OVERDUE PINES ARE 25¢ PER DAY PER ITEM Return to book drop to remove this checkout from your record. We 5” © 1979 MARK HOWARD GRE ENWALD ALLRIGHTS RESERVED THE EFFECT OF DECREASED FREQUENCY BANDWIDTH ON SPEAKER IDENTIFICATION BY AURAL AND SPECTROGRAPHIC EXAMINATION OF SPEECH SAMPLES BY Mark Howard Greenwald A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of MASTER OF ARTS Department of Audiology and Speech Sciences 1979 V, ABSTRACT THE EFFECT OF DECREASED FREQUENCY BANDWIDTH ON SPEAKER IDENTIFICATION BY AURAL AND SPECTROGRAPHIC EXAMINATION OF SPEECH SAMPLES By Mark Howard Greenwald The purpose of this investigation was to determine the effect of decreased frequency bandwidth in addition to talker sex and examiner training on the ability to produce correct speaker identifications and eliminations by aural and spectrographic‘examination of speech samples. Twenty-four subjects, 12 male and 12 female, individually recorded the same sentences twice on two occasions one month apart, through a commer- cial telephone line. These recorded materials were filtered at four different frequency bandwidths (240 Hz - 2000 Hz, 240 Hz - 2500 Hz, 240 Hz — 3000 Hz and 240 Hz to 4000 Hz). Sound spectrograms and record- ings stored on 7" open-reel magnetic tapes served as test materials for two groups of examiners (two professional and two trainee) in 192 dis- crimination-type trials of voice identification. The frequency range of "Known" samples was always 240 - 4000 Hz. The frequency range of "unknown" materials was varied according to the type of trial performed in each instance. A total of 768 decisions was produced by all four examiners employed. From observation and analysis of these data, it was concluded that the effect of restricted bands of speech frequencies does not increase the errors, but rather, increases the percentage of non- opinions. Also, it was concluded that male and female talkers are essen- tially identified and eliminated at equal rates. Finally, it was con- cluded that the training of the examiner has an important effect on the error rate of this method of voice identification. ACKNOWLEDGMENTS I express thanks to Dr. Oscar Tosi for his guidance and time. In addition, a special thank you to Dr. Michael Chial, Dr. Deborah Spates, Mr. Ken Beachler, Mr. Fred Barney, Dr. Martin Fox, Lt. Lonnie Smrkovski, and Mrs. Shirley Taubman for their assistance. Most of all, to my parents for their support and understanding. TABLE OF CONTENTS LIST OF TABLES . . . . . LIST OF FIGURES . . . . . . . . Chapter I. II. III. IV. STATEMENT OF THE PROBLEM Introduction . . . . Statement of Problem and Purpose of Study . . . . . Hypotheses Tested . . Importance of Study . Definition of Terms . REVIEW OF LITERATURE . Introduction . . . . Restricted Frequency Bandwidth Aural-Spectrographic’Experiments DESIGN OF STUDY . . . . Subjects . . . . . . Phonetic Materials Instrumentation . . . Spectrography . . . Speech Spectrograms . Trials of Voice Identification and Elimination . Examiners . . . . RESULTS AND DISCUSSION Analysis of Results . Raw Scores . . . . Results . . . . . . . Frequency Bandwidth Talker Sex . . . . Examiner Training . Discussion . . . . . Effects of Reduced Frequency Bandwidth Effects of Talker Sex . Effects of Examiner Training iii Page vii I-l \JO‘UIb 10 11 16 16 17 18 28 30 31 33 35 35 37 38 38 52 63 74 74 q l 77 Chapter V. SUMMARY AND CONCLUSIONS . . Implications for Further Research . APPENDIXES A. EVALUATION OF SUBJECTS' VOICES B. FUNDAMENTAL FREQUENCY C. LIST OF PHONEMES D. RESPONSE OF TELEPHONE E. EXAMINER'S AUDIOGRAMS F. INSTRUCTIONS G. SCORE SHEETS H. RAW SCORES . LIST OF REFERENCES . OF EACH 0 O O O SUBJECT iv Page 78 81 82 95 98 99 108 112 116 118 120 LIST OF TABLES Table Page 1. Match Trials: Number and Percentage of 5 Alternative Decisions as a Function of Frequency Bandwidth for Each Examiner . . . . . . . . . . 39 2. 'X2' Values for Frequency Bandwidth in thh Trials 0 O O O O O O O O I O O O O O O I O O I C O O 42 3. Match Trials: Number of Correct (1 and 2), Error (4 and 5) and No-Decisions (3) as a Function of Frequency Bandwidth for Each Examiner . . . . . . . . . . 43 4. 'rs' Values for Frequency Bandwidth and Examiners' Decisions in Match Trials . . . . . . . . . . . . . . . . 44 5. No-Match Trials: Number and Percentage of 5 Alter- native Decisions as a Function of Frequency Bandwidth for Each Examiner . . . . . . . . . . . . . . . 47 6. 'X2' Values for Frequency Bandwidth in No-Match Trials 0 O O O O O O O I O O 0 O O O O O O O O O O O O O O 49 7. No-Match Trials: Number of Correct (5 and 4), Error (1 and 2) and No—Decisions (3) as a Function of Frequency Bandwidth for Each Examiner . . . . . . . . . . 50 8. 'rs' Values for Frequency Bandwidth and Examiners' D8C1810n8 in No-MatCh Trials 0 e e e e e e e e e e e e e 51 9. Match Trials: Number and Percentage of 5 Alternative Decisions as a Function of Talker Sex for Each Examiner O O O O O O O O O O I O O O O O O O O O O O O O O 5 3 10. 'X2' Values for Talker Sex in Match Trials . . . . . . . . . 56 11. Match Trials: Number of Correct (1 and 2), Error (4 and 5) and No-Decisions (3) for Each Examiner as a Function of Talker Sex . . . . . . . . . . . . . . . . . . 57 12. No-Match Trials: Number and Percentage of 5 Alternative Decisions as a Function of Talker Sex for Each Examine r O O O O O O O O O O O O O O O O O O O O O O O I O 5 8 Table 13. 14. 15. 16. 17. 18. 19. 20.. 21. 22. 'X2' Values for Talker Sex in No-Match Trials . No-Match Trials: Number of Correct (5 and 4), Error (1 and 2) and No-Decisions (3) for Each Examiner as a Function of Talker Sex . . . . . . . . . . . Match Trials: Number and Percentage of 5 Alternative Decisions by Professional and Trainee Examiners as a Function of Frequency Bandwidth and Talker Sex . . . 'XZ' Values for Examiner Training in Match Trials . . . No-Match Trials: Number and Percentage of 5 Alternative Decisions by Professional and Trainee Examiners as a Function of Frequency Bandwidth and Talker Sex . . 'Xz' Values for Examiner Training in No-Match Trials . . Number and Percentage of Cases in Which a Higher Level of Confidence was Given to a Lower Frequency Bandwidth List of Each Subject's Fundamental Frequency . . . . . . Examdners' Responses Within each Frequency Bandwidth for Each Talker in Match Trials of Voice Identification Examiners' Responses Within each Frequency Bandwidth for Each Talker in No-Match Trials of Voice Identification . . . . . . . . . . . . . . . . vi Page 61 62 64 68 69 73 75 96 118 119 10. 11. 12. 13. 14. LIST OF FIGURES Schematic Diagram of Equipment Connections to Produce the First Generation Tapes . Schematic Diagram of Equipment Connections to Produce the Second Generation Tapes . Schematic Diagram of Equipment Connections to Produce the Third Generation Tapes . Schematic Diagram of Equipment Connections to Produce the Fourth Generation Tapes . Schematic Diagram of Equipment Connections to Produce the Fifth Generation Tapes . Average Amplitude Display for the Same Speech Sample at Two Different Recording an erat ions 0 O O O O I O O O O I O C 0 Frequency Calibration Markings Reference in the Experimental Speech Spectrograms . . Labeling of Speech Spectrograms . . . . . Match Trials: Median Values for Each Examiner's Deci- sions as a Function of Frequency Bandwidth . No-Match Trials: Median Values for Each Examiner's Decisions as a Function of Frequency Bandwidth . Match Trials: Median Values for Each Examiner's Decisions as a Function of Talker Sex . No-Match Trials: Median Values for Each Examiner's Decisions as a Function of Talker Sex . Matdh Trials: Distribution of Professional and Trainee Examiners' Decisions for Male Talkers O O O I O O O I O O O O O O O 0 Match Trials: Distribution of Professional and Trainee Examiners' Decisions for Female Talkera C C O O O C O O O O O O O O O 0 vii O Page 19 21 23 24 26 27 28 32 40 46 54 60 65 66 Figure Page 15. No-Match Trials: Distribution of Professional and Trainee Examiners' Decisions for Male Talkers . . . . . 70 16. No-Match Trials: Distribution of Professional and Trainee Examiners' Decisions for Female Talkers . . . . . . . . . . . . . . . . . . . . . . . . 71 17. Comparison of the Same Speech Samples from 2 Different Frequency Bandwidth Speech Spectrograms . . . . . . . . . . . . . . . . . . . . . . 76 18. Distribution of Subjects' Fundamental Frequency . . . . . 97 19. Schematic Diagram of Equipment Connections to Produce Frequency Response of Telephone ... . . . . . . 102 viii CHAPTER I STATEMENT OF THE PROBLEM Introduction Tape recordings of conversations are sometimes the only evidence or part of the evidence available in criminal offenses (e.g., assassina- tions, bomb threats, kidnapping, rape, false fire reports, obscene tele- phone calls). These recordings might be used to establish whether any suspected person's voice is the same as the questioned through voice identification techniques. Quite often recordings yield speech samples with a restricted range of frequencies. This limitation is not unusual due to the wide variety of room acoustics, transmission media and record- ing devices utilized in the recording process. Professional voice examiners trained in voice identification techniques may determine a speaker's identity by aural and spectrographic examination of speech samples from the unknown (questioned) and one or several known (suspected) persons (Tosi, 1972). The examiner selects one of the following alterna- tive decisions after completing examination of speech samples: a) Positive identification (a high degree of confidence that the unknown voice is the same as a known one) b) Probability of identification c) Positive elimination (a high degree of confidence that the unknown voice is different from the known one) d) Probability of elimination e) No opinion one way or the other (Tosi, 1979). In some instances, the examiners are unable to make a determina- tion of same or different because of limited amounts of speech frequency information. The advantage of the five alternative decisions is in allowing the examiner not to make an indication of same or different, hence, avoiding the chance for error. Opponents to the use of aural and spectrographic examination of speech samples as evidence in a court of law claim that this technique is not valid under limited frequency bandwidth conditions. These claims were made by some expert witnesses during court room testimonies.l These claims cannot be supported or denied a priori because of a lack of formal aural-spectrographic studies which indicate whether correct speaker identification under limited frequency bandwidth conditions is possible. However, one study (Pollack, Pickett and Sumby, 1954) employ— ing only aural examination of speech samples2 found correct speaker iden- tification was possible under severely restricted frequency bandwidth conditions. In addition to the Pollack.g£_gl, (1954) study, the claims pre- sented by the opponents do not consider the examiners' alternative to offer no opinion one way or the other after examination of restricted frequency range speech samples. These samples may include widely fil- tered band(s) of speech frequencies, masking of speech frequencies by noise, or both. The end result is a restricted bandwidth of speech 1Transcripts of the cases California vs. Kelly, 1973 (Mr. Fausto Poza, expert witness for the defense; Massachusetts vs. Lykus, 1975 (Dr. Louis Gershman, expert witness for the defense; and California vs. Chapter, 1973 (Dr. Harry Holien, expert witness for the defense). 2Recently, a study employing three methods of voice identifica- tion indicated examiners using the aural or spectrographic method alone did not perform as well as when they used these methods together (T031 and Greenwald, 1978). frequency information. Under these conditions, the examiner is able to produce only a non—opinion decision; therefore, the issue of validity seems inappropriate in this instance. Presently, there seems to exist a disagreement among some pro- fessionals in the audiology and speech sciences regarding the range of frequencies which constitute restricted bandwidth conditions. 