetitors. The analysis indicates a significant semantic content effect for the single-ground carpeting message. The ETA-squared value indicates 34% of the dependent variable score can be accounted for by the serantic cantent factor, suggesting a mild-to-moderate experimental effect. ‘Ihree pairs of competitors were carpared to investigate possible differences in masking efficiency levels as a function of differences in a foreground-background carpeting message containing semantic content in (1) background only, (2) foregrormrd only, and (3) back— ground and foreground. Table 24 displays the result of the ane-way between-subject ANOVA of masking efficiency levels for carpeting messages MI' 8 + DEN l and DEN MI‘ + DEN l. A significant difference in the mean 50% masking efficiency levels of the two carpeting messages was determined, indicating a significant background serantic content effect. ‘Ihe (MEGA-squared value is .39, suggesting approximately 40% of the score variance can be attributed to background serantic content. Again, this indicates a mild-to-moderate experimental effect. ANOVA resultsofthecarpetingmessages DENMT+landDENMr+DEN1masking 102 Table 23. One-way within-subject analysis of variance of 50% masking efficiency levels for competing messages MI‘ 8 and DEN MI'. _ 2 SOURCE dF 33 MS F obs F crt— .05 w Carpetitor (A) l 20 20 13.3** 4.84* 0.34 Subjects (5) ll 21 AxS ll 17 1.54 Total 23 58 *Fram F distribution table (Winer, pp. 804-869, 1971) **Significant beyond P 01 _<_ .05 level. 103 Table 24. Che-way between-subject analysis of variance of 50% masking efficiency levels for competing messages MI' 8 + DEN 1 and DEN MT + DEN l. SOURCE (315‘ SS MS Fcbs Fcrt = .05 w Competitor 1 47.1 47.1 16.52“ 4.30* .39 Error 23 62.8 2.85 Total 24 109.9 *From F distribution table (Winer, pp. 804—869, 1971) **Significant beyond P a _<_ .05 level. 104 Table 25. Che-way between-subject analysis of variance of 50% masking efficiency levels for competing messages DEN MI‘ + l and DEN MI' + DEN l. SOURCE dF SS MS F obs Fcrt_ .05 Competitor 1 6 6 2.36 4.30* Error 23 56 2.54 Total 24 62 *Fram F distribution table (Winer, pp. 804-869, 1971) 105 Table 26. Che-way within-subject analysis of variance of 50% masking efficiency levels for carpeting messages MI' 8 + l and DEN MI‘ + DEN 1. _ 2 SOURCE (31“ SS NB Fd)s Fert— .05 n Competitor (A) 1 485 485 404** 4.84* .96 Subject (S) 11 6.7 A x S 11 13.3 1.2 Total 23 505 *From F distribution table (Winer, pp. 804-869, 1971) **Significant beyond P 0L 5_ .05 level. 106 Table 27. (he-way within-subject analysis of variance of 50% masking efficiency levels for carpeting messages MT 8 + DEN l and DEN MT + l. _. 2 SOURCE dE‘ 55 MS F obs Fcrt .05 n Competitor (A) 1 20 20 13.3** 4.84* .34 Subject (8) 11 32 Ax s 11 17 1.5 Total 23 58 *From F distribution table (Winer, pp. 804-869, 1971) **Significant beyond P a _<_ .05 level. 107 Table 28. (he-way between-subject analysis of variance of 50% masking efficiency levels for competing messages MI‘ 8, MI‘ 8 + 1, and MI‘ 8 + DEN l. _ 2 SOURCE (1? SS MS Ems , Port .05 8) Competitor 2 313 156.6 l6.l4** 4.14* .37 Error 33 323.8 9.8 Total 35 636.8 166.4 *From F distribution table (Winer, pp. 804-869, 1971) **Significant beyond P a _<_ .05 level. 108 efficiency levels (Table 25) indicate no significant foreground semantic content effect. Table 24 contains the results of the one- way within-subject ANOVA of competing messages MI‘ 8 + l and DEN MT + DEN l 50% masking efficiency levels. This analysis indicates a significant foreground-background content effect. ETA-squared suggests a very high percentage of score variance (96%) can be accounted for by He foreground-backgromd semantic content factor. Collectively, these three analyses indicate semantic content has a significant effect on masking efficiency levels Wren it occurs in the background (multi-talker) or background and foreground (mmlti-talker and single-talker) .1 Single-talker foreground semantic content does not exhibit a significant experimental effect. TheNE8+DENlandDENNE+10cmpetingmessage comparison was intended to investigate the effect of semantic content position and type on masking efficiency. The MI‘ 8 + DEN 1 carpeting message has a multi-talker background with a single-talker speech modulated noise foreground. Conversely, the DEN MI‘ + 1 competitor has a multi- talker speech modulated noise background and a single-talker fore- ground. ‘Ihe ANOVA results in Table 27 reveal a significant experi- mental defect for semantic content position and type. The EPA-squared value (.34) suggests a mld-w—mderate experimental effect. It is important to note that this carparison examines two aspects of semantic content sixmltaneously. Mean 50% masking efficiency levels obtained using the MI‘ 8, MI‘ 8 +1, andMI‘8+DENlcatpetitorswerecwparedtodetermine theeffect of changing foreground status in a competing message containing a speech background. The ANOVA results shown in Table 28 indicate a significant 109 difference in 50% masking efficiency levels among the three carpetitors. The experimental effect, estimated by (MEGA—squared, was mild-to-moderate (.37) . Tukey's test of specific carparisons revealed the MI“ 8 + DEN l carpeting message 50% masking efficiency level differs from those of the MI‘ 8 and MI‘ 8 + 1 carpeting messages, but the masking efficiency levels of the latter two carpetitors do not differ fram each other (see Table 29) . Thus, a differential effect occurred only when the noise foreground was coupled to the speech background. A final carparison determined the effect on masking efficiency of changing the foreground status in a noise background carpeting message. Table 30 displays the ANOVA result for carpeting messages DEN MI‘, DEN MI‘ + l, and IEN MI‘ + DEN 1. The significant F-ratio indicates a significant difference in mean 50% masking efficiency levels among the three carpeting messages. The .74 (MEGA-squared value reflects a relatively strong experimental effect. 'I‘Lflcey's test of specific carparisons reveal a differential experimental effect occurred for the DEN MI‘ carpeting message only (see Table 31) . 'Ihe effectsofcarpetingmessagesIINM-l-landDENM‘I‘+DENlon 50% masking efficiency levels were equivalent. SW SSI performance scores were obtained as a function of MIR for six different carpeting message cmditions. Nban psydmetric functions were determined from individual subject data. The general calfigurations of all six mean psycharetric functions were similar. Various aspects of test reliability were estimated using several 110 Table 29. Tukey's test.of specific comparisons of mean.50% masking efficiency levels for carpeting messages MI‘ 8, MI' 8 + l, and MI‘ 8 + DEN l. MT 8 MI‘ 8 + 1 MT 8 + DEN 1 (—11.4) (-ll.6) (-17.7) MI‘ 8 (-ll.4) - -0.2 -6.3* MI‘ 8 + 1 -- -6.2* (-ll.6) MI‘ 8 + DEN 1 - (-17.7) *Exceeds critical value (mean) at .05 probability level. 111 Table 30. One-way between-subject analysis of variance of 50% masking efficiency levels for carpeting messages DEN MI‘, DEN MI‘ + l, and DEN MI‘ + DEN l. SOURCE dF SS MS Fobs Fort .05 8) Competitor 2 294 147 54.4** 4.14*" 0.74 Error 33 92 2.7 lDotal 35 386 149.7 *Fram F distribution table (Winer, pp. 804—869, 1971) **Significant beyond P or i .05 level. 112 Table 31. Tukey's test of specific carparisons of mean 50% masking efficiency levels for carpeting messages DEN MI‘, DEN MT + l, and DENMI-l-DENl. 50% MASKING EFFICIENCY MEANS. AND MEAN DIFFERENCES (dB) DEN MT DEN MI‘ + L DEN MI‘ + DEN 1 (-14 . 0) (-19 . 5) (-20 . 5) DEN MI‘ -- -5.5* -6.5* (-14.0) DEN MI‘ + 1 - -1.0 (-19.5) DEN MI‘ + DEN 1 (-20.5) *Exceeds critical value (mean) at .05 probability level. 113 statistical procedures. Test-retest reliability was lower than expected for most carpeting message conditions. However, test-retest data was obtained within the most variable portion of the psycharetric function and the correlation coefficients should be considered very conservative estimates of test-retest reliability. Intra-list reliability was assessed using the odd-even method and the binanial test. These procedures failed to reveal any significant differences in odd item versus even item performance. SEMs calculated for this study were, generally, much Smaller than SENS calculated for the ' other SSI studies reporting dispersiar data. wean psycharetric function slope values for the six carpeting messages were similar, ranging from 7."0%/dB for the MI‘ 8 + DEN l carpeting message to 8.5%/dB for the MI' 8 + 1 carpeting message. The relatively narrow range in which these values occur indicates that semantic content had little, if any, effect on how quickly performance rises as a function of changing MCRs. Masking efficiency varied significantly as a functiar of carpeting message cardition. The MP 8 and DEN MI‘ comparison revealed that semantic cartent caused an increase in masking efficiency (2.6 dB) in a single-ground continuous spectrum carpeting message . The effect of semantic cartent was also determined for foreground-background carpeting messages having temporally continuous backgrounds and temporally discontinuous foregrormrds . Generally, semantic cartent had little effect on masking efficiency when it occurred ally in the temporally continuous background or temporally discautinuous foreground. However, semantic cartent in a temporally discartinuous foreground and 114 in a temporally continuous background had a marked effect on masking efficiency. CHAPTERIV DISCUSSICN IN'I'RODUCI‘IGQ Chapter III describes the ipsilateral competing message SSI test performance of 36 normal-hearing srbjects divided equally into three experimental groups. Each group listened monaurally to SSI target sentences in the presence of two different carpeting messages at each of four MIRs. Mean psycharetric functions were determined from individual subject data. The two dependent variables of psychometric function slope values and 50% masking efficiency levels were derived from individual psycharetric functions . Psycharetric function slope values indicate how quickly performance changes as a function of IVER , whereas 50% masking efficiency levels estimate the IVER at which a 50% correct performance should be achieved. The psycharetric function slope values and 50% masking efficiency levels were analyzed to determine significant differences as a functiar of carpeting message condition. Whal significant differences were found (. 