0n Febru- ary 9, 1979, the Columbia Broadcasting Company (CBS) had in their posses- sion a tape recording allegedly spoken by the Shah of Iran as well as three confirmed recordings of the Shah. The CBS network requested inde- pendently to both Dr. Tosi and Dr. Papcun to examine these tapes without disclosing the name of the alleged speaker (Broad, 1979). Both Dr. Tosi and Dr. Papcun came to the same conclusions; a) the tapes contained speech frequencies from 150 to approximately 3800 Hertz (Hz) and b) the tapes were produced by the same speaker within a high level of confi- dance. The National Broadcasting Company (NBC) at that time requested Mr. Poza and Dr. Hecker to perform the same analysis as Dr. Tosi and Dr. Papcun. According to statements from Mr. Wilson, the producer of - CBS,3 Mk; Poza declared that it was not possible to conduct any examina- tion on these tapes for purposes of voice identification because of their restricted range of frequencies (150 to 1500 or 1800 Hz). The CBS television network then requested the services of Bolt, Beranek and New- man Inc. to make a further determination as to the range of frequencies of these tapes. Dr. Bolt found a range of frequencies of 150 to 2500 Hz without any previous electronic elaboration on the tapes. Dr. Tosi and Dr. Papcun independently found the range of frequencies to be 150 to 3Telephone communication from Mr. Wilson to Dr. Tosi, 9 February, 1979. approximately 3800 Hz after improving the signal-to-noise ratio by proper filtering and amplification. A paper published in the Proceedings of the 1974 Carnahan Crime Countermeasures Conference (Poza, 1974) described and compared the condi- tions under which the "Voiceprint Method of Identification" has been validated in laboratory studies to those encountered in forensic cases. One laboratory study (Tosi, Oyer, Lashbrook, Pedrey, Nicol and Nash, 1972) as well as several forensic cases were employed in the comparison. The conditions compared were signal-to-noise ratio, speech frequency bandwidth, time elapsed between known and unknown recordings, selection of speech materials, examiner training and the duration of examinations. As a result of that comparison Poza concluded that a restricted frequency bandwidth among other parameters ... may adversely affect the error rate of professional examiners working on real cases ..." and that these parameters ... have not been adequately investigated." Poza suggested that these conditions should be further investigated. Statement of Problem and Purpose of Study Law enforcement agencies and society in general are interested in having a reliable technique of voice identification and elimination that can be used to acquit or convict a defendant. The technique cur- rently accepted by the Michigan State Police and the International Asso- ciation of Vbice Identification (I.A.V.I.) is aural and spectrographic examination of speech samples by a trained professional examiner. The problem is that there are no formal aural-spectrographic studies which describe the performance of these examiners under the conditions of decreased frequency bandwidth. The object of this study was to determine the effect decreased frequency bandwidth had on the professional examiners' and trainees' capability to make correct speaker identifications and eliminations by combined aural-spectrographic examination of speech samples. The bands of frequencies tested were 240 Hertz (Hz) to 2000 Hz, 240 Hz to 2500 Hz and 240 Hz to 3000 Hz. Control tests utilized speech samples with fre- quencies from 240 Hz to 4000 Hz. 4 The specific questions were: 1. Are correct speaker identifications possible under restricted frequency bandwidth conditions? Are correct speaker eliminations possible under restricted frequency bandwidth conditions? Are female speakers more difficult to identify than male speakers? Are female speakers more difficult to eliminate than male speakers? Do professional examiners perform differently in tests 'of speaker identification than trainee examiners? Do professional examiners perform differently in tests of speaker elimination than trainee examiners? Hypotheses Tested These questions were tested using the following null hypotheses: 1. 2. There are no significant differences between examiners' decisions in match trials by using phonetic materials within 240 to 4000 Hz and a) 240 to 3000 Hz, b) 240 to 2500 Hz and c) 240 to 2000 Hz. There are no significant differences between examiners' decisions in no-match trials by using phonetic materials within 240 to 4000 Hz and a) 240 to 3000 Hz, b) 240 to 2500 Hz and c) 240 to 2000 Hz. 4These ranges of frequencies are usually used in forensic cases because of the limitations introduced by telephone communications, mostly encountered in evidence recordings. 3. There are no significant differences between examiners' decisions in match trials including male or female speakers. 4. There are no significant differences between examiners' decisions in no-match trials including male or female speakers. 5. There are no significant differences between the decisions of professional examiners and trainee examiners in match trials of voice identification. 6. There are no significant differences between the decisions of professional examiners and trainee examiners in no-match trials of voice identification. Importance of Study The absence of formal aural-spectrographic studies on the pro- fessional examiner's ability to make correct speaker identifications/ eliminations under limited frequency bandwidth conditions has been stated by expert witnesses during court room testimony. Poza (1974) indicated that the effects of limited frequency bandwidth and examiner training needed further research. The present study was designed to explore the effects of a) decreased frequency bandwidth in speech samples, b) examiner training and c) talker sex on the examiners' ability to make correct speaker identifications/eliminations utilizing the combined aural and spectro- graphic examination method. Additional information concerning the dis- tribution of examiners' alternative decisions under these conditions is reported. This type of information could be valuable to those employing professional examiners as expert witnesses in a court of law and to those working on special projects concerned with speaker identification and transmdssion of speech signals. Definition of Terms Aural Method of Voice Identification: The technique of aurally compar- ing speech samples utilizing the short-term memory process. The short- term memory process is used when the unknown and known voices to be compared are not familiar to the examiner and they are continually and permanently available through magnetic tape recordings (Tosi, 1979). Spectrographic Method of Voice Idegtification: The technique of visually comparing same text speech spectrograms from unknown and known subjects. The speech spectrogram displays a continuous succession of short-term speech spectra (Tosi, 1979). Aural and Spectrographic Method of Voice Identification: The technique of aurally comparing speech samples by the short-term memory process and their correlated speech spectrograms; a combined coordinated task. Frequencyggandwidth: The difference between the lower and upper limit- ing frequencies of speech samples, expressed in Hz. Fundamental Frequency: The frequency F of the lowest harmonic compo- 0 nent of a voiced speech wave. The fundamental frequency is also the frequency F of a periodic speech wave. 0 ‘Trials of Voice Identificatigg; Tests where the unknown and known voice samples are compared to determine whether they were produced by the same persons. These trials can be subjective (aural and spectrographic method) or objective (computer methods). Open Trials: Those tests of voice identification in which the examiner is told that the unknown voice sample may or may not be included among the known voice samples. He has to decide whether the unknown sample is one of the known ones; and if he decides positively or probably, he must determine which known talker is same as the unknown (Tosi, 1979). Closed Tests: Those tests of voice identification in which the examiner is told that the unknown voice sample is included among the known voice samples. He has to decide which known sample is the same as the unknown one (Tosi, 1979). Discrimination Tests of Voice Identification: An open test where one unknown voice is compared to only one known voice. The outcome of this type of test is either that the two voices are the same or that they are different. However, in practice the examiner may choose not to decide one way or the other. Match Trials: Those experimental open or discrimination tests of voice identification in which the experimenter has included a voice sample from the unknown talker among the known talker voice samples (Tosi, 1979). No-Match Trials: Those experimental open or discrimination tests of voice identification in which the experimenter has not included a voice sample from the unknown talker among the known talker voice samples (Tosi, 1979). CHAPTER II REVIEW OF LITERATURE Introductiog_ Several laboratory studies on speaker identification (also called talker identification and voice identification) by the aural method (McGehee, 1937, 1944; Pollack.g£“gl., 1954; Bricker and Pruzansky, 1966; Coleman, 1973; Stevens g£_gl,, 1968), by the spectrographic method (Kersta, 1962; Young and Campbell, 1967; Hazen, 1973; Tosi 35.51,, 1972), and by the aural-spectrographic method (Smrkovski, 1975, 1976; Tosi and Greenwald, 1978) are available. These methods of voice identification are subjective since an examiner decides whether an unknown voice belongs to the same speaker. The published studies on voice identification using these subjective methods have investigated the effects of a wide variety of variables, e.g., voice disguise, context, speech sample duration, short and long-term memory, signal transmission and restricted frequency bandwidth, on the ability of examiners to identify/eliminate speakers from one or a group of speakers. This review will be limited to one study which explored the effect of decreased frequency bandwidth and to those studies which utilized the aural and spectrographic method for examination of speech samples. Studies of aural and spectrographic examination of speech samples are of interest since this method was adopted by the Michigan State Police and the I.A.V.I. as part of a stan- dard for the investigation of crimes where the voice is part of the evi- dence (Tosi, 1979). 10 Restricted Frequency Bandwidth In reviewing the literature on this subject, no formal aural- spectrographic experiments of voice identification testing the effects of restricted frequency bandwidth were found. Only one voice identifica- tion experiment utilizing aural identification by the long-term memory (Pollack g£_gl,, 1954) mentioned the effects of the frequency range of the speech signal utilized. In that experiment 16 male talkers were tape recorded speaking monosyllabic words from the Psycho-Acoustic Labo- ratory PB list. Tape recordings were played through low pass and high pass filters with 17 dB/octave roll-off characteristics. Six different cut-off frequencies were used for both low pass and high pass conditions. The resultant frequency bandwidths ranged from approximately F to 8000 0 Hz and F to 100 Hz for low pass conditions and 100 Hz to approximately 0 8000 Hz and 5000 Hz to 8000 Hz for high pass conditions. Tests of voice identification performed by seven listeners, who were all familiar with the talkers through daily contact, included identifying the unknown talker from a) one group of four known talkers and b) one group of eight known talkers. Listeners were not allowed to express a non-opinion decision. The results of these tests indicated that as the frequency bandwidth decreased, the number of correct speaker identifications decreased too. Nevertheless, correct speaker identification was possible in a majority of the trials even under severely restricted frequency band- width conditions (e.g., from F to 1000 Hz). Details specifying the 0 open or closed nature of the task and the number of false identifica- tions and eliminations were not reported by the researchers who concluded: ... that the identification of a speaker's voice is not critically dependent upon the delicate balance of dif- ferent frequency components in any single portion of the speech frequency spectrum (Pollack.g£.gl,, 1954). 11 Extrapolations from this study to forensic cases where the aural and spectrographic method'is employed, where the examiners are not familiar with the talkers through daily contact, and where the aural identification by the long-term memory is not used, should be cautiously applied. The listeners in Pollack's study were biased by the fact that they were familiar with the unknown and known talkers through daily contact, whereas professional examiners rarely encounter this situation. Indeed, professional examiners do practice the process of becoming famil- iar with the talker's speech behavior by listening to available record- ings, but not under the conditions of informal day-to-day contact. The following question could be asked: would the listeners in Pollack's study have performed better if they were also allowed to compare the talkers' speech spectrograms in conjunction with expressing their opin- ions as one out of five alternative decisions (previously mentioned on p. 1)? This aspect, among others, is now considered in experiments on voice identification. The following is a review of experiments employ- ing the method of voice identification by combined aural and spectro- graphic examination of speech samples. Aura1;§pectrggraphic Experiments The combined aural and spectrographic method has been utilized in forensic practice since 1971 (Tosi, 1972), but one of the first formal laboratory experiments employing this method was performed by Smrkovski (1975). The purpose of his study was to determine whether professional examiners would make valid determinations under field recording condi- tions by using that combined method. Three known voices were compared to one unknown voice in each of four open trials of voice identification. 12 Of these four trials, three were recorded from a telephone and one from a fire station's tape monitor (Dictaphone 4000). The calls consisted of one bomb threat, two false criminal reports and one extortion case. Three of the trials contained a matching speaker. Five professional examiners and two less-experienced examiners (with less than two years' training) were employed to examine these four trials. Examiners were instructed to analyze speech samples recorded on magnetic tape and to reach one out of five alternative decisions in each trial. All exam- iners were instructed to base all positive (same speakers) decisions on similarities found in more than 10 pairs of equal words from the unknown and known voices. Trials were the discrimination type. The answers to these trials indicated that professional exam- iners committed no errors of either false identification or false elimi- nation. In one case an error of false elimination was produced by a less experienced (trainee) examiner. In trials where the identity of the unknown speaker was difficult to determine, a no opinion decision was rendered by the professional examiner (7.52 of total trials performed by professional examiners). Smrkovski concluded that the method of aural and spectrographic examination of speech samples, when used by profes- sional examiners, is a valid method in certain forensic situations. The study did demonstrate the ability of professional examiners to make correct decisions under certain conditions, despite the fact that a small number of trials was produced. Smrkovski (1976) performed another experiment to further increase the amount of experimental data on professional examiner performance in trials of voice identification. 13 Fourteen speakers (seven male and seven female) of General Ameri- can English dialect each recorded nine words twice in a natural working environment on two occasions spaced 15 months apart. Twelve examiners (four professionals certified by the I.A.V.I., four trainees with less than two years' practice and four novices) performed a total of 120 dis- crimination-type trials of voice identification utilizing the aural- spectrographic method. Fifty percent were no-match trials, and 502 were match trials arranged in random order. Each examiner was instructed to choose one of five alternative decisions in each of a total of 10 trials. The results of that experiment indicated no errors of false identifica- tion or false elimination were committed by professional or trainee examiners. In addition, both types of errors were committed by the novice examiners. Professional examiners were more efficient than the trainee examiners, averaging higher correct positive identifications and positive eliminations. Smrkovski concluded that the experience of the examiner has an important effect on the validity of the aural-spectro- graphic method of voice identification. The most recent experiment on voice identification by the aural- spectrographic method investigated the effect that training, minority group voices, speaker sex, noise and non-contemporary speech samples had on examiners' ability to make correct speaker identifications (Tosi and Greenwald, 1978). Twenty-five female and 25 male talkers were randomly obtained from the Chicano population of Lansing, Michigan. Each talker recorded over the telephone from his/her home to the Michigan State Uni- versity Speech & Hearing laboratory. Each talker repeated three sen- tences and counted from 1 to 10 (in Spanish) on two occasions spaced six months apart. A cassette tape recorder, connected to the M.S.U. 14 telephone through a magnetic pick-up coil, was utilized for this purpose. The first recording was obtained in a quiet environment. The second recording consisted of repeating the same phonetic materials, once in a quiet environment and once in a noisy environment (radio or television in the room); noise was monitored at the M.S.U. laboratory to reach a 0 db S/N ratio. Eight-track loop tape cartridges and broad band (300 Hz) bar speech spectrograms (100 Hz to 4000 Hz) were prepared with the original cassettes to serve as test materials for three types of voice identification tests: 1) trials of voice identification by the spectro- graphic examination method; 2) trials of coice identification by the aural examination method, and 3) trials of voice identification by the combined aural-spectrographic examination method. All trials of voice identification were the discrimination type, 501 match and 50% no-match. Also, 50% of the trials were quiet recording condition and 502 noisy recording condition materials. Sixrhundred trials were produced, 200 by aural, 200 by spectrographic and 200 by aural-spectrographic examination methods. Two professional examiners certified by the I.A.V.I. and 10 untrained examiners were instructed to express one out of five alterna- tive decisions in addition to qualifying their opinions with a percentage rating of self confidence from 51 to 1002 identification or elimination. Fifty percent was considered a non-opinion. A total of 7,200 decisions were produced by the examiners as a group. Results from this study suggested that: 1. Examiner training is vital to the validity of results of a subjective method of voice identification based on combined aural-spectrographic examination of voices. 15 2. The conditions of sex, non-contemporary samples and minority group voices do not result in significant errors of identification/elimination provided profes- sional examiners are employed. 3. Noisy voice recordings yield a larger percentage of no- opinions from professional examiners. 4. Untrained examiners produce a greater variety of errors ranging from 22 for some examiners and up to 362 for others. The professional examiner's prudence in decision-making is reflected in the large percentage of no-opinion decisions made by them as compared with the untrained examiners. The experiments reviewed in this section have covered the effects of non-contemporary speech samples, noisy speech samples, talker sex, examiner training and minority group voices on the examiners' ability to make correct speaker identificationsleliminations utilizing the aural- spectrographic method. In general, the professional examiner is a pru- dent and efficient decision-maker when employing this method of examining speech samples. Although these experiments did not investigate the effects of restricted frequency bandwidths, one may apply the profes- sional examiner's no-opinion decisions produced under noisy conditions as an indicator of the examiner's unwillingness to make decisions of same or different when the speech samples were masked by noise. Such could also be the case if limited frequency bandwidth conditions are sub- stituted for noisy conditions. An experiment which investigates this effect is warranted. CHAPTER III DESIGN OF STUDY Subjects Twenty-four Caucasian subjects, 12 male and 12 female, between the ages of 25 and 32 years were employed in this study. The number of subjects was chosen for purposes of limiting the total time necessary to complete the experiment, considering statistically 24 subjects would yield a reasonable amount of data. The range of age, 25 to 32 years, was selected because of the availability of subjects in this age range, which also aided in developing a homogenious sample of voices. All subjects were selected by incidental means, obtaining their telephone numbers from friends known to the experimenter. The first 24 subjects who met criteria for acceptance were used as the sample group of voices in this experiment. Thirty-six voices were recorded before the sample yielded 24 acceptable ones. The criteria for accepting a subject for the study were as fol- lows: 1. Each subject would allow recording of his/her voice twice over the telephone, once initially (designated the "Unknown") and once a month later (designated as the "Rnown"). 2. Each was native to the Midwest Region of the united States of America. 3. Each had, up to the time of this study, resided in the Lansing area for at least five years. 16 17 4. Each spoke with no speech defects or marked phonetic idiosyncracies as determined by a speech pathologist. Refer to Appendix A for more information on each sub- ject's speech. 5. Each spoke within a 16.52 variation of the average usual fundamental frequency. This stipulation was made for the first call ("Unknown") only. The reference values were 125 Hz for male talkers and 200 Hz for female talkers (Boone, 1971). Refer to Appendix B for infor- mation on how these values were determined. Phonetic Materials Each subject was asked to repeat his/her name, the present date, his/her age and his/her telephone number. This information was recorded in the beginning of all speech recordings. All subjects were asked to repeat after the experimenter the fol- lowing monosyllabic words twice: A. lhit - hat - hat - hot - hut/ B. lhid - hat - had - hod - hud/ The monosyllables were not used in the actual trials of voice identifica- tion but for purposes of obtaining additional data of the habitual funda- mental frequency of each speaker. In addition to the monosyllables, all subjects were asked to repeat the following sentences twice: C. Julie ate clam chowder for lunch. D. The daytime flight is very boring. E. I'll speak with you later Monday night. F. Vision my boy sharing hot dogs. These sentences were employed as speech samples in the trials of voice identification. Together, these sentences contain most of the phonemes included in the American English Midwest dialect. The presence of each phoneme afforded each examiner the opportunity to sample a broad 18 speech repertory of the subject. A list was used to record the occur- rence of phonemes within each sentence (Appendix C) to show that these sentences represent a phonetically balanced list. Instrumentation All subjects were recorded through a commercial telephone line from their home telephones to the Michigan State University Speech & Hearing laboratory telephone. An induction coil (Dictaphone) was fitted to the telephone receiver at the Michigan State University telephone and connected to the microphone jack of a tape cassette recorded (Sony TC- 205). The recorded speech signal level was controlled within the cas- sette tape recorder by an automatic gain control circuit. The electrical line was used to power the cassette tape recorder, avoiding the chance of battery failure and slow speed recording. Cassette tape utilized was TDK D-C60. The recording of subjects' voices from the telephone was designated as the "First generation" recording (Figure 1). These cas- sette recordings were used to prepare master tapes only. These tapes were then kept in storage. A second recording was necessary to provide a master 7" open reel. These master tape recordings were produced by dubbing the sub- jects' voices from the cassette tapes to low noise 7" open reel tapes (Scotch Tenzar 1/4" x 1.5 mil). Tape cassettes were played back in a tape recorder (Sony TC - 205) connected from the ear monitor jack to the line input of an open reel "Crown" tape recorder built in the sound spectrograph (Model 700, Voice Identification, Inc.). The output signal level of the tape cassette recorder and the input signal level of the open reel tape recorder were set to 0 VU. All master tape recordings (both Known and Unknown) were recorded at 7.5 ips-full track with proper 19 inductlon subiects coll ' home telephone \ e._..._. |OCO| ca“ x~ \ cassette tape laboratory _ I telephone mic. input Figural Schematic Diagram of Equipment Connections to Produce the First Generation Tapes 20 speed equalization. These master open reel tape recordings were desig- nated as the "Second generation" recordings (Figure 2). A third tape re-recording involved the production of two complete sets of 7" open reel tapes, both copies from the Second generation tapes. One set of tapes was employed to produce speech spectrograms. The other set of tapes was employed to produce the aural materials for the trials of voice identification; the reason for this was that during the process of spectrogram production, the sound spectrograph removes a relatively slight amount of oxide from the tape. The removal of oxide represents an attenuation of the speech signal noticeable mostly in the high fre- quencies ( 4000 Hz). Third generation recordings