05 level) appropriate strengths of association measures were used to determine the amount of dependent variable variance accounted for by the independent variable. A discussiar of the experimental findings and their implicatiars follow. 116 EXPERIMENTAL FINDINGS Reliability Odd-Even and the Binomial Test. The use of odd—even correlation coefficients and the binomial test to assess intra—list reliability allows examrination of two aspects of response carsistency at the level of the item. The odd-even method indicates a fair amount of variability among individual subjects in the number of the respective odd and even incorrect responses , whereas the binamial test suggests the lack of a difference between odd and even item incorrect response rates at the group level. Close agreement between the percentage correct of response items across subject group when several individual subjects within the group exhibit poor agreement between odd and even performance indicates differences in individual odd and even perfor- mances are randan in nature. That is, the experimental procedure did not systematically favor either odd or even item performance. ‘ Test-Retest Reliability. Tast reliability across successive administration of the SSI test was assessed by retesting all subjects at both carpeting message conditions. Generally, test-retest reliability was low to moderate except for the MI' 8 + l carpeting message. This carpetitor exhibited a test-retest correlation coefficient of +. 84, indicating highly consistent performances across successive test presentations . Several factors may have cartributed to the relatively low test-retest reliability for five of the six competing messages. An obvious potential hindrance to high test-retest reliability in the present study was learning effect. It is possible subjects learned to deal with the carpeting messages. Most subjects stated 117 they adopted one or more strategies during the course Of the test procedure. Jerger (1970) noted that subject strategies ranged fran listening for entire sentenCes at easy MCRs, listening for words or word pairs at harder M185 and finally listening for temporal patterns at the most difficult MCRs. Martin and Mussel (1979) found synthetic sentences could be identified on the basis of single word recognition. Strategies reported by present subjects included trying to ignore the carpetitor, listening for one word, listening for particular phonemes, and listening for sequential patterns of the sentenCes. Certainly, the status of almost all subjects changed from naive to experienced by the end of the experimental procedure. Also, some subjects may have recognized reoccurring patterns of the primary message set toward the end of the testing procedure. The two retest sessions occurred following the eighth administration of the target sentences. Although previous experience suggested this type of learning effect developed sometime after the twelfth test presentation, it is possible that it occurred earlier. The effect of learning on SSI test performance has been studied by Beattie and Clark (1982). They presented five synthetic sentences ateachcf sixNCsthichvariedinhvostteps fram-6dBto+4dB for two trials (60 synthetic sentences) ,3 five synthetic sentences at each of four IVERs ranging from -2 dB to +4 dB during a third trial (20 synthetic sentences) and ten synthetic sentences at each of the original six NCRs for the fourth trial (60 synthetic sentences). Significant improvement in mean scores occurred between the first and 118 second trial for IVERS of 0 dB, +2 dB, and +4 dB while score improvement was delayed until the third trial for the ‘-2 dB MIR value. Mean scores remained consistently low at MCRs of -6 dB and —4 dB for trials are, two, and four. In all instances the carpetitor was a four-talker complex. These results suggest that learning effect is greatest and occurs fastest at easier I‘CRs. It appears that subjects are initially confused by the carpeting speech message, even at favorable MIPS . Another cartributor to the relatively low test-retest reliability observed in this study was the use of the 50% performance level for administration of the retest. Thornton and Raffin (1978) found the greatest degree of subject variability near the 50% correct performance region for ten and 25—item speech discrimination tests. Theoretically, the 50% correct performance region of any psycharetric function is inherently the most variable segment of the function. However, it is also the region in which performance is leaSt affected by extremely difficult NCRS (low performance) or extremely easy NCRS (high performance) . Test-retest reliability would be artifactually high if successive administrations involved MIRS where subjects performed near 10% or near 100% correct. Thus, the present test-retest correlation coefficients should be viewed as very conservative estimates of reliability. less conservative and more complete estimates would be those based upar repeated measurement of the entire psycharetric function. ' Standard Error‘of the Dean. A relatively small numerical value for the SEM indicates clustering of individual scores about the group mean, independent of sample size. Mean SEN values for this study were 119 relatively small in numerical value and narrow in range (2.6 for DEN MI‘ carpeting message to_3.‘8 for DEN MI' + l competing message). This indicates there was close clustering of performance scores around group mean values which remain consistent across competing message carditions . Reliability Indices of Other SSI Studies Indices of reliability used in this study (odd-even, test-retest, SEN) have not been routinely used in other SSI studies . Four studies (Keith, 1977; Goldschmidt, 1979; Webb and Greenberg, 1980; and Beattie and Clark, 1982) did report standard deviations fram which SENS can be Calculated. Goldschmidt (1979) was the only study that reported test-retest data. All of these studies erployed normal listeners under earphones and a ten-item test. Beattie and Clark (1982) used a four-talker carpeting message while the other three used a single-talker carpeting message. Mast SSI studies have ignored dispersion and reliability data entirely, reporting only mean performance scores as a function of MIR. SENS calculated for the four SSI studies reporting dispersion data were, in most cases, significantly larger than SENS calculated for the present study. Mean SENS collapsed across MIRS in the present study range fran 2.6 to 3.6 for the six carpeting message conditions. The mean SEN across MZRS of -20, -10, and 0 dB for Keith's study was 5.3,. Webb and Greenberg's study yielded a mean SEN of 3.1 across MEIRs of -24, -20, and -15 dB. Goldschmidt's data produced mean SENS ranging from 6.1 to 6.6 for her six different single-talker carpeting message conditions . Finally, the mean SEN calculated for Beattie and 120 Clark's data across MCRS which produced scores fran 20-80% correct (i.e., linear portion‘of psycharetric functial) was 3.9. Dispersion data from this last study have particular relevance to the present study since both employed a continuous spectra carpeting speech message. The mean SEN of3.2 calculated from DCRS within the linear portion of the psycharetric function for the MI‘ 8 carpeting message in the present study reflects less variance than Beattie and Clark's data under similar carpeting message conditions (continuous speech spectra). This reduced variance may be largely due to the use of a 25-item test in the present study rather than the ten item test employed by Beattie and Clark. According to Thornton and Raffin, the 95% confidence interval for a ten item test is significantly wider than the carfidence interval of a 25—item test at any mean percent correct value. In other words, the 25—item test exhibits more measurement precision than the ten item test. More precise measurements produce less test variance . Test-retest correlation coefficients in the present study were sporadic, ranging frarr -.33 to +. 84 across six carpeting message carditions with a mean correlation coefficient of +. 31. Tea-retest measurements were taken at the MIR value which yields the closest to 50% correct score for each subject. Test-retest correlation coefficients obtained for a small number of subjects (N = 3) in Goldschmidt's study also varied significantly across six carpeting message conditials at IVERS of -12 and -18 dB. Correlatial coefficients ranged fram -.76 to +.56 with a mean of .0 at -12 dB IVER and fran -.70 to +.99 with a mean of +.60 at -.18 dB MCR. Correlation coefficients obtained at the -24 dB MIR exhibited a smaller range (+.28 to +.99) across carpeting 121 message conditions and overall revealed high reliability with a mean of +.84. These data indicate that Goldschmidt obtained consistently high reliability only at the most difficult IVER. One goal of this study was to reduce the variability of the SSI test. Examination of dispersion data (reflected by SENS) indicates that score variance across subjects was reduced carpared to previous SSI tests. Assessment of intra-list reliability suggests nothing in the experimental procedure systematically affected odd or even item performance. The status of test-retest reliability remains unclear, however. High test-retest reliability across experimental conditions either did not occur or was obscured by the method used to estimate it. Psycharetric Function Slope Values A main objective of this study was to determine what effect (if any) semantic content has on SSI psycharetric function slope values. Speaks, Karmen, and Benitez (1967) flattened the SSI psycharetric function slope by using a single-talker carpeting message and attributed the effect to randan masking (temporal discartinuity) and disruptive features (semantic cartent) inherent in the single- talker stimulus. The present study isolated the semantic content factor by cartrolling for terrporal discartinuity and formrd no semantic content effect on SSI psycharetric function slope values for any of the experimental conditions , except the single-ground carpeting message condition (MI‘ 8). Here, the mean psycharetric function slope of the mrulti-talker speech carpetitor indicated a steeper rather than flatter slope than its matched noise counterpart. It is questionable whether the mrulti-talker carpeting message slope was actually steeper than the 122 noise spectrum