from "Unknown" subjects were made by filtering master tape recordings at different range of frequencies (240 Hz to 4000 Hz, 240 Hz to 3000 Hz, 240 Hz to 2500 Hz and 240 Hz to 2000 Hz). This range was selected on the basis of the frequency response of the telephone and the amount of area in which to display the band of frequencies under investigation. The response of the telephone for local calls (Appendix D) indicated the bulk of frequency information lies between 300 Hz and 5400 Hz (ref. - -36 dB below the peak). Also, a third generation tape recording from "Known" subjects was obtained by filtering master tape recordings through a 240 Hz to 4000 Hz filter set- ting producing one tape containing the same Known speech samples. In total, two copies of third generation tapes were produced, one complete set from Known subjects and one complete set from Unknown subjects. One copy was used to produce spectrograms and the other copy to produce aural materials. Sound Spectrograph 21 ‘ Cassette tape aux. output Figure 2 Schematic Diagram of Equlpment Connections to Produce the Second Generation Tapes 22 All third generation tapes (Figure 3) were dubbed from the line output of an open reel tape recorder (Magnacord 1022-P, half track) through a band-pass filter (Allison Labs 25) to the line input of an open reel tape recorder (Model 700). A fourth tape recording was needed to produce the aural compo- nent of all the subjects' voices into the actual trials of voice identi- fication. The production of that recording involved the use of three open reel tape recorders (Model 700, Revox 77-A and Revox 700-A). One tape recorder (Revox 77-A 1/4 track, 1 channel) played the Third genera- tion unknown recordings. The third tape recorder (Revox 700-A 1/2 track, 1 channel) recorded the voice samples sequentially from either of the playback tape recorders. By dubbing from either line output of the play- back tape recorders into the line input of a third tape recorder (Fig- ure 4), the actual trials of voice identification were constructed onto the fourth recording tapes (Scotch 206 1/4" x 1.5 mil). The following was the format for presenting one trial: 1. The trial number was verbally spoken by the experi- menter ("Trial number one") followed by a three second lapse of silence. 2. The Known voice was recorded speaking the first repetition of the first sample sentence followed by a half second lapse of silence. 3. The Unknown voice was recorded speaking the first repetition of the first sample sentence followed by a half second lapse of silence. 4. The Known voice was recorded speaking the second repetition of the first sample sentence followed by a one second lapse of silence. 5. Beginning with 2, above, the same format was repeated substituting the second sample sentence for the first through the fourth, where the same format was again repeated for the third and fourth sample sentences. Following the last spoken sentence by the Known voice was a three second lapse of silence, a lapse preceding the presentation of the next trial. 23 Magnacord \ Bandpass filter aux. output 6} @J ...... Figure 3 Schematic Diagram of Equipment Connections to Produce the Third Generation Tapes AC. g5: :5} E El °°°-‘J * Sound Spectrograph 700 ' , o e 0 9 . . . ,\ e 9 line input Revox E u ‘ oo Em ca 3.0. \ aux. output 0.6. ‘f' > " Sound Spectrograph ""7‘00 4*: :©§ \ ' .o o headphone output Figure 4 Schematic Diagram of Equipment Connection to Produce the Fourth Generation Tapes 25 The recording of Known and unknown voice samples onto the fourth record- ing tapes was designated the "Fourth generation" recordings. Note that within those recordings only one repetition of the Unknown voice samples was included. The inclusion of only one repetition closely resembles the forensic field conditions where the Unknown speaker makes only one non-repeated statement. The production of two copies of the Fourth generation recordings was necessary to provide the trials of voice identification to two dif- ferent examiners (Figure 5). Therefore, a fifth recording was constructed by dubbing the Fourth generation recordings from the line output of an open reel tape recorder (Magnacord 1022?) to the line input of a second open reel tape recorder (Model 700). In total, the Fourth generation recordings were dubbed twice from a l/Z-track tape recorder onto 7" open reel tapes (Scotch 206 1/4" x 1.5 mil) at 7 ips-full track. That pro- vided two different examiners aural materials to compare during the same period of time. Those tapes were designated the "Fifth generation" recordings. Care was taken throughout the recording process to be certain that the record and playback heads were demagnetized and free from oxide. In addition, tape guides and pinch rollers were routinely cleaned to assure the best possible tape handling. The signal-to-noise ratio (-38 dB) did not measurably decrease with the increase in recording generations 2 through 5 (Figure 6). This was accounted for by the high quality of tape recording equipment and magnetic tape employed in the voice record- ing process. 26 Magnacord Re. T outpuN Sound Spectrograph 700 0 a 2 ac ° DDDD .\ O. \ ) line input Figure 5 Schematic Diagram of Equipment Connections to Produce the Fifth Generation Tapes 27 All .33 05:. 223350 95:33. 223:0 oz 3 295m :8on 25m 2: .8 >235 32:33 00323 2 c. GO O.0.< _ _= a. m U 5:92:25... < .r N 032...... a a . 5.33.. A. n -2. [OAS-I “DIP." oauopguoo ‘0 41 and 240 Hz to 2000 Hz frequency bandwidths for each examiner in match trials. The results of that test, summarized in Table 2: indicated that as a group there was no significant difference between the 240 Hz to 4000 Hz and 240 Hz to 3000 Hz frequency bandwidths. There were signifi- cant differences between the control group and the two most restricted frequency bandwidths at the d- 0.05 level. These results suggest depen- dence between the 240 Hz to 4000 Hz and 240 Hz to 3000 Hz frequency band- widths, but not for the two most restricted frequency bandwidths. With this information, Hypothesis 1 would be accepted in the case of 240 Hz to 3000 Hz, but rejected in the case of 240 Hz to 2500 Hz and 240 Hz to 2000 Hz frequency bandwidths. Further tabulations of the examiners' correct, error and no- decisions were conducted for each frequency bandwidth in match trials (Table 3). The number of errors was measured in a one-tailed confidence interval test to determine whether these errors were within an acceptable margin of error (arbitrarily set at 11). The errors of examiners C and D at the most restricted frequency bandwidth.were significant at the °(- 0.05 level (Z - 3.61 for examiner C, 2 - 5.66 for examiner D; N - 24). These results suggest the most restricted frequency bandwidth may elicit significant error responses from less experienced examiners. The proportion of correct decisions as a function of frequency bandwidth sup- ports the notion that correct speaker identification is possible within the frequency bandwidths tested. The results of a Spearman Rank-Order Correlation(rs) between fre- quency bandwidth and examiner responses are presented in Table 4. These results indicated the probability the true relationship of frequency bandwidth and examiner responses is zero (no relationship) is one chance 42 ‘00. 82-33 ...-:00 2.2. 3...... e. 3.3:... o... Eons... ... :23... ooodv. .o .32....2- u I 00... 00.0 ¢m._ 0.. .0 3.3.. ... 28.....2. a O ...md. ..md 8.3.... 2.2.. o.~ 2 3.2.58 0... I a u 3.3.. «Cd gown mod... new 2 8.2.58 o... o..." 2 3.2.53 0... 32.5 .53.. ..e... .2330... .23: 522.com 35.62... .8 «2...; 22¢ .22.. ... ...X. N 030... 43 no.0va 2.3.2.3.» u .1 305603 359:. n o no Cassexu 3:23:95. u m :4 ¢~ cu cu cu ta cu ¢N cm cu cu ¢~ ¢~ cu cw ¢N v.9. .20... oo_~ .omu o_:o moto Amvucozawm o o o o o o o o o o o o .n .N o o .omszofiwum em ¢~ mm mm mm cm _~ Nu 0. mu m. m. n. «N x. o. 30: 2.2303 8230 dufi.‘ Goad defifl aomd_ 22.—seam.L - u: 000.. 2 m10¢~ wagon 2 "room a: 009 a. mzm¢~ u: OOON 2 u: ova 2.23:0 3 523390 35:92.... ......» 3 2.2.35 .8563“. :25 ..8 5232.5 3532.... 3 5:33... a mo 3.. 2230352 ...... 3...... S Btu .898: 8230 3 39.52 .32.... 5232 n 2.3 44 Table 4 'r,‘ Values for Frequency Bandwidth and Examiners‘ Decisions in Match Trials Spearman Level Examiner Coefficient of 'r.’ Significance A 0.4577 0.00: B 0. 6 7 I 9 0.001 C 0. 3637 ' 0.00: D 0.5269 0.00: A-B a Professional Examiners C-0 8 Trainee Examiners N 8 96 Trials/ Examiner 45 in a thousand for this sample for each examiner tested. The magnitude of relationship between frequency bandwidth and examiner responses is 'mildly strong.‘ These results suggest that the decrease in frequency bandwidth accounts for some, but not all, of the variation in the exami- ners' responses under match trial conditions. Further analysis continues with frequency bandwidth in no-match trials. The number and percentage of five alternative decisions for each examiner as a function of frequency bandwidth in no-match trials is pre- sented in Table 5. The distribution of these decisions indicated that as the frequency bandwidth decreased the mode level of confidence shifted from the positive to the no-decision level for examiners A and C, proba- ble to no-decision level for examiner B and no change in the level of confidence (positive) decisions for examiner D. To characterize the relationship of frequency bandwidth and each examiner's level of confidence, the median scores of each examiner were graphed as a function of frequency bandwidth in Figure 10. The median scores indicated that the examiner's level of confidence generally remained the same for the 240 Hz to 4000 Hz and 240 Hz to 3000 Hz fre- quency bandwidths. More dramatic changes were observed as the frequency bandwidth decreased to 240 Hz to 2500 Hz and 240 Hz to 2000 Hz. The most dramatic change was noted between 240 Hz to 3000 Hz and 240 Hz to 2500 Hz. Examiner D did not follow these patterns producing a majority of positive level of confidence decisions throughout the frequency band- widths tested. To test Hypothesis 2, a Chi-square (X2) test of independence was performed on the distribution of examiner's level of confidence decisions for 240 Hz to 4000 Hz and 240 Hz to 3000 Hz; 240 Hz to 2500 Hz and 240 46 532.com 35:99.... .0 5:05.... o 3 2.03.3... {3.836 ..oom .8 3...; :23: ......E 5.02-02 o. 2.6.... £232.00 ace-aeo.m\..o.... vu .- 2 22.626 :50... a are .28.. :8... .3532... .23: 32.336 3.3.3.3... u ma< o... o.» QN o.~ 0.. 06 h - - n - n a . . . . . . N .3303. .. n no: 5.355.... v 2.32.. l I 5.35.55 c ......aoa [us-I umpeu esuepg 1was) #0 47 000.0... a 0-0 0.02880... n 0... 30000.023 0.0:... 02.33... 0... 000.300.. .0 3000.2 . 0300:... 0:... 00.0. 0... .... 02.30030. 0.0 «00200030 ox. .0.0h 8.355.... -0 2.0.8.. 52.2.5... ..V 2.2.2.. 020.00.. .. 02 .. m 00:02:30... i N 0.02.0.0 00:02:23... ... 02:00.... .3. .3. .3. ...... 0.0.3.03. .3 0.3.... ... 0... 0.303....3. ....~. 30...... .m. .0: ......0. .n. 8.. 3.. .0. .... .0. 6.20.. ... .0. mm an N. on .0. mm N. N.» om ...m .0 mm 0.. N... v mm ... .... ...... 3. $10.. ...... .0: .N. .0. .... 8.. 3...... .... .... .V mu mm nm .N N.» 0.0 N.‘ m .N 0.» N0 _N .N mm mm ... .... .0. .0. .0. 0.. .... .... .0. 3.3.. .0. ... 3...... .... ...EmNmN m_N_N m. .mmmOO mm ¢mm~mmw .0. .0. .0. .0. .0. ... .0. .0. ... ... .0. .0. .0. .n. .0. .0. ooooo¢oo v¢ooo.~.oo .0. .0. .0. .0. .0. .0. .0. .0. .0. .0. .0. .0. .0. .0. .0. .0. 0000 0000 0000 0000 oom<0o.m<0om<0om< . 030.585 o.¢-0¢~ 0.m to¢N m.~ rowm o.N .. O¢~ .~...v. .2: 0.230000 3.30.3... 30002.00”. .0 3.3... 20_m_0wo 30.0.0.6 000m. 3000.3... .0 00:00:... 0 0.. 2.0.0.000 m .0 002.3030 000 30052 3. .0335... 03.2.3.2 .220 5.22 ...z m 0.0 0h 48 Hz to 2000 Hz frequency bandwidths for each examiner in no-match trials. The results of that test, summarized in Table 6, generally indicated that there was no significant difference between the control and test frequency bandwidths at the 0" 0.05 level of confidence. These results suggest dependence between the distribution of examiners' responses in the control and test frequency bandwidths in no-match trials. With this information, Hypothesis 2 would be accepted at the ct- 0.05 level in the cases of 240 Hz to 3000 Hz; 240 Hz to 2500 Hz and 240 Hz to 2000 Hz. Further tabulations of the examiners' correct, error and no- decisions were performed for each frequency bandwidth in no-match trials (Table 7). The results of a one-tailed confidence interval test revealed the amount of errors of Examiner C were significant at the 240 Hz to 2000 Hz frequency bandwidth, but not at the 240 Hz to 2500 Hz and 240 Hz to 3000 Hz frequency bandwidths (Z - 5.66 at 240 Hz to 2000 Hz, 2 - 1.59 for 240 Hz to 2500 Hz, z- 1.59 for 240 Hz to 3000 Hz; N - 24) at the 4- 0.05 level. The amount of errors of examiner D at the 240 Hz to 2500 Hz was not significant at the d- 0.05 level (2 - 1.59; N - 24). These results suggested the most restricted frequency bandwidth may elicit significant error responses from less experienced examiners. The proportion of correct decisions as a function of frequency bandwidth suggests correct speaker eliminations are possible within the frequency bandwidths tested. The results of a Spearman Rank-Order Correlation(r8) between fre- quency bandwidth and examiner responses are presented in Table 8. These results indicated the probability the true relationship of frequency bandwidth and examiner responses is zero (no relationship) was a median value of one chance in 73 for the four examiners as a group. The magnitude 49 0000 0.0:! 