because of differences in the number of MRS used to estimate the slopes for the respective competitors. It appears, however, that the speech stimulus (semantic cartent) slope was definitely not flatter than the Slope of its associated noise stirmlus (no semantic content). Data from this study also corroborate findings of several other studies concerning the effect of carpeting mesSage semantic content on SSI psycharetric function slopes ._ Dirks and Bower (1969) found slopes of 4.0%/dB and 4.3%/dB for their single-talker carpeting message forward (CMF) and single-talker carpeting message backward (048) carditions. Tramrrel and Speaks (1970) discovered similar single- talker (NF and single-talker CMB slopes of 3.0%/dB and 2.4%/dB, respectively. Goldschmidt (1979) found a steeper (NF slope (3.8%/dB) than CMB (l.5%/dB) occurred between MIR values of -18 and --24 dB. The present study also lends support to a study by Miller (1947) in which psycharetric function slopes were obtained using a single-word target stimulus in the presence of various carpeting messages. Inspection of Miller's data reveals a multi—talker speech stimulus (eight voices) produced a slightly steeper psycharetric function than a continuous spectrum masking noise . Psycharetric function slopes for the multi-talker speech carpeting message and noise carpeting message were approximately 3.6%/dB and 3.0%/dB, respectively. The psychometric function slopes determined in the present study range frarr 7.0%/dB to 8.5%/dB across carpetitor conditions and were much steeper than slopes obtained in the other studies using the single-talker carpeting message. The steeper slopes in the present study probably resulted from the continuous spectra of the carpeting 123 message. That is, carpeting messages exhibiting temporal discartinuity (i.e., single-talker) . Data fran Martin and Mussel (1979) support this contention. They found psycharetric function slopes of approxi- mately 4.6%/dB and 8.5%/dB for a single-talker carpeting message and a single-talker carpeting message mixed with white noise , respectively. The white noise added to the single-talker eliminated temporal dis- continuity and increased psycharetric function slope steepness . Webb and Greenberg (1980) noted their SSI psycharetric function slope ' obtained with a four-talker competing message rose ll%/dB over the linear portion of the psycharetric function. In summary, the present study and several others support the contention that most (if not all) of the single-talker carpeting message flattening effect a: the SSI psycharetric functiar slope observed by Speaks, Karmen, and Benitez (1967) was caused by the temporal discartinuity factor (randarr masking) alone . Although it may generally be accepted that carpeting message semantic content does not affect the shape of the SSI psycharetric function slope, the questial of whether semantic content has an effect upon the location of the psycharetric function along the abscissa is less well resolved. The present study sought to answer this question by comparing the masking efficiency of various carpeting messages. Fifty-Percent Masking Efficiency levels The effect of carpeting message semantic content on masking efficiency was determined by analyzing the 50% masking efficiency levels of various carpeting message carparisons. One comparisar examined carpeting message semantic content effect on 50% masking efficiarcy 124 levels in a single-ground, continuous spectrum carpeting message. Four other carparisars investigated carpeting message Semantic content effect a1 50% masking efficiency levels using various paris of fore- gorund-background carpetitors... The background of theSe carpetitors consisted of either mrulti-talker speech or multi-talker speech spectrum noise (modulated by the MI‘ 8), whereas the foreground was either single-talker speeCh or single-talker speech spectrum noise (modulated by the single-talker) .- The relative amplitude level of the temporally discartinuous foreground in all foreground-background carpeting messages was 10 dB higher than that of the temporally cartinuous background, causing a 10 dB acoustic (perceptual) "window" during the foreground off-time. A final two analyses examined the effect of manipulating foreground semantic content status in temporally continuous speech and temporally continuous noiSe backgrounds , respectively. Several carparisons produced significant results. Single-Gromd Semantic Content. The MI‘ 8 and DEN MI‘ carpeting message carparison allowed examinatiar of semantic content effect on 50% masking efficiency levels for single—ground temporally continuous carpetitors. The carpeting message stimuli in this carparisar had equivalent acoustic parameters , which isolated the semantic content factor. Mean 50% masking efficiency levels of -11.4 dB and -14 dB for the MI‘ 8 and DEN MI‘ carpeting messages respectively were found to be significantly different. Thus, semantic cartent caused an excess in masking efficiency (perceptual masking) of 2.16 dB. The existence of perceptual masking secondary to serrantic content has been obServed in 13:00 studies (Carhart, Tillman, and Greetis, 1969; 125 Carhart, Johnson, and Goodman, 1975) that have particular relevance to the present study. Carhart, Tillman, and Greetis (1969') investigated semantic cartent in continuous spectrum maskers using speech (single-talker) mixed with white noise and white noise alone as carpeting messages. Spardee thresholds were obtained in the presence of both carpetitors. They found 3.2 dB excess masking (perceptual masking) with the speeCh and white noiSe carpeting message carpared to the white noise ally carpeting message. The addition of a secard talker to the white noise and single-talker carpeting message increased perceptual masking another 3.4 dB for a total of 6.6 dB excess masking. Carhart, Johnson, and Goodman (1975) also carpared spa'rdee thresholds obtained in the presence of (1) steady state speech spectrurm noise 2(2) speech spectrum noise modulated by seven talker combinatiars, and (3) the seven talker carbination (l, 2, 3, 16, 32, 64, and 128 voices). They discovered excesses in masking efficiency for all talker carbinations relative to their respective associated modulated noise conditions. Masking efficiency increased to a maximum of 9.8 dB for the three-talker modulatednoise speech spectrum, decreasing thereafter to stabilize at 3 dB with 64 talkers. Collectively, these Studies indicate (1) semantic cartent causes perceptual masking (excess masking) when terrporal discontinuity is either eliminated or controlled for, (2) perceptual masking increases with talker number until three talkers, and (3) Perceptual masking decreases when four or more talkers are included in a carpeting message. These findings agree with the findings of the present study that serrantic content causes perceptual masking and also suggest that the amount of perceptual masking discovered in the present experiment may 126 have been greater if fewer talkers had been used in the mrulti-talker speech background (i.e., MI‘ 3 instead of MI‘ 8) . Apparently, the inclusion of four or more talkers in a carpeting message reduces the intelligibility and therefore semantic content effect. Whereas one or two talkers in a carpeting message still allow the listener to take advantage of the peripheral "window" effect, four or more talkers in a carpeting message cause a reduction in the Semantic content and a subsequent release fram perceptual Evidently. as carpeting message plmemic images impinging upm the auditory nervous systerm becare less distinct, (i.e., less like meaningful speech), the target speech stimuli becare perceptually more distinct. This notion is supported by data from a study by Young, Parker, and Carhart (1975) which assessed digit recognition performance in the presence of forward speech (two and four talkers) , backward speech (same two and four talkers) and speech modulated (same two and four talkers) speech spectrum noise. They found forward speech to be the most efficient masker, and backward speech Secard in masking efficiency, and speech modulated noise the least efficient masker. Again, as the masking agent became less speech-like, masking efficiency was reduced. It could be that perceptual masking which occurs at the central level operates a1 a principle similar to physical masking at the peripheral level . According to the critical band theory, the most efficient masker of any particular tone is a narrcwband of frequencies centered around that tone. If the narrow-band masker and target tone are analyzed in the Same critical band alarg the basilar membrane of 127 the cochlea, masking efficiency will be increased carpared to a situation where the masker and the target tale are analyzed in separate critical bands . Masking interference is greater when the target and masker stimuli are analyzed by common physical structures. Likewise, an excess in perceptual masking may occur at the central level when target and masker Manic stimuli are mediated by carrron processes . Experimental results reported by Triesman (1964) also have relevance to the MI‘ 8 versus DEN MI‘ carparison. Subjects in the Triesrran study shadowed narrative extracts from a novel read in English by a female talker in the presenCe of several different carpeting messages . Subjects received a binaural presentatiar (presumably, diotic) of target and carpeting message stimruli. Carpeting message conditions pertinent to the present study include (1) same-voice, forward English speech, (2) same-voice, forward foreign speech, (3) same-voice, reversed speech and (4) different- voice (male), forward English speech. Treisrran found (1) the lowest scores were obtained using the same-voice, forward English speeCh carpeting message, (2) intemediate scores were obtained using the Same-voice , foreign speech and same-voice reverse speech carpeting messages, and (3) the highest scores occurred with the different-voice, forWard English speech carpeting message . Triesman explained her findings that subjects performed poorer with the same-voice backward or same-voice foreign speech carpeting message than with different- voice (male) forward speech by suggesting that subjects were able to reject a carpeting message differing in physical features (man's voice) from the primary message (woman's voice) at an early stage in a perceptual