30.00 .00.. .000: 0. 020.003 0.0 0.000... .0 000.000 000.0v 0 .0 300.200..- .. I no.0v0 .0 2003.030 .- C 00.0 00.0 00.0 00.0 0.0 2 2.2.08 0.0 I . . 0 00.. 00.0. 00.0. 00.0 0.0 2 3.808 0... - . N . 3.0 0.0 0.0 00.. 0.0 2022.58 0.0 . 0 0 0:0. ...-.... a u m 0.00—0.30 300.000.... .000... 22.. .222 ....z ... 0.0.3005 ..000000... .0. 000.0> .0X. 0 0.00... 8'7 50 00.0 v. 0.8.0.3.. . ... 0.00.505 000.0... a a no 200.606 300.0030... 0 m |< VN tN VN VN ¢N VN *N ¢N VN VN VN VN tN ¢N ¢N VN . .20.. ..0 w m N 00 v N mN_o .mm... :3 00230.0.” 00 O o O .0 O . .0 o o0noo .NG:000.0.009 .0..w nN ON a. o. NN m. m. ON .N c. N. o. nN n. o n. 2060. 000.0.00a 32.00 0 0 0 0 aldflu 0 a ulna 0 0 jldl! 0.00.0.3. . Tllllllll Jl 0.. 0000 2 £000 £0000 0.00000 020000 2.0000 £0000 22.000 .........0 .0 0.0.3000m 300000... - 00.0w .0 0030.008 30.0.0.0 000m .0. 0.0.3000m 300000... .0 00:00:. 0 00 .m. 000.0.000s02 ..00 .0000...0..m .20 0009 30:00 .0 .3052 ”0.0.... 00.02-02 .. 0.00... 51 Table 8 'r,‘ Values for Frequency Bandwidth and Examiners‘ Decisions in No-Match Trials Spearman Level Examiner Coefficient of 'r.’ ' Significance A 0.2366 0.02I B o. 3 489 o. 00: C o. 1574 0. I26 D o. l 083 0.292 A-B a Professional Examiners C-D 8 Trainee Examiners N I 96 Trials / Examiner 52 of relationship between frequency bandwidth and examiners' responses is 'weak.‘ These results suggest that the decrease in frequency bandwidth accounts for little of the variation in the examiners' responses under no-match trial conditions. In retrospect, these results suggested that decreased frequency bandwidth affects the examiners' level of confidence decisions by gen- erally shifting from a higher level to a lower level. Experienced exam- iners did not produce any errors, whereas the less experienced examiners produced errors of false identification and false elimination, some of which were significant at the most restricted frequency bandwidth (240 Hz to 2000 Hz). Talker Sex The number and percentage of five alternative decisions for each examiner as a function of talker sex in match trials are presented in Table 9. The distribution of these decisions indicated that examiners A and B produced more decisions with a lower level of confidence for female than male speakers, examiner C male more than female, and examiner D generally indifferent. To characterize the relationship of talker sex and each examiner's level of confidence, the median scores of each exam— iner were graphed as a function of talker sex in Figure 11. The scores indicated that the examiner's level of confidence was generally divided between examiners A and B producing more lower level of confidence deci- sions for female than male talkers and examiners C and D male more than female speakers. The amount of difference between the levels of confi- dence for male and female talkers does not appear as strong for examiners C and D as it does for examiners A and B. The amount of difference in level of confidence suggested no difference between male and female 53 000.0... 0 o lo .000.000.0.0 0 0-4 0002.00.00 0.0:: 00.00020 0.0 0020.000 .0 .0005: 0 0.2.0.00 .0..0 00.0. 0... .... 030000.00. 0.0 0000.03.00 .... .00. .00. .00. .00. .00. .00.. .00. .00. .390 00..00.E=mr .m. .m. .02 ...... .m. ...... .m. .m. 20.3.. 0 .0. .0. 8. 8. ... .0. .0. a. 5:025..- 0 0 0 0 0 0 .0 .0 .3000... 0 .... .0. .00. .0.. .0. ... .0.. .0. 00.0.000r02in 0. 0 00 .00 .... 0 00 0. .0.. .... ...: .0.. .0.. .0.. ...0. .0.. 00.30....003- 00 00 00 00 00 00 00 0 .332. 0 .00. ...n. .... .00. .00. .00. .0.. ..00. 00:00....000. .. . No .... mN 90 0.0 mm 5 030 02.700 0 U m < O 0.0:450xm 3052.000 0.060.. 0.02 .0 .20... 00m .0....0... ZOE-own .00.E00m 000m .0. 000 .0..0. .0 02.0000 0 00 000.0.000 00:00.22 0 .0 00200200 000 .0003: "as—...... 50.02 m 030... 00m .0..0.0 .0 00:00:. 0 00 0002000 0.000003 000m .0. 000.0> 00.00... 3.0.... 00.0! .. 0.00.0 000 ..00.0.. ~32... 00 0 a 0.00.0.0: 000.0... a 0-... 00m .0..0... 0.00.305 .000.000.0.0 0 o-< madam“. u..<2 . «I . . 0. 00:000....2 A- . 03:000 0 . . /\r\0 \\ \ a 0 s 7. .. 0 0 .. 02.8.0.5... -.. N 03000.0 00.0.000 ..w m .. 0.. «009qu la nae-l uogpsw 55 talkers in match trials of voice identification. To test Hypothesis 3, a Chi-square (X2) test of independence was conducted on each examiner's distribution of decisions as a function of talker sex in match trials. The results of that test, summarized in Table 10, indicated that there was no significant difference in three out of four examiners' decisions between male and female talkers at the c(- 0.05 level. With this information, Hypothesis 3 would generally be accepted at the at- 0.05 level of confidence. Further tabulations of the examiners' correct, error and no- decisions as a function of talker sex in match trials are presented in Table 11. The results of a one-tailed confidence interval test found the amount of errors of examiner C for male speakers was significant at the Gt- 0.05 level (2 - 2.20). The amount of errors of examiner D for male speakers was not significant (Z - 0.75) but was significant for female speakers at the d- 0.05 level (2 - 2.20). These results sug- gested significant errors may be elicited from less experienced exami- ners for male and female speakers in match trials. In addition, the pro- portion of decisions indicated correct speaker identifications are pos- sible for male and female speakers. Further results are now presented for the talker effects in no-match trials. The number and percentage of five alternative decisions for each examiner as a function of talker sex in no-match trials are presented in Table 12. The distribution of these decisions indicated that examiners A, B and C produced more decisions with a lower level of confidence for female than male, but examiner D produced more decisions for male than female speakers. To characterize the relationship of talker sex and each examiner's level of confidence, the median scores of each examiner ...00 0000 0.0.... .00.00 .00.. .0000 0. 00.00020 0.0 0.0002. .0 000.000 000.000 .0 200.200.» a m 00.000 .0 38:20.. . O 56 I J H.295... 0. 00.00030 .Nd ¢md wad mod. I m...<: _ N N o 0.w0.0.0xmm < 00m .0..0» 20:. 020: 0. .30 .0..0... 3. 82.; JX. 0. 2%. 57 2.00.503 000.0... 0 one 0.00.0.0; .000.000.0.0 0 m..< 3.0x. .0 28:20.. .0 00 00 0.. 0.. 0.. 00 0.. 0.. .0.... .. 0 ON 3 m . N. w .n . 000.0.000 I02 00 o o o _ 00 o o 32080 Ba v. 3.5 mm m¢ 0N .Vm am By mm N¢ 0020.000 .Na: .0830 a u m 0 a o m < 0.00.0.3“. .00000..000 m4 Sm .. s. .0 .0. < .... 00m .0.:00 u < 22203.34 00.... .0..0... .0 00:000.. 0 00 30.0.0.6 000m. .0. .n. 000.3000 ..02 000 .n 000 ... 3.5 ..N 000 .. .0230 .0 300.02 3.0.... 00.02 _. 0.00.. 58 000.0. ... 0 o u o 0300.00.00 0.0...» 00.00020 0.0 000.0.000 .0 .0200: 0 0.00050 .0..0 00.0. 00. .0 00.000200. 0.0 802000.00 00> _Deo....uo.k I a I < .0... 30. .00.. .00. .00. .00. $3 .0... HRH .9... .8. .z E. .3. 83 .... ...: cefizeku -0 0. 00 0 0n 0.. 00 0 .000 0. a 0: .0. .... 0... .... a: as 3.. 5.0.0.0...- 0. 00 .1. 00 0 .0 00 0.. .332. 0 a; an: “may Ah; an. 2: .0: am: :O_Q.OOOIOZIm 0 .0 00 00 : 00 mm 00 .2 .3 8. 8. E a. 8. .3 8:80.050.-. 0 0 o 0 0 0 o o .332. .3 .2 .2 .3 8. .2 8. .2 00080.23.-. 0 o o o o o o 0 00.0000 0 0 m < 0 0 022M020 00002.08 0.0.000 0.0.2 . .o .26... 50 .308. 220.000 L 30.0.0.6 0000 .0. 00m .0..0; .0 000000... 0 00 0030.03 00:00.3... 0. .0 000.000.00 000 .0005: "0.0.... 00.02 ..02 N. 0.00.. 59 were graphed as a function of talker sex in Figure 12. The scores indi- cated that the examiner's level of confidence was generally lower for female than male speakers for examiners A, B and C, but indifferent for examiner D. The amount of difference between the levels of confidence for male and female talkers does not appear as strong for examiners C and D as it does for examiners A and B. The amount of difference in level of confidence suggested no difference between male and female talkers in no—match trials. To test Hypothesis 4, a Chi-square (X2) test of independence was performed on each examiner's distribution of decisions as a function of talker sex in no-match trials. The results of that test, summarized in Table 13, generally indicated no significant difference for all of the examiners' distribution of decisions between male and female speakers at the at- 0.05 level. With this information, Hypothesis 4 would be accepted at the d- 0.05 level. Further tabulations of the examiners' correct, error and no- decisions as a function of talker sex in no-match trials are presented in Table 14. The results of a one-tailed confidence interval test indi- cated that the amount of errors committed by examiner C for male and female speakers was significant at the d- 0.05 level (Z.- 2.20 for male and Z - 3.66 for female speakers; N - 48). The amount of errors committed by examiner D was not significant at the e(- 0.05 level (2 0 0.75 for male speakers). These results suggested significant errors may be elicited from less experienced examiners for male or female speakers in no-match trials. Also, the proportion of decisions indicated correct speaker eliminations are possible for male and female talkers in no- match trials. 60 000 .0..0... .0 02.0000 0 00 0020.000 0000.00.00 0000 .0. 000.0> 00.00... 3.0.... 00.00702 N. 0.00.0 000 .0..0... ‘03.... 00 0 z 0.00.1000 000.0... 0 0-0 00m .050... 0.003000 .000.000.0.0 0 out 0.3200 0...... T . . . .‘ 00.3000 .1 n .02 a . / 03.000000 .0 0.0 000.0 " I m 02.00.53 A... 03:30 souspuuog ,o mas-1 uogpsw 61 Table I3 'Xz' Values for Talker Sex In lilo-Match Trials Talker Sex A Examiners B C MALE com pared to FEMALE l .62 2 2.26 .O.64 5.49 Q s significant at p(0.05 Degrees of freedom are presented in upper right corner wlthin each cell. 62 0.00.033 000.0... 0 are 0.00.0.00u 300.0030... 0 0:0 00.00.. .. 0.80.8: .. e 0? 0.. 0' o? no. at at at .03... _ 0. mm ... 0 z 0. N. .n . 2.2280 IOZ 0 on o o _ 00 o o .8308 .N 6 .. .0..u .... mm mm .m «0 mm mm mm 000.0.000 .000. 32.00 a o m. < a u a < 0.00.690 .8000..000 l .0 .30... u..<.2mu_ xow .0..0... m..=O¢bO=< D .0 GOO—cogm ‘8 500532 «20th. gnu—62:02 N- ...5 .33... ...... 5233.3 353...... .o 5:95... a 2. 22.596 852... 2... 35.382... .3 2.03.30 030... 70 110'- Upper Frequency Limit (Kits) 2.0- U pper Frequency Limit (KHz) Figure l5 O. .0. .- O... O. 0...... to 3.0 — 4.0 - \ No- Match Trials: “#1,, Distribution of Professional " one Trainee Examiner‘s Decisions For Mole Talkers \ p 00‘ 9.9 4'4 (a \ , e, E- \ \ <‘o \ Number ’9 at \ \ Decisions ’0 \ O x 1 0 x \ ’O 06. 0 ha I 6’ ”I; 60 ‘0 C a I.’.q 6’. ”fl-v. Level of Confidence r 0 4,4? ‘29 \ es \ ...::.°: ’0 “°°:. \ Number \ Decisions ’4' \ 0 \ 1 \ \ 0) '6, '09 6'6 0., e e lye" ’e , ”b. 3 Level of Confidence 71 *‘o ......”OO / //// 2.0 - \ \ ..°.°.'...-. "2:32.22... \ Number ’0 2.5 ’ o'e: ..-:.e.e.e 0:... 0' \ \ ‘ I. ...... 3.0 -- \ ' \ \ 0 Limit . \ \ \ (KHz) ‘ \ \ no", 0'06 \ a. \ 0'“. IO) ‘0 I a v. I “I. a 0a,,” 6,. 01”,. 3 Level at Confidence r 04’” 6‘43. \ . a, II \ a O 2.0 \ Number ’o 2.5— of \ Decisions ’a UN \ 0 "W lelt \ (Km) ' gun is No- Match Trials. ‘05,” Ar... 0. 0 ,0 0 Distribution at Professional ‘3 , 4!. 8 Ge,” '6’. 30,-, and Troinee Examiners Decisions ° ' o '0' POW". 70"." Level of Confidence 72 To test Hypothesis 6, a Chi-square (X2) test of independence was performed on the professional and trainee examiners' five alternative decisions as a function of examiner training controlling for talker sex and frequency bandwidth in no-match trials. The results of that test, summarized in Table 18, indicated a significant difference between pro- fessional and trainee examiners at the 240 Hz to 4000 Hz, 240 Hz to 2500 Hz and 240 Hz to 2000 Hz frequency bandwidths for male and female talkers at the 4- 0.05 level. There was no significant difference found for the 240 Hz to 3000 Hz frequency bandwidth for male and female talkers. These results indicated there are differences between the distribution of level of confidence decisions for professional and trainee examiners under conditions of male and female talkers in no-match trials. With this information, Hypothesis 6 would be rejected at the d- 0.05 level. The previous results generally indicated professional examiners produced significantly different distributions of decisions than the trainee examiners, producing lower level of confidence decisions in match and no-match trials. Typically, the professional examiners were more conservative decision makers than the trainee examiners. The fact that professional examiners produced no errors of false identification or false elimination is significantly different from.the trainee exams iners who produced both or these types of errors. Discrepancy of Results Inspection of the raw data for intra-examiner differences revealed instances where a higher level of confidence decision was produced with a reduction in frequency bandwidth. One may observe these instances by contrasting the level of confidence decisions for one examiner across all of the frequency bandwidths for one talker in match or no-match 73 Table I8 ix?“ Values for Examiner Training in No-Match Trials Frequency Talker Sex Bandwidth "ALE FEMALE... 240 to 4000 Hz .8.4l 2 . 8.60 2 24a to 3000 Hz 5.32 2 4.65 2 24omzsoonz ’I l.53 2 . 9.69 .240 t02000 Hz 0 io.42 2 0 l4.0i 2 Q. significant at p < 0.05 .a significant at p < 0.005 Degrees of freedom are presented in upper right corner within each cell. 74 trials. Tabulation of these cases for each examiner is presented in Table 19. A closer look at the speech spectrograms for these particular cases indicated some attenuation of speech frequencies between 240 Hz and 800 Hz was noticeable in the less restricted frequency bandwidth speech samples (Figure 17). The attenuation of speech frequency infor- mation may have influenced the examiners' decisions in border line cases (e.g., between a positive and probable level of confidence). It seems reasonable in this instance to conclude that the examiners had truly ordered the speech samples from greater to lesser amounts and/or quality of speaker information. Discussion ggfects of Reduced Frequency Bandwidth The results indicated that correct speaker identification and elimination were possible under restricted frequency bandwidth condi- tions (Tables 3 and 11). Decreased frequency bandwidths speech samples lower the examiner's level of confidence in match and no-match trials of voice identification. However, in a few instances a decrease in fre- quency bandwidth elicited decisions with a higher level of confidence than a wider frequency bandwidth. Also, the few errors of voice identi- fication and elimination were produced by the less experienced examiners using the most restricted frequency bandwidth samples (240 - 2000 Hz and 240 - 2500 Hz). The more experienced examiners (certified professional examiners) did not produce any errors of false identification or elimina- tion, but rather no-opinion decisions when using restricted bandwidth samples. Therefore, it seems important that examiners should be tested for normal hearing. A loss in the high frequency hearing certainly will interact. 75 22... 2322.62 .6 =26: ... 32.526 x N» u 2 262526 3......» n o i .... 82.386 35.328“. a m i < N o . W a a m 6 3429. w a a a .. o n.2,. M. m. a 32826 a < «.... W W o r u- m N 0 ¢ w4<2mm l. J .V n N O . m4<2 m. a 9.9.2..” o < xaw 3%? 5232.6 352.2... .23.. o 2 .330 no: 3.32.50 .6 .26.. .23.: a 5...? ... nonoo .o 32:33... ...... .3532 m. 036... u§.oo..ooom goo-am £23230 3.8.3.... 20.3.5 N :3... .2955 76 uouonuuw goo-am ......m a... 3 52.22.30 .... 3.3... T ...-3 cl... \mfl om .d,_ an. n1r¢,.\ , . 2.0 . _: r 7 :L— ”jazz: 18.. :m ; .... : ... ... 9—. __ __ . . . .oo.~ _ _ it, - I ! 8d 2-2.2391...I.-=:.¥z+:i.+¥:+rt It «..-: . ...... $2,232.. ...tzviooé .... Mona 2 3“ a... OOON 9. OVN ‘— (zux) muonbug 77 Effects of Talker Sex The results indicated female talkers were identified or elimi- nated at a lower level of confidence decisions than the male talkers, but the difference was not statistically significant either in match or no-match trials, for combined responses from four examiners. However, data from a particular examiner yielded a statistical significant dif— ference from male and female talkers in match trials. Effects of Examiner Training_ Professional examiners performed differently than trainee exami- ners in both match and no-match trials. Professional examiners did not produce errors of false identification or elimination; the trainee exami- ners did. However, trainee examiners produced correct identifications and elimination in more cases than professional examiners. In other words, professionals used more often than trainees the no-decision option. These results indicated that professional examiners are more prudent in their decisions than trainee examiners. CHAPTER V SUMMARY AND CONCLUSIONS Twelve male and 12 female speakers were recorded over the tele- phone repeating four English sentences. Two recording sessions were held for each subject separated by a one-month duration. All subjects spoke with an American Midwest dialect free from marked phonetic idiosyn- cracies. The first recording session was designated the unknown talker speech samOles and the second recording, produced one month later, was designated the Known speech samples. Both Known and Unknown speech samples were originally recorded onto cassette tapes. The phonetic mate- rials were later dubbed onto magnetic 7" reel tapes for experimental manipulations. The Unknown speech samples were filtered four different times through four different frequency bandwidth settings: a) 240 Hz to 4000 Hz, b) 240 Hz to 3000 Hz, c) 240 Hz to 2500 Hz and 3) 240 Hz to 2000 Hz. The Known speech samples were filtered one time through the filter settings of 240 Hz to 4000 Hz. These filtered recordings were analyzed on a sound spectrograph and re-recorded onto magnetic 7" reel tapes. Speech spectrograms (96 Known samples and 768 unknown samples) and their corresponding aural components (6000 feet of magnetic tape) comprised the total material content of the trials of voice identifica- tion. These materials were organized into 192 discrimination-type trials of voice identification, half including the matching speaker and half not including the matching speaker. 78 79 Four examiners, two professional and two trainee, each examined the 192 trials by the aural-spectrographic method of examining speech samples. Each examiner selected one of five alternative decisions after examining each trial. These decisions were the following: 1) positive level of confidence that the two voices belong to the same person, 2) probable level of confidence that the two voices belong to the same per- son, 3) no-decision one way or the other, 4) probable level of confidence that the two voices belong to different persons and 5) positive level of confidence that the two voices belong to different persons. The results of these tests were tabulated and analyzed to determine the effect decreased frequency bandwidth, talker sex and training had on the exami- ners' ability to make correct speaker identifications and eliminations by aural-spectrographic examination of speech sample. Six experimental questions were asked: 1. Are correct speaker identifications possible under restricted frequency bandwidth conditions? 2. Are correct speaker eliminations possible under restricted frequency bandwidth conditions? 3. Are female speakers more difficult to identify than male speakers? 4. Are female speakers more difficult to eliminate than male speakers? 5. Do professional examiners perform differently in tests of voice identification than trainee examiners? 6. Do professional examiners perform differently in tests of voice elimination than trainee examiners? The salient results of this investigation may be summarized as follows: 1. Correct speaker identifications are possible under restricted frequency bandwidth conditions. 80 2. Correct speaker eliminations are possible under restricted frequency bandwidth conditions. 3. Female speakers are more difficult to identify than male speakers. 4. Female speakers are more difficult to eliminate than male speakers. 5. Professional examiners perform differently in tests of voice identification than trainee examiners since the professional examiners produced no errors of false elimination. 6. Professional examiners perform differently in tests of voice elimination than trainee examiners since the pro- fessional examiners produced no errors of false identi- fication. 7. There was no significant difference between the 240 Hz to 4000 Hz and 240 Hz to 3000 Hz frequency bandwidths in match and no-match trials. 8. There was no significant difference between male and female speakers in match and no-match trials for the range of frequencies 240 Hz to 4000 Hz and 240 Hz to 3000 Hz for professional and trainee examiners. 9. Professional examiners produced no errors of false identification or elimination whereas the trainee exami- ners did produce errors; 2.62 false identification and 3.12 false elimination for all of the trials of voice identification. However, for 240 Hz to 4000 Hz, the trainee examiners produced 0.02 errors of false identi- fication or elimination. The results of this investigation are in general agreement with the studies of Smrkovski (1976) and Tosi and Greenwald (1978). These studies indicated that the experience of the examiner had an important effect on the validity of results of a subjective method of voice identi- fication based on aural and spectrographic examination of speech samples. In addition, the results of Tosi and Greenwald indicated the professional examiner's unwillingness to make a determination of same or different speakers when the speech samples were masked by noise. The same unwilling- ness was observed in this experiment except the reduction of speaker 81 information was produced by restriction of the frequency bandwidth. The following conclusions appear warranted: 1. For professional examiners restricted bands of speech frequencies do not lead to errors, but rather, to an increase in the percentage of non-opinions. 2. Male and female talkers are essentially identified and eliminated at equal rates. 3. The training of the examiner has an important effect on the validity of results of a subjective method of voice identification based on aural-spectrographic examination of talker samples. Implications for Further Research Other factors which were not tested in this study which could be explored are the effects of decreasing the frequency bandwidth from 240 Hz and 4000 Hz to 3000 Hz and 4000 Hz on the examiner's ability to make correct speaker identifications and eliminations. Also, the effects of restricted frequency bandwidth with differing durations (e.g., 10 seconds, 20 seconds, 40 seconds of the unknown speech sample) on the examiner's ability to make correct speaker identifications would greatly enhance the field of voice identification. APPENDIXES APPENDIX A EVALUATION OF SUBJECTS' VOICES 82 APPENDIX A EVALUATION OF SUBJECTS' VOICES A.vocal rating of all subjects was performed to determine whether this sample group of subjects were representative of a normal speaking population for the Midwest Region of the united States. The unknown speech samples recorded in the third generation tapes (240 Hz to 4000 Hz) were played back on a tape recorder (Whllensak T 1500) into a set of headphones (AKG Kr240). A speech pathologist listened to each subject repeat the sample sentences and concurrently rated their speech on 10 vocal dimensions. Based on the ratings in the 10 dimensions, the speech pathologist produced an overall rating representing the subjects' over- all speech behavior. The results of this rating process indicated no marked phonetic idiosyncrasies by any of the subjects. The overall con- clusion of the speech therapist was that the group of subjects was repre- sentative of a normal speaking population for the Midwest Region of the United States. Ph -.fects of Decreased Frequency Bandwidth on Speaker Zcexticheticn 2' lurel .3: Visual Braziaation of Speech Sezples 7033 223533103 33213} 3333? Dimension Sea? '23333 Too sort \n O\ 7 £30 loud r4 04 N PITCH Too low 0\ \‘I oa 0 0 high ‘0 mafia ..J P503 Eli: Too 'reathy oapaazzzssu. zoo far 1 2(2):. 3 harsh U. kn U: 0. ‘0 0-3 0 0 0\ \l H O 0 far forward aascxaxsg back XASAL Too 1 2 353C3333£ denasal PAP: Too slow 1 2 3 @5 52:52:; 53:2? 1 2 @h 5 VARIETY foo little 2 VCZIiL 3.3223. Eastern 1 2 3 ® 5 c 2:35:3st " Slurred 1 2 3 A. @ 6 OVERALL Unaccept. l 2 3 5 5 65> 5 5 7 Too nasal Os Too fast Patterned Too much u {=- U. O\O\O\ South :n Over :ecise Qflflfl-QK) Excellent Conner-.23 for Speaker Number if ? ...—nee .. ... “LIE-coed; I J 0 0 m 0 r n ..J M \n 0 ‘9) t 0 0 f4 and v o I n ‘Z‘Z‘ZH foo lot: 1 2 Q) “r 3 O ; Too .1511 - e c - .. o / r PHCL.2;.. Poo areethy l 2 3 5 0 / Too harem -.. . 'eO-I - - a G ’ H . Oirzaza-¢lia rec fer 1 2 3 S 0 ( Too :3: forva:c 3330:3553... back SAEAL Too 1 2 3 5 5 7 'Too nasal 33:62.;33 ienzzal 2...... can 319:: 1 2G) 5 6 7 Too fast a:-:z::-::.. Jerky 1 2 5 6 7 amen-.3; 73.2--2 loo little 1 2 5 5 r Poo :uch iC._L 1 :13. fastern l S 5 7 .Sou'hern 1 A ' ' ". ‘ 0. ' at..- 2 VH-I-\ye-.\o-b Jla-.2’ U| ‘0 (‘3 ‘3 .10 '1 0) '1 m U 0 Do (D .'.‘_-0 IO U U U U U q‘®c@@®“ ®®® -_) 1‘ O 0 u |U (. n O ..J l I ” . '_¢.n“-‘- H...