analysis and thus better attend to the primary message.” If 128 this early rejection on a physical basis cannot be made, then the ability to reject the carpeting message apparently depends on the degree of semantic cmtent (meaning) in the carpeting message. Thus, the same-voice forward speech was much harder to reject than the same- voice foreign speech or same-voice backward speech. Based on Triesman's findings, it would appear that subjects in the present study performed better with the DEN MI‘ carpeting message than the MI‘ 8 carpeting message because the former carpeting message (noise) was perceptually less like the target stimuli than was the latter carpeting message (speech). This allowed the subjects to more effectively reject the noise stimulus than the speech stimulus and better attend to the target sentenCes. Since the DEN MP and MI‘ 8 had very similar physical characteristics (i.e. , spectral and temporal characteristics) any difference in the ability of the auditory systerm to reject either stilmllus probably did not occur at tire most peripheral level (cochlea) .. It seems likely that differential rejection occurred later (more central) in the perceptual process ar the basis of differences in the amount of psychological impact created by the two stimruli. The MI‘ 8 carpeting message was harder to reject than the noise carpeting message because it caused psychological as well as physical interference. It is assured that this occurred because the incoming speech carpeting message and target stimuli were analyzed by carrron neurological structures and processes. Norman and Bobrcw (1975) present a theoretical View of data- limited versus resourm-limited psychological processing that has particular relevance to the present study. They define data-limited 129 processes as being constrained by input signal characteristics (e.g. , physical masking) and rescurce-limited procesSes as being limited by available psychological resources (e.g. , perceptual masking, attention, etc.) . In a data-limited process performance is largely dependent upon the spectral and terrporal characteristics of the input signal, whereas in a resource-limited process performance depends a1 the amount of psychological resources that can be allocated to a given task. It is important to note that resource- 1imited processes are not totally free fran data-limitations because the peripheral mediation of the signal must occur prior to any central mediation. Performance on a data-limited task can be reduced by degradation of signal quality. Performance on a resource-limited task can be reduced by additional carpetition for limited psychological resources (auditory processing capabilities, linguistic knowledge, attention, etc.) . Since the MI‘ 8 and DEN MT carpeting messages are equivalent in the data sense, the more efficient masking ability of the MI‘ 8 carpeting message must be due to additional carpetition for psychological resources. Thus, the speech carpeting message interferes with the processing of the primary speech signal a1 a physical. as well as psychological basis. Semantic 'Cartent in 'Foregrormd-tBackgromrd Competitors . Three paired carpeting message carparisals were analyzed to assess the effect of serantic cartent position within foreground-background carpeting messages ar 50% masking efficiency levels. A fourth foreground- background competing message paired carparisar allowed assessrrent of the effect of serantic content position and type on' 50% masking 130 efficiency levels. A separate discussion of each comparison follows. The MI‘ 8 + DEN 1 versus DEN MI‘ + DEN 1 comparison investigated serantic content effect in the backgrormd only of a foreground- background competing message. Both carpeting messages had a terporally discontinuous foreground void of serantic content, but differed in the presence versus absence of semantic cartent in a terporally cartinuous background. The MT 8 + DEN 1 carpeting message (background semantic cmtent) produced 2 . 8 dB excess masking carpared to the foregrormd—background carpeting message void of serantic content in both grounds. Both carpeting messages exhibited a 10 dB acoustic (perceptual) "window" during the foreground off-time. The results suggest the subjects were better able to utilize the 10 dB acoustic windows available in the foreground-background carpeting messages when semantic contert was absent in both grounds. The DEN MI‘ + 1 versus DEN MT + DEN l comparisar assessed the effects of the presence versus absence of foreground serantic content in foreground- background carpeting messages featuring terporally continuous noise background. Interestingly, it was discovered that semantic content in a terporally discontinuous foreground (single-talker) did not produce the same amount of excess masking observed with the terporally continuous background serantic content in the previous carparison. The DEN MI‘ + 1 produced only a negligible amount of excess masking (1.0 dB) , which was not statistically significant. Evidently, the terporal discartinuity (acoustic windows) of the single—talker in the DEN MI‘ + 1 carpeting message negated any excess in masking 131 secondary to serantic cartent. In other words, when the serantic cartent is terporally discontinuous in nature it (lies not manifest the same degree of masking efficiency as terporally continuous serantic cartent. Perhaps this last carparisa'r explains the findings of Dirks and Bower (1968) that fore-ward single-talker and backward single-talker carpeting messages were equivalert in masking ability when the former carpeting message caltained semantic cartent and the latter did not. In the Dirks and Bower study the acoustic windows existing in the single—talker carpeting message probably were about 40 dB in depth (i.e. , from single-talker peak amplitude to equipment noise floor level), carpared to acoustic windows 10 dB in depth in the present study (i.e., fram foreground peak arrplitude to background average amplitude) . However, both studies indicate that terporal discontinuity of the serantic cartent carponent create perceptual windows that void any excess masking due to serantic cartent. It is also possible that serantic content effect may have been nullified in the Dirks and Bauer study by the intensive subject training prowdure. Whereas foreground terporally discontinuous carpeting message serantic cartelt (DEN MI‘ + 1) caused a negligible amount of perceptual masking and background temporally continuous carpeting message serantic cartelt (MI‘ 8 + DEN 1) caused a moderate amount of perceptual masking relative to their respective noise counterparts , serantic cartent presentinbothgramds ofacarpetingmessage (MI‘8+ l) produceda remarkable 8.9 dB of perceptual masking carpared to its noise counterpart void of serantic contert in both grounds (DEN MI‘ + DEN l) . Theamomtofperceptualmaskingthtoccurredwiththem8+l 132 competing message is especially noteworthy when carpared to the amount ofperceptual maskingproducedbythebfl‘8+DENlandDENMI‘+l carpetitors (2.8 dB and 1.0 dB, respectively). All three carpetitors caitained serantic content and a 10 dB acoustic window separating the foreground and background. However, when serantic content appeared in both grounds, subjects were unable to utilize the temporal discontinuity factor inherent in the single-talker foreground to negate or partially negate semantic cartent effect. This last statement is supported by the finding that mean 50% masking efficiency levels of the foreground- background MI‘ 8 + l and single-ground MI‘ 8 carpeting messages were nearly identical (—ll.6 dB and -ll.4 dB, respectively). Serantic cartent was present in both carpetitors while acoustic windavs were present in the former and absent in the latter. However, the carpeting mesSages produced essientially equivalent amounts of perceptual masking. Apparently, when competing message serantic cartent exists in both . grounds of a foreground-background carpeting message, the size of the windavs must be increased before the central auditory nervous system is able to gain a perceptual advantage (i.e., negate serantic content effect). Presumably, this advantage would be realized using deeper windows (larger difference between relative amplitudes of foreground and backgrormd) "wider" windows (increased off-time of foreground), or both. Several studies have measured the effect of changing carpeting message acoustic window size on word or sentence recognitiar. Miller (1947) varied the al-off time of cartinuous spectrum noise from 20% or to 100% or and found increasingly reduced perforrrance on PB words as the noise arr-time becarre greater. Martin and Mussel (1979) 133 eliminated acoustic windows in a single-talker carpeting message by adding white noise and found increased masking efficiency. They used synthetic sentenCes as the" primary stimuli. Carhart, Tillman, and Greetis (1968) obtained spardee thresholds using a single-talker carpeting message, a single-talker carpeting message mixed with modulated noise, and a two-talker carpeting message. Spondee thresholds were 5.1 dB poorer when the noise was added to the single-talker and 8.0 dB poorer when the two-talker carplex was used. This finding indicates that reduction of acoustic window size alare accounts for sare increased masking efficiency, havever, the effect is even stronger if serantic cartent is increased in caljunctiar with decreaSed acoustic windav size. This supports the finding of the present study that the MI‘ 8 + l is a more efficient masker than any of the other foreground-background carpeting messages. The size of the acoustic windows in all four carpeting messages is equivalent, haoever, the MI' 8 + 1 has more semantic content. Another explanatiar of the remarkable masking efficiercy of the MI‘ 8 + 1 carpeting message carpared to the other foreground- backgrormrd carpetitors is based on Norman and Bobrow's (1975) accormt of data-limited versus resource-limited processes. All foreground-background carpeting messages exhibit similar spectral and terporal characteristics, but differ in the amount and position of serantic cartent. By definitiar, spectral and terporal character- istics qualify as data-limited factors while serantic cartent qualifies as a rescurce-limited factor. The carpeting messages are equivalent in the data sense, but differ in their degrees of carpetition for 134 available