- ll! I” O"? U vent 21:23:33: 1 ...‘fioe-v a “f..- ":‘r-xr \ doc-4-- PECNAEICM - O .-. e-v . OnPrLc. 12...; Julia =:=CQ q- c... o. v unav— \3e3-9 .‘OnaAU RZECTAMSE e-mr ...-1- U 227-313"- M- -.--o0 . -... VAfilllf .0 c o . , . 'C.;£h 3...";8. CONJCXAEZ ' OVERALL Comments for Dinensian '4 "5 soft 103 F‘ r: no N Too 'reathy .-J 0 Too far back Too 1 2 3 denesal Too slow Jerky Too little Eastern Slurred k‘ h‘ k! r. r: r4 Unaccept. Speaker Hunter D: 0 p ... (D 0‘ 0\ O‘~ O\ \3 ~e -q x: -d 'fi loud Too high Too harsh Too O 0 ...: . 3r 0 :'«'3- - Patterned Too much Southern Over precise Excellent r .0.... .Q s . aU..J...‘...—‘J --qQOO EEC::¢\::¥ :- —-.- .... ~ 0.. Cp- 0 . ...l 0...- .-.-J‘I ‘A -P .‘vnédh =:§’ qc‘. ‘ d | ne‘eovev.$c- .Ie -. we? ...-$-d 30...! qt.- . o'--—.qo, o‘-.o‘ ‘0 - ..Ip.. q. .-, 1&4...‘ .‘1..-Aoe Rho- . - ' v u d~e0‘h.a.‘o'§ eo-.‘ . o o 0-3.15... ~ ‘ A-— F». . OM- ~3:.JJ . to“ Zoo soft l 2 I 0 um. t.) N foo H [0 co treathy 0'” p "a J H H lo 0" Ix) Too denesal N 9 U ®~©® Zoo slow .0 lb eéwoe re r4 r4 r4 l4 :4 I) ' we on .11 2d " .- ’5 an -3. ..VBZJC. '- ( ID ,4 :A w ’1 t. H \D ‘I em- “UK” p @ \n L. {r (9 \fl \0‘ \n U: \n \JI Ox Ox U\ Cw. u . 0. u\ U . O \ (h ‘ \‘ \‘ -Q ~ \‘ \‘ \I \l '31- Ezfects of Decreased Frequency Bandwidth on Sceaksr Identification EJ Aurel n; Visual Exanination of Speech Serples y. . C-o-QOO,.“‘Q . ”Tic-q .O:-:?fi ICJ.XL 2&ov;h.;‘Uofi 30‘--.IJ J-odd- Dinensicn Scale Loatzzsss Eoo soft i®3 4 s o 7 3co‘ou‘ / u- PZTCH Too low 0 ’ Toc h g: l ’4- 5 Pz—zcmrzcz: Too breathy 1 @3 a 5 6 7 Too harsh oaassarissit Too far 1 2 3 (E) 5 6 ‘7 Too far forward 9350? “' back 3.3.3.3 Too 1 2 3 @5 5 7 Too nasal 333033333 denasal air: Too slow 1 2 (:> b 5 6 3.31.322; Jerky 1 2 369 5 VAEIEEL Too little 1 (:) 3 u. 5 Too much vowsi 5-21:. Eastern 1 2 3 u 5 5 5 tonnes: " Slurred l 2 3 ® Over precise 7 7 7 7 South rn 7 OVERALL Unaccept. l 2 ® ’4- 7 Excellent Connents for Speaker Number S;C;_£i \n (I ‘. V foo loui ()\ . Too high '0 H OJ 1) :q 0's 0 O .J 9 H H H H to uuee mm Ut ax \\ PhC;.E:.. Eco treathy 2 Too harsh OEPEA'E‘EE-JEAL foo fli' 2 <3 7 Too far :‘orrar'i asso::;._ sack 3.9.3.; Too 1 2 3 5 5 7 Too nasal 333C: .33 'enasal ELIE too slow 1 2 3 5 0 7 Too fast 333333; Jerk; 1 3 3 3 3 7 Patterns: 1" d ‘-‘.. 5... ‘\ *.%.'.l.'...’ .‘30 7.14.13 I 3 - 0‘s ‘3 0 i 0 0 H I .. ... - o - 3C--.nxs£ " Slarrea l 2 D 9 . Cver prezise o-o—a.oo to Q Gill... .n scest. l - ( l r .' o f \ O l. p (u u (u '1 :1 .4 lo U \J U u . . (.3 \n O\ *3 U) 0 9: i 5 \h ‘I (I H U U .1 . a u :3 H .7" ‘0‘0 5 rue e..ec.3 -. .3, 'a- :.*..*:.'!1: .1: U --Oy-nooaov «‘0 a- Q - -- 'I'CZAL as“... Cl”. -..... 86 PECSA3-L3 --..n :Q‘ou q — 0:. :13. .-..ulAL ". \Q' - Q' Q o isebzmh ...; \r 7 -v “Adda 7:”7""‘ -o‘dbo-.\a U..- cvmtn- “-.--... VAflIizz vousi 54:13. ' COKJGSAHE OVERALL Consent: for 200 soft Eco low Tao breashy Unlccept. .4 .«O 14 h‘ r4 t4 ho re DJ r0 (0 N NNNNN to Speaker {runner we? 0'. 0‘- 0‘ O\ 0\ O\ 0\ m0\ 0\ 300 Too Tee Tee . Tao Too Patterned m on much Southern 033313.333 .5511... 3:302: .335 . .1 1-~ ..-‘go. :00 103 Too treachy is {3" back k‘ r4 >4 t4 H l0 IQ '0 L) l') (0 \1\ mm \A A \n Us Us \ \,h 0\ 0\ U\ (h 0 \ 0'\ 0\ \J\ N) ‘l ‘0 Too Too Too Eco fhe -.-- 3 of Decreas-- Frecuency E ndricth on 523332: Zientificeticn a? Aurel and Visual Exxzihation of Speech Sanple ‘--o-.o-'q‘_o ..- ‘o— u- - ...0;.OV;&.O; 0.0‘2100J 30‘0": VCZS . O- .3 Q Q bOu‘JuSJ.) MR" PEC3A3I£H - .0 . .... . u.- . f a :Q yo .- Iii-30.0 "bl I a \. ‘3 V ' CONJChAhQ n OVERALL Speaker Hunter 0 J 0 0 M O n, (1’ ... O\ a“ loui NN high 0\ O\ \\ k‘ re to s» x» \» @® 0 e (365% u. U‘ UI breethy harsh ..J 00 hi Too fer back U‘ 0\ \, far forwnri Too 1 2 3 denesal U 0‘ \) nasal 0\ Too slow r) oo fest Patterned Too much South rn Over precise hi 3.) la n (D b: n D4 r' 04 r0 r4 04 h) n: n1 AS A: h) U U U U U U 0\ 0\ as as 0‘ ~: {.. - 9 & Unaccept. Excellent PEGZL‘CIC l. OEPHAEILJLAL, a ficaef . .v 31.0L..al« a Q. .o‘ : . , “flush :‘ v v -- ~~ 0.--». .o'v'd I J 0 0 Us 0 ’1‘ u n; \J ,\ F (a 0\ m 0‘- \‘ loui Too high 213.3351! 0": O O .4 O ‘1 k‘ r4 r0 r4 to N (u U U U {:1 ~ [I U’\ 0\ a“ far forweri been vac , 5 3 1"“. 5 6 7 'n‘ “ t. 1 iehesal ‘ Too slow 1 2 3 3 3 0 7 too fast Jerk; l 2 ”3 4 5 5 7 Petterne: Jeo liztle 1“: )3 ~ 5 5 7 Too :uoh . .- 1 ..\ p ' ‘- eastern l 3 J ‘«i» 2 9 1 Southern | '9‘ ‘ ’ slurred l 3 3 9- D 3 7 Over crezise In 232-3. 1 2 3 /-4\ 5 5 7 221*l1.n: fl ‘4' - i - ‘3 ihe ;:-ects o: Decreased : - O ' bl . Q . - - ’ - :annriuta on Sheena: ice“. - - . .' ' a ’ f9 , :y lurel en; lieua- -Jazin a. w ogeech azzples f. 0‘ ---o 1.00:- “? ..“TII- ‘ ICU-1L .[looo-‘oové-VOO 3“.-00J Scale 2.02.321 3.3.5 P-?CH \‘Q ‘3 - 9 blfloc‘U =7 3'... ' a. ‘ q ..--Vo-o‘o'v; 3313 C a o urns. -.--.0 VAflISIZ vowsi AHIIS. couscnse: * 01:23.11; Comments for Dinensisn '4 soft low r4 r4 to av I» breathy ..J ‘0 too far back Too 1 2 3 h denesel Too slow 1 Speaker Number §;;§ dr (A UN 0" (h C\ 0‘ O\ O\ U\ 0\ K) x) ‘Ni V V” loui Poo high hersn far forward Too much South rn Over precise Excellent f'ro—I-oo ..- “#5-“;43 ‘°finu On-.. — I .I q .- :hc“tx-:\- .4 3.2502: .1‘:4 _ . . o0A‘;e‘-h 7373"" ‘ ..-‘C‘. o 4‘ "-O,fio-o. ....--O‘e- ... - D ..' I.‘-~'.---- . “..., .-, g [\- 4..‘.‘-l 3..--40. QR ‘ '. "o. d~o'~‘.'l O 0".—:;25\L- 0“ ® 89 ‘0 ‘0 I.) l .n ( .J . A‘ : > 0 V .... ‘ a Ul‘-.2‘c‘ 1 ~ 3 4 ”n «n‘ ) - l q : I. be. .u~o-IV. . Has .‘an ° —I-a‘ 13;! .14- O-o- ...a.-—-- 0\ 0\ C\ ()1 ()3 (J\ (J. l)\ .“ \i ‘4 “J x! ‘l - ' ' "V‘ s - ure- 3.3-1.99 7"QQv. ;'-- -‘~“¢- cod .‘ un~ ‘ooerq’fi 3.3:: .0 J Sgt-ad- 89 VCZAL 313335133 «OJ—5 1“ fl - 0 P501 P r 3 ... .a ... r. . a u f. 1. . by h s 1. h. a... “a n 5 n. S 3 c u. n. 1.. u 5 a“ a nu a“ g-“ o h. C. r a ho .V. fl. ‘5 W— .w e O c O O u 3 W W O o. c o 0 x m... T 3.. ms ...c. m. 3 a; p:- r’ 7 7 7 7 7 7 ..O ’0 r0 6 r0 r0 6 5 5 :J {J :J 5 5 I ...a ® a me C ® ... ... 3 3 3 3 3 3 ® 3 2 7. 9. 2 2 2 2 l 1.. 1 l 1 l 1 1 1 n4 ...... o ...... w p. u... o. n 1. O n 9. O a. a. a l u. e l .3 a; 3 s ”J a c k a o W n m 0 3 C n . » w w nee five 0 a. 3 r a. a. «....3 «.74.. q¢ J 7a U C a“ C o _ . ... .. u ... ... 7. «o. .1.\. .5 . k .. m . a «w . .- 2 2 m3 a. .3 ....m m P “u mm Vaum .uuh ”a .ru be A ..m no .‘ «LC .4” «M Tu m. J. ”J a . Q. ‘0 I ‘. o . » ...“ a: .9 a . ~ . mm m "V c o 0 3 a? 0'. 3 o h :0. “0.941 ~3 q o '1 .~ lou uthzrn Eco ’°* to 0 Too nasa .I ‘GO 3 7 7 231756 123®557 l -0?! 1 Speaker Number Eco . “ ‘ “d “19“" -0 0.. ‘.o ‘70. U.‘O\'.-’- to . \i.- .2 . \ HT 0.....- d ....q-_. ...-i... C: o - WCJ .. Qua—.o . g . “KI-ad a .-.? olfl‘.’u a- ‘01-. ..--o ......o—.. ' 5 ‘ - "I u I". .Q ‘ v I ....'..\'...- a- t' b 7 3 O Era-u o COCO. “'3‘ -~-u--, {“22 n-‘.‘». --.-.‘ i. Q- ' 0.1:...: 1-. Jd-X“ -‘I‘ .... :zaCunu,; .I-‘vf‘u ::‘.' .‘ . ..--I‘V- \ovv- "CO-fi... .— - I e “......0 V q C -... nixegi VCUSL 5hEIC. COKJVLAL ': ll n- ~90 ' - .ae e.-ec:3 c: Decreased - a ‘5 -. ‘o - :‘.‘..“.~.‘.'1.:..s.1‘. on 3;.324.3.‘.‘ 4‘18 3? Aural an; 71:33; ’“a‘ a - a speecu staples Q? ‘q “f"_ 0:Ff\_\f ..fife-Q le.‘ ~-.‘l‘..~“.. 2.‘.--0J Q . acage W . -‘¢- 1 -oo eo‘. - 10:: breathy 30c far back 200 denzsal Too slow K“ \n Us U‘ U\ 0 \ O“ 0‘ O\ O\ O\ 0\ ON 0'\ 0\ s “ \‘ ~¢ -q -o -q \3 \3 ice Too harsn far forward dghvso;-Jv: .. .- "V"; b ----b O':-_ ' . . "rm—‘1 1... o dfiih ‘ .-a‘no . , C ‘I uo—GU-o 1" O.- O . 1 . - .03 or.“ q. C a" ' ‘ 00”..-. . - ‘ :00 low q '~ .- -co -re .hy fee far back ‘5 -cc qen‘sal f! H f-O .-0 i—J h) N U [J U \M m0\ Ox U~ U\ (L \‘ \) ." loud high harsh .Q far . Ave-.9 cud V- I3- . C v- u ”1‘0: “4%. Fez--..? 7..“:VQ ..- -¢.vo-- ... \O H O C J :0 .H- ' 532:: . . creased :requency In '. - . .Q - :xnivlitn on 393- 3: eieazl.zca.;cn . . Y c. ‘ .0. , O p‘ :3 Aural an. lleuae eaaziazceon c- a - a Q speeca aazpees .. .Q q-.. -90-, ‘1. a,“ .O_‘ ‘.-.-.-.-, JC-ng -L..;...“.‘.. 3.‘-I.DJ don‘t-o 9123:5133 aca-e '3'- - ~-- -1 .u. 1 /‘\ ' ' -- <-| . . . bUbJuLJJ ‘30 30.. - 2‘ 3 -4 3 o / .gg eon: ~7fi~2 roaJeg “mu-r“ '-: rooVO‘fl-I\voo .— .... agno-q - 03?..: --‘C CO. .. I 3;;OLnfle; Q‘o ? .1' aauia fl- Cc." ‘o-‘ 9 4 u I “--‘n‘ .Xo‘i’ ‘ - -..“ é a 2.1“:. 0- b...- -..-.0! VAlIiEZ :QQ-J‘I U ? vcyiz Anfzc. COKJCLARZ ” OVERALL Consents Jinansisn for -\‘ 0\ low roe nigh m U| O\ \‘ breath? Tco harsh r! F‘ r4 n: _ (E) too far back far forwa:u Ln 0\ cu Tco H N Eco Toc nasal denasal U ( é kn (h \) [0 Too slow Too 0\ "\) \l U\ Pa::e:nei 0\ ‘1 Too aueh ‘3 Southern 0\ 0\ K) Over precise r‘ 04 r4 04 k0 +4 0\ \) Excellent 0... e? O Y .A .“ “Ubdoo ...-:3 P'finc $-J0. P.ICIJ.\".‘Z£ L. chop-...- . ‘- ~o . 0...- b .4..- «...—.-- b ‘ ~-1d\&-— ‘J c \ n U. .0 I". 0 0 rt r4 r: N N (9 u (9 \n UI ~ 0\ h \‘ ’\‘ H O 0 :J‘ I m :3‘ Kn C\ ‘0 5) U Kn (h N) ‘ 333a; r-OJ mo :50 H I) (‘2 ’ n . I .oo ‘lou l 2 O 1 .20 .32: O I.- a I o I: a h -- ua:“' l -< 3 >L 2 - 7 :1.tern:- ‘ --~ 3 .' ‘ e: n q . -31 11..-: 1 - x3 , h 3 v , .30 :42h .‘ "' vi ’4‘ A 5' ‘ ; fl - .‘ .2323.5 1 - J { ~> 3 ~ - :c'.u3:n 1 7 n ' 4 7 ~ ;-e;r2" l - 3 h 3 (-) - eve: 3:2:Lsa §PI .. _ Q .‘ : I . o _: R q. 1‘ - N... .33230. ‘ - 3 - 2 '4 ' édhoe--:nc n a" dice. V. .9-.. \ i- - - d.---- S .3 c . '7~_0‘-'\.\09 ..----o.‘ 1:;1 92 -..-Q3- '- 5 RATE“ 7' 9n- - s b-¢o—o-U-v-I Sseecn “fl 3"; ‘----..---qu. . n L -‘o A g. Aurel 5y 7 -. , a a n... e .. n.- c 1. be a. . .1 C . . i h m . .... t e h u e n n u .e r s 3 «u c r r a. m. o 1. x a u a . u a“ a. l l 7. h“ h I. . a; e 2 v... 7. .. ..w .... .w o C O O C O t O . a A. W O C 3 O a O C W I «Me n... T m. a... m; a. P a... 3 v. . 7 o/ n, 7 7 7 hi hi 7 7 7 .o .c ,o ,0 ,c ,3 ,0 ,0 ,0 ,0 Ce .qa .‘3 .s. ..4 .‘4 ps1 .5; .s1 .~a 3 2 1.. a 12 3656 7 ’4‘ v’ 1 a. A J ‘5’.“ .--.- H co harsh a Toe ch :ucn Sou: 7 7 low 3 10:! ‘ ..‘i‘l; 00 0C lenzsal . _\ .-'-v- u o- .3} 3 30:) ’fl 5 5 \/ \ s‘ IV . e C a ; t _ ® 3 3 3 G 3 3 3 3 .m. 3 A3 ..a \9... 2 2 2 2 2 2 2 Z 2 ..m. .9}. Z 1. l l l l 1. l l 1. "u. .s o. g m a“ w Pu .. o ".0 ... ... m... a a. ...u. n .3 . N . u . . . m o u“ L fl. 1 ... ”u .m e e w ... l .0 a. S S 7 .... e n. c v“ ..a 3 k t m“ c . O a. a. a. .u .. nu r“ nu ». Au .9 a“ .e n. 3 o c Ga. ave o e a a .; r P a. a; a. as «.3. abau n. .a n. 7. .a nu .9 .. Ta 0 O ...u av I f . . ... n . . ... ... ... 2. w... 3 J ... . . . .. . , 3 . x» _ t v ... q. ...... .... I .. a. . ... t u u "a .a .. u .. L" p ... ... . L ..u .fi 3 ..." ... .4 a. r“ swan _«.e .e a. .e n; ”a * m .. .w L n. . L P 3 . . ... a; ... :u .u .u m. ...“ "U C .-. u“ ..u 2. ... ... .m .3 .a C .c C J . a ... P P C i .3. 3 1. a V 7 C 0 ., .‘o' -- -.. ..‘gu . q¢ ‘ . i- . L...£-I.~QQ na- qu..-- ......:¢.- 03¢-.- = L - ...‘1-d '2'.- ‘n.. v\- \Doo. ‘ . k ‘ --'.o‘.-- n"° .‘Q", -u- -...ccc cf D-c:czsec Frecuency :gniviich on Sgsexzr Zian=Lchachu 3y Aurel .3: 713231 :xuziaaczon of Speech 51-9135 o . --Oorct-’ ‘Q. .15 con -.- u—q IC“ --..;u---. 3a-I“4 czecc Dicensicn Scale 0 at- .o g . .0 ‘ ‘ I | 73~2 --JOQ FEM LXIIC{ CEPH§EIL.3-AL :.JKJLc .15- a .. - :Q, V.a..I- ' VC' «’5 3.4213. n n .0 ‘0‘ K boy: 6L“ .1.J.\ 2 0"EPmZ... Ccanencs for IQ ..a to U U U \u 0\ \\ r3 0 0 '3‘ M 3 Too low 1 @® U‘ I O \ \D 0d O O {I ... 0. 13' Too breachy IO U\ 0 \ \I I8 0 O D N 0 i I so far 1 ck Too 1 denzsal [0 U (‘7 (FT) mmmmmm 0‘ kn (h fl 0 O O E In p .4 Zoo slow 0\ \) x) *4 0 O "a W M ct at J Pa..e::e- O\ U\ ‘3 @~ \ @mvv Too :uch Southern to p I» fl “9 '1 31 H H H H H H N N N :6 N N u: x» \» ( Us \l x) x: ’ 3 c. '1 '0 '3 l. fl .. m (b O\ (a 0 (I I 0“ 00 :1 t I Unaccepc. peel-(er number E§ J‘ :L:en:-an Sczle . Q0-.. . .- . - 2 ‘ ..Uc-..:...-a -cc act. 1 2 ,' , 0 7 tea laud. 0"«5': - o ‘0. “ts. . q P..Uo;.\-:;;- 1-10.. “ 0\ J O O ‘1' '8 09 $3 ( I :7 q H H H h) h) \J U \11 U\ 0\ \‘ ‘J 03 ‘ 0 O '0' P. M 7 (aé®®® Us Ox ‘4 3 D O ' 5 u 1 O I, .0 E w p-‘rJ u. 0 '2': O ..a C») U U (h K, O 0 O B La 3) .4 o -"‘\ ' p q 0 -cu lox l 2. 2 .4 2 O 7 .co fa-: ‘~-" '0 , “h " ‘ fl '. '- - u ¢-é\.[ l - 1' J b 3 ‘ I {13363313- \~’ 0 4 I £)o 113:1: l 3 f3‘ 4 J 3 7 2c ‘13? l ‘ I p . -133323 1 3 3 : vf 2 3 . :cu.?3:: - f n -lu;re- 1 2 3 ’=“‘J 3 :’ CVe: crazize v no - ‘ 1 a.”\ : fl .. a Q - c“.c:3,.. - - 3 ~g , a . -“ca---n. She 35522:: c: Decreased Frequency 3 adviizh on Speaker Zienclficezzcn 3y Aurel an: Visual quanczicn c: Speech Szzcles 7C:.\L 2:13.337- ‘-" 7""7'“. :2???" ”3.9.- oo.‘l.--oo doc-co- O ‘0-.. g l . ‘ : b‘dbdno ‘d (u rfi~2 -040. 0-;\9‘ . Q? ~-- :“L-‘C‘Q-UD- C::’:":' ‘_.‘ f In. 0.... 0-. -fivod -..Q.‘oo . 3;;UQ-flQOV‘ a Orlcuflzo- - . b.oo-co-.-o VAiIiII .IC.-‘I.£L 5....‘1‘30 COKJcfiAfiZ " Y" ‘7. 1&nb-l- Cczzencs for 0\ fl L'cc loud r J U 0 In 0 HO cl .4 I .0 +0 fcc high G M U I \n \n U‘ 0‘ ‘0 O\ \‘ 1 Eco breecuy l 2 c ch hersu TOO far 1 Too far forward back to p Ut \ V 200 1 Z 3 & denesal coo slow 1 2 @ L; 5 UI O\ \I r O O 3 k I» 9) ..J 0‘ \I Too fast on- ..x 23.. :32. ‘ o «L '1 9‘ ‘3 H N F. U\ 0 \ '0 Too much 0'4 0 O H P d t t ...4 m ...; U {:- kn O\ ‘1 ‘1 Southern 3 04 fl 94 s F. 0 It .‘2 a u 0 h‘ r1 N N (9 u " (5 \n U\ 0\ 0\ O\ \l \l M a 2 0 (D (O H .J r0 a: w I; a 9 ct u ... In a Q ..‘~-..‘. -U.-..‘.‘...‘.; Q-flfl'O- :-- do. -Ut‘.' ' ‘17 “ :..-..C‘C-5ol r. q I w 0 fl '. ‘ I "...—-..