psychological resources. Apparently, the existence of serantic content in both grounds of a foreground-background carpeting message creates significantly more carpetition for available psychological resources than the non-existence of serantic cantent or the existence of serantic content in just are ground. Thus, the MI‘ 8 + 1 carpeting message is inherently a more efficient masker than the other foregrormd-background carpeting messages because it causes an equivalent amount of data interference in addition to an increased amount of resource interference The last foreground-background carpeting message pair allaved examination of the effect of serantic content position and type an 50% masking efficiency levels. The MI‘ 8 + DEN 1 carpetitor consists of a terporally continuous speech background and a terporally discontinuous noise foreground while the DEN MI‘ + l carpeting message is comprised of a terporally continuous noise backgromnd and a terporally discontinuous speech foreground. It was discovered that the MI‘ 8 + DEN l carpeting message produced 1.8 dB more perceptual masking than the DEN MI‘ + 1 carpeting message. Again, the key to increased masking efficiency is related to the terporal status of the semantic content. When serantic content is terporally continuous , masking efficiency is greater than when serantic content is terporally discartinuous. This is true even though the terporally discontinuous foreground serantic cartent was 10 dB higher (louder) than the terporally continuous background serantic cartent in the respective carpetitors. ' The Effect of Changing Foreground Serantic Content Status. Tm carparisons examine the effect of changing foreground status in 135 carpeting messages with terporally cantinuous speech and noise background, respectively. Manipulation of foreground status included (1) adding a single-talker to the terporally continuous background (2) adding a single-talker speech modulated noise (DEN l) to the terporally continuous background, and (3) elimination of the foreground. In the last condition the carpetitors actually became single-ground stimuli. The first analysis revealed significant differences in the mean 50% masking efficiency levels obtained using MI' 8, MI‘ 8 + l, and MI‘ 8 + DEN l carpeting messages. Subsequent use of Tukey's test for specific carparisons determined the MD 8 + DEN l carpeting message mean 50% masking efficiency level (-l7.7 dB) differed from MI‘ 8 and MI‘ 8 + 1 carpeting messages mean 50% masking efficiency levels, but levels of the latter two did not differ fram each other (11.6 dB and 11.4 dB, respectively). Again, this finding confirms earlier results indicating subject inability to effectively utilize the 10 dB acoustic window in a foreground-backgrormd carpeting message when semantic content appears in both grounds . TheDENMI‘, DENMI‘+landDENMI‘+DENlcarparisar also revealed significant differences in mean 50% masking efficiency levels among carpetitors. Tukey's test of specific carparisars found masking efficiency of the DEN MI‘ 9-l4.0 dB) differed from DEN MI‘ + 1 and DEN MI‘ + DEN 1 masking efficiency levels, whereas masking efficiency of the latter two were similar (-l9.5 dB and -20.5 dB, respectively). Here , terporal characteristics of the masking spectrum appear to account for increased masking efficiency of the DEN MI‘ carpeting message. Interestingly, the terporally cantinuous noise spectrurm void 136 of serantic cantent is a more efficient masker than the spectrum consisting of the same noise Stirmlus (DEN MI‘) used in a background mode and coupled to a terporally discantinuous single-talker fore- ground. In the first instance, the terporal consistency and frequency composition of the masking spectrum more adversely affects central auditory nervous system processing,while in the latter instance the perceptual advantage gained by the 10 dB acoustic window manifests itself as Serantic content occurs only in a terporally discontinuous foreground. These findings provide additional evidence that serantic content must appear in both grounds of a foreground- background carpeting message to eliminate the acoustic window effect. CLINICAL D’IPLICATIONS The traditional SSI test has already been used as a clinical tool in the differential diagnosis of central auditory nervous system (CANS) disorders (Jerger and Jerger, 1974; Speaks, 1975; and Jerger and Jerger, 1975a and b). The SSI test material and carpeting messages used in the present study represent an improverent over the traditional 10 item test and the single-talker carpeting message. The 25—item test and continuous spectra stimuli served to reduce subject variability and increaSe test reliability. The eight alternative lists may serve to reduce learning effects associated with several administrations of the test. Evidence from the present study suggests the MT 8 and MI‘ 8 + l carpeting message cause more perceptual masking than the traditional single-talker carpeting message and thus may be more effective in identifying pathological conditions within the CANS. Pathological 137 conditions may result from the aging process, stroke, space occupying lesions, or trauma. The SSI test used in the present study may be effective in identifying these pathologies and in monitoring improvement in central auditory processing abilities following any medical intervention. For stroke and traurra victims , periodic testing throughout the therapeutic process may provide prognostic as well as diagnostic information. Pre-operative and post-operative testing of patients undergoing surgery to rerove tumors within the CANS could provide inforrmation regarding the success of the operation. The test could be applied to any literate population capable of understanding the task. The response mode (circling the correct item) could be modified for individuals with handicapping conditions that preclude marking ability.) Test results should provide the clinician with reliable and valid indicators of an individual's auditory processing skills. The SSI test used in this study in the present forrm would probably not be useful in testing young learning disabled (ID) children. Although use of the multi-talker carpeting message would be very appropriate in identifying a weak or underdeveloped CANS , the synthetic sentences may be too difficult for the ID children to comprehend and recall. The identificatian of a synthetic sentence an the response sheet requires reading skills ID children may not have. Relatively easy to understand and recall primary stimuli such as sentences used by Willeford (1977) or Central Institute for the Deaf everyday sertences may be more appropriate for this clinical population. It would also be necessary to establish chrarOlogical age, group means for children with normal auditory processing skills before an adapted 138 versian of the present 881 test would be useful in identifying children with poor auditory skills. IMPLICATIONS FOR FUTURE RESEARCH Before the present SSI test can be erployed with an adult clinical population, a clinical test procedure must be defined. Additional research is needed to determine which carpeting messages and NCRS most effectively differentiate between normal and impaired CANSs.‘ Presurrably, the MI‘ 8 carpeting message presented at a MCR which normal-hearing subjects score 70-80% (-8 dB) would provide the most useful diagnostic information. Extremely eaSy or extrerely difficult NCES would render the SSI test insensitive to the identification of pathological CANSS. Easy IVCRS would not tax even an impaired system while difficult MCES cause lav performance scores for normal neurological systems. Harever, this needs to be verified.) Another possible clinical procedure would involve adjusting the MCR until a 50% correct response rate is obtained. This adaptive procedure was used by Dirks, Morgan, and Dubno (1982) to determine the MIR at which normal and hearing-impaired listeners obtained 50% correct. The adaptive procedure effectively differentiated between normal and hearing-impaired subjects' ability to recognize spondees and monosyllables in the presence 0f speech babble. The adaptive method may also be effective in quickly identifying impaired CANSs. Normative as well as clinical (pathological) data must then be collected using tlne defined clinical testing procedure. The MI‘ 8 carpeting message used in the present study could also be used in the auditory assessment of ID children. However, as noted 139 in the. previous section, this would require development of alternative primary stimuli and scoring procedures. Again, it would be necessary to develop a clinical procedure that differentiates between children with normal auditory processing abilities and children with redmd auditory processing abilities. It would also be necessary to collect data an normal and ID children. Future research might deal with more experimental issues such as the size of the acoustic window in the foreground-background carpeting messages, especially the MI‘ 8 + l carpeting message. It may be of interest to vary the depth or Width- of the acoustic window in this competitor and measure the effect.) For example, the single-talker carpeting message (foreground) could be time—expanded in specified amounts for several experimental conditions. Time-expansion of the single-talker carpeting message would also expand the width of acoustic windows. It would then be possible to determine how much . expansion (in percent) is necessary before normal CANSS can success- fully negate the excessive semantic content effect inherent in the MT8 +1carpetingmessage.l(i.e., MI‘ 8+lscores=DENMI+DEN1 scores at same IVER) . The length of the acoustic windows could also be increased by reducing the single—talker speaking rate. The effect of increasing acoustic window depth could be Studied ' by systeratically laaering the background (multi-talker) amplitude. Another focus of future research might entail changing the amount of serantic content in either the carpeting messages or the primary stimuli. Carpeting message serantic content could be varied by making the Carpeting mesSage more or less intelligible (increasing or decreaSing talker number) or by stringing together serantically 140 unrelated phrases or sentences in the carpetitor. Synthetic sentence serantic cartent could be manipulated by using successive order approximations to real sentences (i.e., first order,“ second order, third order, etc.). The SSI test developed for this study could be applied in several clinical or experimental endeavors. Various aspects of the test may need to be modified to accarmodate different clinical populations or to answer different experimental questions. However, use of the 25-item four-alternative forced-choice response paradigm, cartinuous spectra carpeting messages , and precisely calibrated acoustic stimuli should serve to reduce subject variability and increase test reliability in any application. CHAPTERV SUM/JARS! AND CONCLUSIONS INTRODUCI‘ICN The SSI test allows assessment of an individuals ability to identify syntactically constrained sentential stimuli without the problems inherent in the use of real sentence stimuli (Speaks and Jerger, 1965). The traditional SSI test includes use of a ten- alternative closed message set and a single-talker competing message. Several studies (Dirks and.Bower, 1969; Garstecki and Mulac, 1974; Trammell and Speaks, 1978; and Goldsdhmidt, 1979) have investigated the effect of competing message semantic content on SSI performance by carparing carpeting message foreward and carpeting message backward scores and found conflicting results. Several methodological differences which.may have accounted for the discrepancy in results include differences in test stimuli, levels of subject sophistication, response modes, and.presentation:modes. Another reason for conflicting results may have been the high level of test variability reported in Goldschmidt's study. High test variability may be due to the use of a small ten~item.test (Thcrtcn and Raffin, 1978; and Raffin and Shafer, 1980) and theatemporal discontinuity inherent in the single-talker carpeting message. Several studies (Miller, 1947; Carhart, Tillman, and Greetis, 1968; 1969; Speaks,‘Wigginton, and Germono, 1971; Carhart, JChnson, and.Goodman, 1975; Y0ung, Parker, and carhart, 1975; and 142 Martin and Mussel, 1979) have reduced or eliminated competing message terporal discantinuity and found a clear semantic cantent effect a1 speech intelligibility scores using a variety of primary stimuli and carpeting messages. No are has studied the effect of carpeting message semantic cantent an SSI psycharetric function slopes or 50% masking efficiency levels using acoustically identical speech and noise carpetitors. Also, no are has examined serantic content effect in foreground-background carpeting messages . PURPOSE The purpose of this study was to obtain SSI performance scores for normal-hearing subjects as a function of MCR for six different carpeting messages . TWO dependent variables , psycharetric function slope values and 50% masking efficiency levels, were interpolated from individual subject data. These dependent variables allowed direct carparison of the six carpeting messages an SSI performance. A 25—item test was employed to reduce subject variance. EXPERIMENTAL DESIGN Subjects Thirty-six normal-hearing adult subjects were used in this study. Each subject displayed normal sensitivity for pure tones, spondees, acoustic reflexes, as well as normal reflex decay, middle ear function, and word discrimination scores. Twelve subjects were randomly assigned to each of three experimental groups. Stimuli Synthetic Sentences . Three hundred third-order synthetic sentences were generated for this study. One hundred and four synthetic 143 sentences met linguistic criteria employed by a three-judge panel. These synthetic sentences were recorded and measured for duration and relative amplitude. Twenty-seven synthetic sentences had durations within 100 m/sec of the mean duration and relative amplitudes within 0.5 dB of the mean relative sentence level. Twenty-five of these senterces were randomly chosen and used for target sentences . Different random orders of the 25 sentences were re—recorded on each of eight cassette test tapes. The 77 synthetic sentences that met the ‘ linguistic criteria, but did not meet the duration or amplitude criteria were used as distractor sentences in eight different SSI lists. Each of the eight lists exhibited a target sentence order that corresponded to the presentation order on one of the eight cassette tapes . Three distractor synthetic sentences were randomly combined with one target synthetic sentence for each of the 25-iters an all eight lists. No two sentences within a four-alternative test item could begin or end with the same word or phareme. Target senterce position was counterbalanced within each test item for all eight lists. Carpeting Passages. This study employed two classes of carpeting messages; single-ground and foreground-background. TKwo carpeting messages were used to determine serantic contert in continuous spectrurm single-ground stirmrli. Four other carpeting messages allowed examination of semantic content effect in foreground-background carpeting messages having a terporally cantinuous background and a terporally variable foreground. The single-ground carpeting messages were (1) multi-talker speech consisting of eight equally intense talkers and (2) speech spectrum noise derived from and modulated by the mrulti-talker stimulus. The foreground-background carpeting messages 144 included (1) the multi-talker background coupled with a Single-talker foreground, (2) the multi-talker speech modulated noise background plus the single-talker speech modulated noise foreground, (3) the multi- talker background with the speech modulated noise foreground, and ( 4) the mrulti-talker speech modulated noise background carbined with the Single-talker foreground. A 10 dB differenCe in relative amplitude separated the background and foreground in these competing messages , thus creating a 10 dB acoustic window during foreground off-time. The two single-ground competitors had nearly identical physical character- istics while the four foreground-background carpeting messages were alwo nearly identical to each other in physical characteristics. This allaaed isolation of the serantic content variable across carpeting message conditions . PROCEDURE The experimental procedure for each subject in this study included (1) signing a release-consent form, (2) undergoing a hearing screening, (3) participation in a practice session, and (4) completion of the main experiment. During the practice session, each subject listened to five non-experimental synthetic sentences presented to the test ear at 40 dB SL in quiet to gain experience with the four-alternative forced- choice test paradigrm. Eighteen other non—experimental synthetic sentences were then presented at 40 dB SL with the carpeting message adjusted to obtain a -2 dB MCE to provide the subjects with nine listening experiences with each of the two carpeting message conditions according to group assignment. The subjects task for each of the 23 145 items was to listen for the presentation and choose the correct answer an the response sheet. All subjects scored 100% on the practice test. In the experimental session, each subject listened monaurally to a 25-item test list at each of four MCES for both carpeting messages. The target stimuli were presented at a constant 40 dB SL and the intensity levels of the carpeting messages were varied. Again, the subject's task was to listen to each presentation and circle the correct answer on the response sheet. Retest data were obtained for both competing messages at the MCR which produced the nearest to 50% correct score for each srbject. FINDINGS The experimental findings of the present study are as follavs: 1. Mean monaural SSI psycharetric function slope values obtained with normal listeners differed as a function of the presence versus absence of serantic content in a single-ground ipsilateral carpeting message. Mean psycharetric function slope values were 8.5%/dB and 7.0%/dB for the MI‘ 8 and DEN MI‘ carpeting messages, respectively. However, this staterent must be qualified by two considerations. First, the strength of associatian measure indicated only a trivial portion of the variance in slope could be attributed to the semantic content factor. Second, carputation of the DEN MI‘ slope was performed using three MCES while the MI‘ 8 slope was determined using four IVCRS. Psychometric function slope values computed using mean scores at three corresponding MCES revealed similar slopes (7.5%/dB for MI‘ 8, 7.0%/dB 146 for DEN MT) . 2. Mean monaural SSI psychometric function slopes obtained with normal-hearing listeners did not differ as a functiar of differences in foreground-backgrormd ipsilateral carpeting messages containing semantic cantent in: a. Background anly; b. Foreground ally, and c. Background and foreground. 3. Mean monaural SSI psycharetric function slopes obtained with normal-hearing listeners did not differ as a function of serantic contert positiar and type in a foreground-backgrormd ipsilateral carpeting message (i.e. , terporally continuous background serantic content versus terporally variable foreground serantic content) . 4. Nban monaural SSI psycharetric function slopes obtaired with normal-hearing listeers did not differ as a function of differences in foregrormd status in ipsilateral carpeting messages cantaining a speech background. 5 . Mean monaural SSI psycharetric function slopes obtained with normal-hearing listeners did not differ as a function of differences in foreground status in ipsilateral carpeting messages cantaining a noise background. 6. Mean monaural 50% masking efficiency levels obtained with normal-hearing listeners differed as a function of the presence versus absence of serantic cartent in a single—ground ipsilateral carpeting message. Masking efficiency was 2.6 dB greater when semantic content was present. 147 7. Mean monaural SSI 50% masking efficiency levels obtained with normal-hearing listeners differed as a function of differences in foreground-background ipsilateral carpeting messages containing semantic content in the background only and both foreground and background. Background-foreground carpeting messages containing semantic content were more efficient maskers . Mean values did not differ as a function of semantic content in the foreground only. 8. man monaural SSI 50% masking efficiency levels obtained with normal-hearing listeners differed as a function of semantic content position and type in a foreground-background ipsilateral carpeting message. Masking efficiency was greater for the temporally cantinuous background semantic content carpeting message . 9. Mean monaural SSI 50% masking efficiency levels obtained with normal-hearing listeners differed as a function of differences in foreground status in an ipsilateral carpeting message containing a speech background. The noise foreground competing message exhibited reduced masking efficiency carpared to the single-ground carpeting message and the speech foreground carpeting message. The latter two were similar in masking efficiency. 10. Mean monaural SSI 50% masking efficiency levels obtained with nonral-hearing listeners differed as a function of differences in foreground status in an ipsilateral carpeting message containing a noise background. Greater masking efficiency was obtained with the single-ground competing message than with carpeting messages containing either a noise or speech foreground. The latter two were similar in masking efficiency. 148 ll. Reliability (consistency) within experimental conditions was sporadic for the various carpeting messages as a functian of MCR, using the odd vs. even procedure. The attainment of high reliability with this method depends on cansistency at the individual subject level. Application of the binanial test yielded these scores which indicated equivalency between odd and even item performance for every experimental conditian at the group level. The binomial test takes advantage of the "averaging out" that occurs at the group level and is not sensitive to odd and even item performance of individual subjects. Standard error of the mean values for the six carpeting messages all occurred within a narrow range indicating consistency across experimental conditions . 12. Test-retest performance scores were not consistent for any of the carpeting conditions except MT 8 + 1. Poor test-retest reliability was probably related to selection of the retest MCR value (near 50%) at which performance inherently variable. It is also likely that learning effects cantributed to poor test-retest reliability. C(NCLUSIONS In addition to the above findings, the following conclusions are offered. 1. Generally, ipsilateral carpeting messages containing semantic cantent (speech) and acoustically similar ipsilateral carpeting messages void of semantic content (noise) have an equivalent effect on normal-listener SSI psycharetric function slope values. The various carpeting messages exhibited slope values within a narrow 149 range fram 7.0%/dB to 8.5%/dB. 2 . Competing messages containing semantic content caused physical masking at the peripheral level related to their spectral, amplitude, and temporal characteristics and perceptual masking at the central level related to their ability to interfere with psychological processing (i.e. , linguistic processing). 3. The perceptual advantage gained by acoustic windows in foreground—background competing messages is partially reduced when semantic cantent occurs in the temporally continuous background only and carpletely negated when semantic content occurs in both the temporally cantinucus background and temporally discontinuous foreground. However, the perceptual advantage remains when semantic content occurs only in the temporally discantinuous foreground. 4 . 'Ihe 25-item four-alternative forced-choice SSI test paradigm coupled with the use of a temporally continuous carpeting message served to increase the precision of measurement in the present study carpared to previous SSI studies. The present SSI test can be a useful clinical tool following collection of normative data. APPENDIX A APPENDIX A. 151 SUBJEXZT SCREENING FORM Project Name Sub. No. Name Age Examiner Date PURE TONE AlFt CON DUCTION ‘R + Pass HTL= 15dB L - Fail 250 500 1000 2000 4000 Coding Audiometer ACOUSTIC REFLEX _ 500 Hz 1000 Hz 2000 Hz R. Decay Test HL HL HL HL +Pass =100dB Tone L Fl Tone - Fail Test HL HL HL HL ARD+=10$ Audiometer TYMPANOMETRY Pressure , , mm H20 ”35,,“ pt (cc) + 200 +100 o -100 - 200 Compliance R Fl R Fl R R R cc L L L L L L Impedance Bridge SPEECH TESTING HISTORY SRT NU-6 ' - o R R + Pass = SRT Currently active URI. 15 dB HTL or Family History of less Hearing Loss? +Pass = 90 °/o . . . . Histo of Tinnitus L L or higher . or Verrtyigo? APPENDIX B 153 APPENDIX B. EIGHT SSI TEST LISTS a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. SSI LIST - FORM A FEW YARDS TILL HEALTH STORES WARM PRESSES IASI‘INBGDYI‘DVEMENT'IHESWEETMONEY SCENETODAYGKMSLGVGQUITETRUEWITI—IIN UNDER ICE FOR FISH ESCAPE BEYOND WAR TOOBJECTWILESPIECOUIDCLEARW MUST ACI‘ BUSY TUNDRRGN AND STOP FOR CLEARFRO’ICIIIJDSABOVEEAR'IHOPENSWIDE CHILDREN noo SPOKE LAST BECAUSE SEVEN MILES HEAVY LADIES THINK PRACTICE KEPT WHO DEALS MORE MESHADNEIERCHANGEDMEMBERS THAT WHILE UNDER NINE BIG MEETINGS BELONG KNCMN BEDS CAN'T KNCM MUSIC AS POST CHIEF HEAT SEEIVB DIFFERENT SINCE GOOD MANNERS CAN'T TAKE OUT WHOIE NUMBERS ALWAYS WENT AWAY CEASE CHANGE WEAR DRESSES WHILE LISTENING MADE EGGSGRQVWGCIIBEPPETNCIEANSNGVLES LIEANDNG'I‘I-IOSEWIIDCOUN'I‘RYGARDENBUIWED CENPSSGWETIMESKEEPBETWEENWIDEENCIEHSHOES PLEASURE SIMPIEFRIENDCIJI‘WHATALLNINE WATER FISH UNTILNEXTSET ISLIKE OVERBUILDINGWXLKHARDLIKEVDODCN CLOTHES ACIIIJNT NEARLY PAYS YCIJR BOY CANNOT ALTHOUGH GROUPSTALKLATEYE'I‘MOSTMAKE WEREPIEASANTWHIIEWARMDAYSACCCXJNTFOR FOR WHCM MOST SPARE MILLIONS SAFELY MTG-I GROUPS BE CCIVI’DN WHlIE UNDER STREAMS BUILDING CLEAN T'HRCIJG-I PAPER WHICH PRCIVLLSES ME GREAT ISCUI‘LASTWEEKSCDNGAVEPRACI'ICE ALREADY TCIDK CW WANTS OLD TRUTH KEPT WHIIE LESS CLOSE EXCEPT BEEN WHITE GARDENS PEACE SEEMS KNCMN SPIRITS ME]? LITI‘IE SINGING SEEMYOUNGHOTEVERSENDSTRAIGHTAWAYFRQI NEWS PIEASED EIGHT GIRLS UNDER BOXES PREPARE SHORT HUMAN HAIR FLIES GOING TILL THAN GUARDS SPRINGIMS WHILE HANGING WHERE WEIGHT TIES CDUNI‘RY LED WAR OF MUSIC DANCED ALL FORM A' 10. ll. 12. l3. 14. 15. l6. l7. 18. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. 154 (Cont.) STILL PLEASED SINCE C(Il’LLNG OUTSIDE STORES BUILD IVDRE HILLS LAY VILLAGES MILES W'T PART C(MPLETE HIS NDVEMENT ARRIVES NEAR SCHCDL (N GOVERNMENI‘POSTS YOURPAIRTOOKWALKSFOR DCN'TWRITELIKE STONEWILLGIVE PLEASURE WENT STRANGE QUICKLY MADE SEVEN TRAINS WON'T PEACE AROUND EARTH VDN'T CARE PAST DANCING EXPLAINS SCMETIHING WE WATCH OIL PROBLEMS FELL NORTH PAST A BANK REMAINS ABLE TO SAVE SOULS NEARLY LOST INSIDE FUIURE PAIR DAY SAVE LIES IN BRCMN (LILY ACCORDING C(IDKS HEAT ALIVDSI‘ TWO WORDS PREPARED BEFORE VVAITEASYLJNTIL'IAKENDDRESCIiCDLSOLHfP 'IHESEESCAPECIASSESINSTEADSAVETREESTILL CRONDSWLENMDNETURNS SUCH HAPPY CCMPANY 80w CAN WALK FAST EXCEPT BUILDING A LEGSCERIADIMSEXPRESSHERWICEGHIT ANKNGACROSSTHEBWSTOREICEIN DARESBUI‘SEEIECCMPLE‘IEANKXJNTSFEELCKDSSES MILES WHILE QUICKLY SING CUI'SIDE IN HCNOR ALTHOIH-IVDODACISGAINALLVIEWDIGFUI‘URE FOODSBUILD STRAIGHT TODAYSHOTSIXTREES EARIHCNSOHEATALIWSFANDUSHAT 'LAKEANY'IHINGABOVECLHJDSSEEBRIGHI‘WIIE ALTHCXKSHDEEPREDCOALSJUSTNEEDSWG SNELLTRIPS VlDN'TBUY BROKEN LIGHTNEXI‘ LAYSUCHFEARPAIDAHEAVYPURPwnE FELLEIGi'I‘ERDSSI—IGJLDTLBNSO’IETHJNGSMOKE IFBROKENMENADDWXTEREVENTO TIASI‘ESGREATDURINGDEATH RATHER BREAK METER SHIPSUNDERTALLMETHODSREACHABOVEREAL WARMDAYSMCNEWHILE FAVORS DON'TLE'I‘ CANNOT FIT SHCES FAST RATHER LOSE AFTER THOUG‘I WOKIH IIDLLARS IF SUCCESS MUST WT BEHIND BUT JOY OF PUBLIC LAND SOVIETD’IES SDICEGUARDSOLDIERSOFCONTROLKEEPBEAUI‘Y FORMA’ 19. 20. 21. 22. 23. 24. 25. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. 155 (Cont.) SAFE IF BETWEEN EARTH ITS WILD HCM ICE FINALLY SIOPPED EIGHT STATES VALUE OF UNDER NOBE BURNED FILE NEARLY SINCE COLD LDLLARSWARYHANDFREEHOIDYQIRPRETTY GAMES THAT BEING PAST YOUR VILLAGE FOLLON REAL GIRLS KEEP LESS CHANCE RESULTS CCNSIDER BROTHER PAID BRLGIT MUSIC DOUBLED IN SCHCIDL MONEY WENT STRAIGHT HOURS WERE MOVING ALONE EGGSSEVENDRYCARSGATHEREDSUQ—IFLOERS HANGHEAVYBUTWHYUNTILANARMY DURINGSERTICEEEELSWRCNGTIILITPRESSED FAIRSENSEEXCEPTITSELFSPREADSWESTUNTIL ACROSS WHILE CCNIIJG FOR NOE WATCH FULL NOI‘ T‘CMARD WINDCWS [\EAR CHAIRS THUS ARE SINGLE FDE SEEMS WITH WEIGHT HANGING LOTS SHOP BEST DURING PARTIES THEIR YEAR NEARLY WESTTHLLHT‘INGCLIJBSKILIEDBIRDSBEHIND NORIHBELWTHOSETREESSHADEFARMPHJVIDES SLM’ERSHOULDDANCESCDNAIVDNGBCXIGBURN HURTSBEYQHDLWGREACHISEASYTILL DIRECI‘SMARKETOFVALUECAMEFIVEDAYS YETCAUGITWINDFIRSTBECIMESJUDCEABLE SHOULD SITUNDERWATERVDRLDLEADSAGAINST JUST MODERN OPINIONS STAY CLEAR NOT WILD VALLEYCFTENGREWWISEIASTHOURWENT SAFEIFBETWEENEARTHITSWILDNON ANDWAIT'IHREECHMANDSWSTMEETCLOSE DOLLARSWAIUHANDFREEHOLDYGJRPRETTY a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. d. a. b. c. d. a. b. c. d. a. b. c. d. 156 SSI LIST - FORM B GOVERNMENT POSTS YOUR PAIR TOOK WALKS FOR IF BIDKENMENADD WATEREVENTO EGGSSEVENDRYCARSGATHEREDSUCHFLOAERS WAITEASYUNTILTAKENIVDRE SCI-IUDLSOUGHT CENTS SCME‘I'I'MES KEEP BETWEEN WIDE ENOUGT SHOES FAIR SENSE EXCEPT ITSEIF SPREADS WEST UNTIL MUST ACT BUSY TOVDRRON AND STOP FOR CLEAR FRGVI CLOUDS ABOVE EARTH OPENS WIDE PEACEARCIJNDEARTHVIN'TCZXREPAST DANCING GRINDSMENHCMENONETLENSSUCHHAPPYCO’IPANY ANDWAITTI-LREECCI‘MANDSWSTMEETCLOSE NORTHPASTABANKREMAINSABLETO LASTINBODYW’ENTTHESWEEI‘I‘DDEY WARMDAYSI‘UIEWHIIE FAVORS IIJN'TLET SWVIERSHCIJLDDANCESCXNAMONGBWSBUHT ACIDSSM-IILECQ’LDEEORNQEWATCHFULL GRHJPS BE W WHILE UNDER STREAMS BUILDING TASTES GREAT DURIM; DEATH RATHER BREAK WATER MEY WENT STRAIGHT HGJRS WERE IVDVING ALONE HEAVY LADIES THINK PRACTICE KEPT WHO DEALS HANGHEAVYBUTWHYUNTILANARMY YETCAUGH'I‘WINDFIRSTBECII/IESJUDGEABLE ALTHQH-IDEEPREDCOALSJUSTNEEDSLWG WHILELESSCLOSEEKCEPTBELWWHITEGARDENS UNDERICE FOR FISH ESCAPE BEYQQDWAR PEACE SEENB KNOVN SPIRITS MET LITTLE SINGLES VALLEYOF'I‘ENGREWWISELASTHOURWENT IAYSUCHFEARPAIDAHEAVYPURPOSE MILES WHILE QUICKLY SING OUTSIDE IN HWOR SHORT HUMAN HAIR FLIES GOING TILL THAN STILL PIEASED SINCE CCMING OUTSIDE STORES BUILD BROTHER PAID BRIGHT MUSIC DWBLED IN SCI—ICDL REAL GIRLS KEEP LESS CHANCE RESULTS CONSIDER BEHINDBUI‘JOYCFPUBLICLANDSQVETIMES WESTTILLLIETINGCIUBSKILLEDBIRDSBETHND SMALL TRIPS MEN'T BUY BHDKEN LIGHT NEXT 157 SSI EORMB . (Cont.) 10. ll. 12. l3. 14. 15. 16. l7. l8. 19. a. b. c. d. a. b. c. d. a. b. c. d. a. b.- c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b.- C. d. a. b. c. d. THOUGH WORIH DOLLARS IF SUCCESS MUST WAIT GAMES THAT BEING PAST YWR VILLAGE FOLIO/VS NOTTYMARDWINDCMSNEARCHAIRS T'HUSARE FEEL EIGHT YARDS SHOULD BURN SOMETHING SMOKE TAKEANYTHINGABGIECLQJDSSEEMBRIG—ITWHILE JUSTGLADDIFFERENTVALIJE‘STEACHMINDSCZJITE NEWS PLEASED EIGH'T GIRLS UNDER BOXES PREPARE UNDERNONEBURNED FINE NEARLY SINCE CDLD WEN'T STRANGE QUICKLY MADE SEVEN TRAINS VIN'T THESE ESCAPE CLASSES INSTEAD SAVE TREES TILL FELL WHILE (IDLD SOFT BEINGS BUT THOUSANDS SAVE SOULS NEARLY LOST BESIDE FUTURE FAIR ICE FINALLY STOPPED EIG-IT SIATES VALUE OF HEATSEEMS DIFFERENT SECEGCDDMNNERSCAN'T EmDSBUILDSTRAIGHTTODAYSHOTSHTREES JUST IVDDERN OPINIONS S'IAY CLEAR NOT WILD COOKSALIVDSTTmmRDSPREPAREDBEEORE SCENETGJAYWSLLIQGQUITETRLEWITHW LIEAMCNGTHOSEWILDCCXJN'IRYGARDENBURNED GUARDS SPRINGDE WHILE HANGING WHERE WEIGHT TIES EXPLAINS SOVIETHING WE MTG-I OIL PRIBLEMS FELL IVDRE HILLSLAYVILLAGESMILESWN'T PART ALTHOUGHVKDDACI‘SGAINALLVIEWINGFUTURE SHOPBESTDURII‘EPARTIESTHELRNEARNEARLY mLLARSWAIWHANDFREEHOIDYOURPRETTY FORWNDSTSPARE MILLIO‘IS SAFELXWATCH SEE/[YOUNGHGIEVERSENDSTRAIGHTAWAYFHI’I CLCIHESACCOUNTNEARLY PAYS YGJRBOYCANNOT TOCBJECTWHIIESHECCIJLDCLEARENCIJG-I ADUJNTACRJSSTHEBNIKSTORE ICEIN CLEAN THROUGH PAPER WHICH PRCMISES ME GREAT HURTSBEYCNDLLNGREAGI ISEASYTILL HOUSE EAST LAND YOURSELF LINN WITH THOSE CEASE CHANGE WEAR DRESSES WHILE LISTENING MADE DAY SAVE DCISS IN BRONN (NLY ACCORDING BEDS CAN'T KNCW MUSICAS POST CHIEF ISCUTLASTWEEKSGJNGAVEPRACTICE ‘I‘AKEOUTWI'DLENUMBEIEALMYSWENTAMY SAFEIFBE'IWEENEARTHITSWILDHG’V WATER FISH UNTILNEXT'SETISLIKE SSI 20. 21. 22. 23. 24. 25. 158 FORM B (Cont.) a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. C. d. WERE PLEASANT WHILE WARVI DAYS ACCOUNT FOR ALREADY TUDK CDMMJN WANTS OLD TRUTH KEPT SO(N CAN WALK FAST EXCEP'T BUILDING A NORIH BELON THOSE TREES SHADE FARM PROVIDES EGGS GRGNING CLOSE PRE'ITY CLEAN SNG'V LIES WHILE UNDER NEE BIG MEETIMSS BELONG KNGNN MORE CLOIHES HAD NEVER CHANGED MEMBERS THAT CANNOT FIT SHOES FAST RATHER LOST AFTER FEW YARDS TILL HEALTH STORES WARM PRESSES CCMPLE'IE HIS WT ARRIVES NEAR SCHCDL ON SHIPS UNDER'JALLMEIHODS REACHABOVEREAL CAN'T WRITE LIKE STONE WILL GIVE PLEASURE COUNTRYLEDWAROFMJSICDANCEDAIL EARIHONSOHFATALWSFANDUSHAT DARES NOT SERVE COMPLETE AMOUNT FELL CROSSES PLEASURE SIMPLEF'RIENDGOTMIATALLNINE ALT'HOUGH GROUPS TALK LATE POST MAKE DURING SERVICES FEELS WRONG TILL IT PRESSED OVER BUILDINGS WALK HARD LIKE MIDD ON CHILDREN TOO SPOKE LAST BECAUSE SEVEN MILES LIKE CERTAIN ROCNS EXPRESS HERVOICE OUGHT DIRECI‘SMARKETOFWIJECAMEFIVEWAVES SINCE GUARD SOLDIERS OF GDNTROL KEEP BEAUTY TASTES GREAT DUle DEATH RATHER BREAK AFTER a. b. c. d. a. b. c. d. a. b. c. d. a. b. C. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. 159 SSI LIST - FORM C - YETCAUCH-ITWINDFIRSTBECCMESJUDGEABLE CO’IPLETEHISMOVEMENTARRIVESDEARSG-ImLCN LEWGTHOSEWIIDCOUNITUGARDENBURNED HOUSE EASTLANDYCURSELFLXWQNWITH THOSE SAVE SOULS bEARLY LOST BESIDE FUTURE FAIR THESE ESCAPE CLASSES INSTEAD SAVE TREES TILL I\EWS PLEASED EIGHT GIRLS UNDER BOXES PREPARE LAY SUCH FEAR PAID A HEAVY PURPOSE CANNOT FIT SIDES FAST RATHER LOSE AFTER SEEM YCIJNG HOWEVER SEND STRAIGIT AWAY FRIVI GUARDS SPRINGIBE WHILE HANGING WHERE WEIGHT TIES SHIPS UNDER'IALLMETHODSIEACHABOVEREAL REAL GIRLS KEEP LESS CHANCE RESULTS CCNSIDER WWI-LENNONETURNSSUCHHAPPYCQ’IPANY GROUPSBECCMVHTWHIIE UNDERSI‘REAMDBUILDING WATERFISHUNTILNEXTSETISLIKE TAKEOUTM-IOLENUMBERSALMYSWENTAMAY WESTUNTILLIFTIMSCLUBSKILIEDBIRDSBEHIND EARLHCNSOI-EATAIIONSFAMIJSHAT CLOTHESACCGJNTNEARLYPAYSYOURBOYCANNOT NORIHPASTABANKRFMAINSABLETO ICE FINALLY STOPPED EIGHT STATES VALUE OF EARTHONSOHEATALIKMSFANDUSHAT LASTINBODYI‘DVEMENTTHESWEETMI‘EY UNDER ICE FOR FISH ESCAPE BEYOND WAR HANGHEAVYBUTWHYUNTILANARMY PLEASURE SIMPLEFRLENDGOI‘WIATALLNINE NDRE HILLSLAYVILLAGESMILESW'T PART ECESSEVENDRYCAIEGATHEREDSUCI‘IFLWERS WHILEUNDERNINEBIGMEETINGSBELONGIQ‘ICWN IDLIARSWARYHANDFREEHOLDYGIRPIETTY HURISBEYONDLLNGREACHLSEASYTILL SCEBETODAYGKNVSLCNGQUITETHJEWITIW WEREPLEASANTWHILEWARMDAYSACCDUNTFOR BROTHER PAID BRIGHT MUSIC DOUBLED IN SCI-KIDL I-EATSEEWBDIFFERENTSINCECUDDMNNERSCAN'T FORMC 10. ll. 12. 13. 14. 15. 16. 17. 18. 19. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. C.‘ d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. 160 (Cont. ) IE‘BKDIENMENADDWATEREVENTO ACROSSWILECCNINSFORNONEWATCEFULL SHORT HUMAN HAIR F'LIES GOING TILL THAN BEHINDBUT JOYOFPUBLICLANDSCIVIETIMES ALTHOLU-IGHXJPS'IALKLATEYETNDSTNAKE FOR WHOVI NEST SPARE MILLIONS SAFELY MATCH mN'TWRITELIKESTOEWILLGIVE PLEASURE EGGSGRCWINGCLOSEPRE‘ITYCLEANSNWLIES SAFEIFBE’IWEENEARIHIT‘SWILDHW UMDERNCNEBURNEDFINENEARLYSINCECOLD HEAVY LADIES THINK PRACTICEKEPTWHODEALS EXPLAINS SCMETHING WE WATU-I OIL PFDBLEMS FELL NDFEYWENTSTRAIGTTHOURSWEREI‘DVINGALQE FEWYEARSTILL HEALTH STORESWARMPRESSES ALTHOUG-IWCIDDACISGAINALLVIEWINGFUIURE MITEASYUNTILTAKENI‘DRESCI-IOOLSOUGHT STILL PLEASED SINCE CCIVLING OUTSIDE STORES BUILD COUNTRY LED WAR OF MUSIC DANCED ALL FELL WHILE COLD SCFT BEINGS BUY T'HOUSANDS DAY SAVE LIES IN BHMN CNLY ACCORDING DURIMSSERVICEFEELSWRONGTILLITPRESSED TASTE‘SGREATDURINGDEATHRATHERBREAKWATER AMIJNTACROSSTHEBACXSTOREICEIN DDRECLOITESHADNEVERCHANCEDMEMBELBTHAT TOOBJECTWHIIESHECDUIDCLEARENOUCE MUSTACTBUSY TUDRRQVANDSTOP FOR LEECERIAIIIROQ/JSEXPRESSI-ERVOICEOUGHT CLEARFKJMCLOUDSABOVEEARTHOPENSWIDE SCHVCANWALKFASTEXCEPTBUILDINGA OVERBUILDINGSVWKHARDLIKEWDDCN VALIEYOF‘IENGREWWISELASTHOURWENT DIRECI’SMARIEI‘OFWUECAMEFDIEWZWES SINGLE FILE SEEMS WITH WEIGHT HANGING LOTS CLEAN THROUQ-I PAPER WHICH PRCMISES ME GREAT ALTHOUGH DEEPREDCOAIS JUSTNEEDS LONG MILES WHIIE QUICKLY SILK; OUTSIDE IN HCNOR IS CUTLASTWEEK SOCNGAVE PRACTICE BEDS CAN'T KNCM MUSIC AS POST CHIEF CmKS HEAT ALMOST Tm WORDS PREPARED BEFORE THCIJGH WORTH HDLLARS IF SUCIESS MUSTWAIT FORMC 20. 21. 22. 23. 24. 25. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. 161 (Cont.) PEACE SEED/B KNOW SPIRITS MET LITTLE SINGING FAIR SENSE EXCEPT ITSELF‘ SPREADS WEST UNTIL GAMES THAT BEGIN PAST YOUR VILLAGE FOLLONS WARM DAYS MOVE WHILE FAVORS LDN'T LET WHILE LESS CLOSE EXCEPT BELGV WHITE GARDENS JUST MODERN OPINIONS S'LAY CLEAR NOT WILD SMALL TRIPS MDN'T BUY BROKEN LICi-IT NEXT WHILE LESS CLOSE EXCEPT BELON WHITE GARDENS GOVERNMENTPCBTSYOURPAIRTIDKWKSFOR NOTTG’EXRDWINLIWVSDEARCHAIRSTHUSARE FELLEIGHTYARDSSHOULD BURN SOVIETHING SMOKE SHCIJLDSITUNDERWATERmRLDLEADSAGAJNST CLEARFRCMCLOUDSABOVEEARTHOPENSWIDE ALREADYTOOKCWMNTSOLDTRUTHKEPT SHOP BEST DURING PARTIES THEIR YEAR DEARLY FELL WHILE COLD STT BEINGS BUY THOUSANDS TAKEANYTHINGABOVECLUJDSSEH’IBRIGH’TMIILE JUSTGLADDIFFERENTVALDESTEACI‘ERMINIEQUITE FIDDS BUILD STRAIGHT TODAY SHOT SIX TREES SINCE GUARD SOLDIERS OF CCN'TROLKEEPBEAUI‘Y CLEARFRCMCLOUIBABOVEEARIHOPENSWIDE CENTSSCMETIMESKEEPBETWEENWIDEENOUGHSHOES WENTSTRANGEQUICKLY MADE SEVENTRAINS WON'T TOOBJECTWHILESHECOULDCLEARENOUGH a. b. c. d. a. b. c. d. b. C. d. a. b. c. d. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. 162 SSI LIST - FORVI D VALLEYOF'IENGREWWISELASTHOURWENT IF‘BIKDKENMENADDMTEREVENTO (IDMPLETEHISD’OVEMENTARRIVESDEARSCHCXDLON FCDDS BUILD STRAIGHT TODAY SHOT SIX TREES NORTH BELON THOSE TREES SHADE FARM PROVIDES HURTSBEYONDLOIGREACYISEASYTILL SAVE SOULS NEARLY LOST BESIDE FUTURE FAIR WATERFISHUNTILNEXTSETISLIKE NOTTCMARDWINLXINSNEARCHAIRSTHUSARE STILL PLEASED SINCE COVIIDE OUTSIDE STORES BUILD ALLNGD’DVINGWENTSTRAIGHTHCIJRSWERE EGGSGRONINGCLIEEPRE‘TTYCLEANSNGVLLES BROTHER PAID BRIGHT MUSIC LIDUBLED IN SCHCDL WHILE LESS CLCBE EXCEPT BELW WHITE GARDENS MUST ACT BUSY TOVDRRON AND STOP FOR UNDER NONE BURDED FLDE NEARLY SINCE COLD GUARDS SPRINGING WHILE HANGIDE WHERE WEIGHT TIES YET CAUGHT WIND FIRST RECIPES JULIE ABLE [IDLIARSWAWHANDFREEHOLDYOURPRETTY CANNOT FIT SHOES FAST RATHER LOSE AFTER MILES WHILE QUICKLY SING OUTSIDE IN HONOR TOCBJECTWHILESHECOULDCLEARENOUCH HANGHEAVYBUTWHYUNTILANARVIY CIDUNTRYLEDWAROFMUSICDANCEDALL BEDS CAN'T KNOV MUSIC AS PCBT CHIEF FELL EIGHT mRDs SHOJLD BURN SOVIETHING SMOKE HEAVY LADIES THINK PRACTICE KEPT WHO DEAL EARTH (N SO HEAT ALLONS FAD’OUS HAT UNIDER ICE FOR FISH ESCAPE BEYOND WAR SHOP BEST DURIDE PARI'IES THEIR YEAR NEARLY FELL WHILE COLD SOFT BEIDBS BUY THOUSANDS CLOTHES ACCOUNT NEARLY PAYS YOUR BOY CANNOT FOR WHOM D’DST SPARE MILLIONS SAFELY WATCH ALTHOUGH DEEP REID COALS JUST NEEDS LONG JUST DDDERN OPINIQVS STAY CLEAR NOT WILD SMALL TRIPS W'T BUY BROKEN LIGIT NEXT EORMD‘ 10. ll. 12. l3. 14. 15. l6. l7. 18. 19. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. 163 (Cont. ) DAY SAVES LDGS 1N BRONN ONLY ACCORDING CCDKSHEATALIVDSTTmmRDS PREPARED BEFORE WXRM DAYS NOVE WHILE FAVORS DON'T LET SHORT HUMAN HAIR FLIES GOING TILL THAN LIKECERTAINROCMSEDCPRESSHERVOICEOUGHT mN'TWRITELIKESTKNEWILLGIVEPLEASURE CWDSWHENNCNETURNSSUCHHAPPYCCMPANY ADDUNTACROSSTHEBACKSTOREICEIN WHILEUNDERNINE BIGWBELWGW GAMES THAT BEGIN PAST YOUR VILLACE EOLLONS DDRECLOIHESHADMIERGIAI‘KIEDMEMBERSTHAT LEAWGTHOSEWILDCCIJNIWGARDENBURNED ALTHOLHTVKDDACI‘SGAINALLVIEWDIGFUTURE SLM’IERSHCIILDDANCESGJNAMJNGBODKSBUW PEACEAROUNDEAR’IHWCN'TCAIEPAST DANCING LAYSUG-IFEARPAIDAI-LEAVYPURPOSE CENTS SCMETIMES KEEP BETWEEN WIDE ENOUGH SHOES ALTHOUGHGKXJPSTALKALTEYETMOSTM JUSTGIADDIFFERENTWUESTEAO—IMLMJSQUITE WENTSTRANGE QUICKLY MADE SEVENTRAINS WIN'T OVERBUILDINGSWKHARDLIKEVDODCN SCENETODAYGRONSLOTGQUITETRUEWIT‘HIN EXPLAINS SOVIETHING WE WATCH OIL PROBLEMS FELL HGJSE EAST LAND YCIJRSELF m WITH THOSE THOUGH WORTH mLLARS IF SUCCESS MUST WAIT CEASE CHANGE WEAR DRESSES WHILE LISTENING MADE NEWS PLEASED EIGHT GIRLS UNDER BOXES PREPARE DARES NOT SERVE COVIPLET'E ANDUNTS FEEL CROSSES SEEMYCIJNGHCMEVERSENDSTRAIGH‘TAWAYFIU’I WPOSTSYCXJRPAIRTOOKWAIKSEOR CLEARFROVICLOUDSABOVEEARIHOPENSWIDE PEACE SEEMS KMMN SPIRITS MET LITTLE SINGING SINGLE FINE SEEMS WITH WEIGiT HANGING LOTS WERE PLEASANT WI-LLLE WARM DAYS ACCOUNT EOR HEAT SEEMS DIFFERET SINCE GOOD MANNERS CAN'T DURIBE SERVICES FEELS WRONG TILL IT PRESSED REAL GIRLS KEEP LESS CHANCE RESULTS CONSIDER LAST'INBODYMDVEMENI‘THESWEETI‘DNEY ALREADYTLXDKCOMDmWANTSOIDTWIHKEPT SCXNCANWALKFAST EXCEPT BUILDIDEA FORTD. 20. 21. 22. 23. 24. 25. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. a. b. c. d. 164 (Cont. ) ISCUTLASTWEEKSODNGAVE PRACTICE IVORE HILLS LAY VILLAGES MILES W'T PART DIRECTSMAMperimental Design. New York: McGraw-Hill Book Carpany (1971) . Wofford, M.J., cited by Martin, F.N., and Mussel, S.A., In Pauses in the Carpeting Signal a1 Synthetic Sentence Identification Scores. Journal of Speech and Hearing Disorders, 44, No. 3, 282-287 (1979) . 201 Young, L.L., Parker, C., and Carhart, R., Effectiveness of Speech and Voice Maskers on Numbers Ehbedded in Discourse. Journal of Acoustical Society of America, 58, No. 1, 535 (1975), MICHIGAN STATE UN IV. LIBRRRIES 111111 Ill1111|11||H111111111111111 31 9 10 714111 23 1 5