--.” I 0 0 0. O “I (I 'J h) I J 0 O ‘4 Q ,a (\J \J U U (“x ’ I U U\ ti 0 O 3' I :3 (1 G) 0\ \‘ F1 0 O E m 5 I J 0 (3 u' '1 m 3.) (I I: 4 *4 (\J r- KI ab 0 Q :I 0 0 ' 5 V I“ H I.) I.) U U ‘r (‘3 k \n U m 0‘ *I N) r) H O O O O a: " .. t: p .c-J lo. 0 '1 S I J. L . o w 0 ': v z: [.1 ...l 9 I“ I“ la [u U U 9 ‘o U U; U 0 \1 a) 'U NI Id 0 (r 0 (f (U "a 'I (h " l'o III (I O u b .-l ..- u «I .I (I :4 t4 In I) \J D \u (A ‘I DJ 0 O U (C t .J .0 '3 I a .c U) APPENDIX B FUNDAMENTAL FREQUENCY OF EACH SUBJECT 95 APPENDIX B FUNDAMENTAL FREQUENCY OF EACH SUBJECT Unknown speech recordings (one repetition of monosyllables and one repetition of the sentence "Vision my boy sharing hotdogs") were analyzed on a sound spectrograph (Model 700, voice Identification Inc.) to determine the average usual speaking fundamental of each subject during the first call. The analysis utilized narrow-band (45 Hz) speech spectrograms, produced with high-shaping frequency emphasis, within a linear scale of 150 Hz to 4000 Hz. The speech spectrograms were tar- geted and appropriately labeled. The speech spectrograms were then measured at the subjectively determined middle of the steady state por- tions of vowels, applying a procedure as described by Hall (1976). Approximately 10 samples were collected from each subject's speech sample. These 10 sample values were summed together and divided by the number of samples to obtain the average usual speaking fundamental. The results of that analysis determined which subjects met the t 8.52 variation of .the average usual fundamental frequency stipulation. The fundamental frequencies for each subject accepted in the experiment were plotted for both the first and second recording sessions. 96 TABLE 20 LIST OF EAQi SUBJECT'S FUNDAMENTAL FREQUENCY Fundamental Frequency (Hz) Subject First Call Second Call JA 112 105 or 103 113 MW 99 100 GM 105 110 BF 108 101 f: as 116 122 < CJ 101 96 z .m 112 116 PS 114 111 PG 116 113 us 115 119 RS 116 117 CF 195 191 cc 190 192 w 188 186 on 192 193 n: CJ 189 176 j .11. 206 194 z DM 192 184 : m 212 209 ms 209 211 so 210 211 .m 208 206 MG 212 208 97 FIGURE 18 DISTRIBUTION OF SUBJECTS' FUNDAMENTAL FREQUENCY Male Subjects Frequencyfl-h) 90 |00 “0 I20 I30 I40 ‘ I ‘ l ‘ I‘ i ‘ l ‘ F ‘1 I ‘ C C Q F2 62 62 A2 A2 82 c3 0 O O. O O O C e e e I I I I I I III.- II . W I I II I. I II II I O 0 O O O O O 0 O 0 I. e e F3 F3 53 63 A3 1 1- - 4 - -L e - - l - - 4. I70 l80 ISO 200 ZIO 220 Frequencyutz) Female Subjects e . First telephone call recording of one subject. I . Second telephone call recording of one subject. Average age was 28 years for male or female subjects. APPENDIX C LIST OF PHONEMES 98 LIST OF PHONEI'IES Vowel Nuclei Consonant Nuclei Sentences * Sentences “mm" i 2 3 4 2 3 4 i e e e e I ee ee ee e e e B e e 8 e e e u _ e e a. e e 0 e e O. A e e e e e 0 ll 0 e ee 0 O o e e 0‘ e 0 ee 0 SI 0 e e e e 81 ee ee e ee ee OI e e e an e e 3 e e e e e e * The sentences are: l dauli eIt kleem thud? for InntS 2 ‘A deI tum Halt I: veri borIg 3 all spik vale ju let? mendeI nut 4 V1380 maI boI Ssrlg til-dogs APPENDIX D RESPONSE OF TELEPHONE 99 APPENDIX D RESPONSE OF TELEPHONE A calibration session was conducted to establish the characteris- tic frequency response variations within one telephonic environment. The response information would provide a general conception of the amount of variation we may expect with the telephonic environment used in this experiment. The specification of the elements which may be expected to change during the course of the experiment was necessary to allow pro- cedures to be initiated to control for these variations. The following questions were asked: 1. What is the characteristic frequency response of the experimental telephonic transmission system when ... a) different microphones are used? b) different headphones are used (speaker element in the telephone)? c) different telephone lines are used? d) different positions within the test box (microphone position in telephone)? The method used was to plot frequency vs. amplitude where ampli- tude is constant (95 db SPL) and the frequency is swept from 20 to 20000 Hz. The signal source, analyzer and other instruments are listed in the following pages, along with an equipment connection diagram and the resultant graphs . The setup and analysis was performed on January 9, 1978 in the Speech and Hearing Clinic at 9:00 p.m. to 12:00 a.m.. Two telephones were placed at opposite sides of the test room (about 20') and their respective elements disconnected and removed to avoid any acoustic 100 interference with the elements being tested. Different calls were made to compare variability within the transmission system. Pre-calibration of the measuring feedback microphone was performed by the following process: 1. The 4230 compressor microphone was connected to the 2607 Measuring Amplifier. 2. A General Radio frequency calibrator type 1562-A.was fitted to the 4230 microphone. 3. A 1 K32 tone at 94.6 dB is generated at the microphone and the 2607 Measuring Amplifier input sensitivity is adjusted equal to the nominal output of the calibrator. 4. The same calibrator is connected to the 4144 Mic which is connected to the Audio Frequency Spectrometer (2112). Again, the same signal is applied and the Audio Frequency Spectrometer is adjusted. A total of 18 calls was made, producing 18 different frequency response curves. The observations made of these 18 different curves consisted of the following: Comparison Observations Different telephone Virtually all the curves are identical, lines showing very few differences in the transmission of the signal. Different telephone The curves show very minor and insigni- receiver elements ficant differences due to changes of the telephone receiver elements of this type. The only consistent difference was a slight 2 dB increase in the dip between 1 K32 and 1.5 KHz. Different telephone The curves appear to be the same for microphones telephone microphones of this type. There were differences in frequency response irregularities below 1 KHz. Two of the Mic.'s produced some peaking in the area of 500 Hz whereas the other mic. produced a smoother curve below 1 K32. Also, the three mic's appear to produce somewhat different patterns of peaking between 1.5 KHz and 2.5 KHz. Different positionings Variations in the curves appear to be minor. of the headset in test The microphone was not changed from event chamber to event, but the telephone was successively removed and repositioned in the test cham- ber. Whatever positional effects occurred were probably limited to frequencies above 4 KHz. 101 Response curves generally depicting the usual response of the variable tested are presented for each condition. The following conclu- sions were made: The results of the calibration supported the notion that the telephone response is very consistent for these conditions; microphone, receiver earphone, telephone lines and positioning in test chamber. The largest variation was observed in the positioning in the test chamber. 102 23.32.... 3 02.2.3». 3:252". 3.62.“. 2 23:02:30 EOE-:39” .0 52020 235.com .o—oEozooem 3:252“. so: / .3. 1.2a. 5:26... a. 952m V y / e e 0 O E . . O O O O o 5333. .964 2.395 333:3 £851.26 E 3.59:4 E 953322 3m .3» ‘2. ‘IO 20 103 Different telephone lines (Record 0 Li) / ___. l w. W W. 1...... m» 1...... ... mu... 4.1%,... (Record '11) .. w l ‘ A i\ —J W / i. ll! loosens in, mono HZ 104 859 Different telephone microphones @eurd’ H) A / '“ j‘v \/ Um no so too one an loco sane sane usual-Ix Q d8 50 Different positioning in test chamber ' (““4”A l'f . ,0 M AD 50 inc 330 So lam anon 5m loans H2. 105 A B B A Record Number Transmitting Receiver TelephOna Telephone Telephone 71'"- Telewhoue Ear- P] ace mutt in ”umber Phone Nwfiber TeSt Box 1. II 121963 ILB. U3 j-333ITHJ CK 121748 2. n n 3)? .90 u 3. n u 3’)“ ‘1 in 'Le ' ’ 3379; n :1. u it 337:2 n K II ‘I ‘ f l‘ N O J-ll 7 . . , 13, 21 " ‘. " ” 3-J333791 " 9. vr 7' ;-3;337J2 0! 1C. n u 9-31/3‘733 n 11”. u u 1-35337 . II 12. " ”.8. U3 33730 " 123968 13. " " “4533789 " 1h. 11 12274 " 33780 £3 15. TI 121676 " 33780 - m 1 1..- -— ' '_-_:I.__~. 106 a. Reference for the output signal was 3F 69 JPL at 1 K53 on the graphic level recorder, which corresponded to 95 dB SPL output from the speaker in the test box, for the same frequency. b. The same telephonic exchange beginning with the number 9 denotes the use of the outside Bell Telephone system. The five digit numbers (99999) denotes the use of the on- Campus Centrex switching system, located on the campus of Michigan State University. Components - (Connections and Controls) Audio Frequency Spectrometer - U«% K #2112 r Automatic Switch OFF Amplifier Input Input from the Pioneer Mixer Output Condenser MIC . Input Potentiometer O Veter Switch . Fast 3M3 } Recorder Output to the Graphic Level Recorder #2305 3 Range Wultiplier 0 dB x 1 § Function Selector 2 - hO KHz L. Weighting Network ‘ Linear Pioneer' Mixing Amplifier - #HA-62 Frequency Response 20 - 15 KHz (+O,-3 dH) Input Sensitivity 0.25 mV Tnput Imped ance 11.7 K52 Rated Output Level 330 mV Signal/Boise -52 d8 (IHF Shorted, A Network) WTC input channel #1, Induction coil 'pick-up' Gain 8 of possible 10 Master Gain Between 7 & 8 of possible 10 Dictaphone Coil - #377232 0 . fi Induchon type donut 'ple-up' goon. Coil placed over receiver telephone earphone Telephones & Vicrophone Western Electric #330 CT (receiver 5) pg} 217hh manufactured 1 - 75' #121963 “KS. I‘73 'I.E. U3 Western Electric “ 00 C-0 (tra :mitter \) MTC 4 TI 121963 "anuf‘actured 1 - 69' # TI 122711 ' # TI 121375) Anechoic Test Box - B e K fh212 107 Graphic Level Recorder - H a K $2305 Potentiometer Range dB Rectifier iesponse Lower Limiting Frequency Writing Speed Input Logarithmic Potentiometer Input Potentiometer Input Attenuator Paper Style Drive Speed Paper Speed 50 1’13 20 Hz 125 nm/sec Audio Frequency Spectrometer SO dR, 10 mV - 3.16 V, ZR COOS Variable Variable #QP llzh 36 3 (small numbers) or 30 (large numbers) Drive mechanism connected to lower right side mechanical access port, with button pulled out (speed control). Other and connected to the Sine - Random at the left side port. Sine Random Generator - B & K #102h Compressor Speed Bandwidth Selector Meter Switch - Power Frequency Random/Sine hatching Impedance 2H3 Matching Transformer Compressor Output Level Compressor Input Measuring Amplifier - B h K Gain Control Input Input Attenuator Output Section Attenuator Meter Function Averaging Time Output Coupling Bandpass .Eeter I”anal Transducers - 7 ‘ K hbdel Fhth l" #hhlh l" 100 Sine 0.3 sec Compressor Input, 12 V Full Scale Ref. '3 D? . . 6 Ohm 3 Ohm - connected to Anechoic Test Box Between 5 and 6 Connected to output of 2607 Weasuring Amplifier #2007 ,0 o- ' calicrxted Fre-amplifier from the feed-back VIC B & K fhlhfi within the test box. 0.3 V x 0.3 Lin-ear RJ'IS Fast A060 2205 LIZ to 9.20,: Wis Read in sound level linear, SA 0056 A? l-J_.--—A ‘, APPENDIX E EXAMINER' 8 AUDIOGRAMS 108 APPENDIX E EXAMINER'S AUDIOGRAMS To obtain information on whether there could exist interactions between the examiner's hearing and the frequency bandwidths utilized in the aural tests of this experiment, audiograms of three (A, B, D) out of four examiners were obtained. Examiner C had previously reported good hearing, but this was not possible to confirm through audiological evaluation since she left East Lansing before the audiograms were obtained. Audiograms of the examiners A, B and D suggested that no interaction between their hearing frequency response and the frequency bandwidths utilized in the tests of voice identification could have existed since their hearing was within normal limits through 4 KHz. 109 moms»: sun UNIVERSITY ' Speech and Hearing Clinic PURE-TONE SCREENING AND THRESHOLD AUDIOGRAM Nu. fixammgg A A go Examiner Audioaatar '. l 0 Date . Purpose of test Complaint (if any) hammer (Ht) 125 250‘ son 1000 2000 woo soon .19 -10 G g; 0 0 . 10 10 H 2 20 20 g so so s 40 ‘° 3 .. 5., so 3 _§ 60 60 g .70 7o 80 .0 90 90 mo .100 Code to threshold measurements: Bar COLOR AC (earphone) BC (bone vibrator) Oright rod 0 (I left blue X ':> Test-retest reliability oLchQ_ Audiologicsl impression u 110 MICHIGAN STATE UNIVERSITY Speech and Hearing Clinic PURE-TONE SCREENING AND THRESHOLD AUDIOGRAM Ida-amt” B J Ago Examiner O Audioaoter Mnlco Date :Zig Purpose of test Complaint (if any) .19 -10 g) o O H . 10 10 H 2 an 20 a 3° so “:7 ‘° ‘° ... SO 50 :3 60 00 3 ,0 so IO 90 90 100 .100 Code to threshold measurements: Bar COLOR AC (earphone) BC (bone vibrator) right rod 0 (I left blue X '> Test-retest reliability—M Audiological impression 111 MICHIGAN STATE UNIVERSITY Speech and Hearing Clinic PURE-TONE SCREENING AND THRESHOLD AUDIOGRAH Nam f L x &1m M 2 E ‘2 . Age Examiner . . Audiometsr Bfltgmg‘afi | Date 5‘ if Purpose of test Complaint (if any) from-Ber (Hz) 125 250 500 1000 2000 4000 8000 ~10 -10 a: § 0 O . lO 10 H 3 20 :o o: 3° so g 40 40 3 60 GO ‘3 g .70 10 80 IO 90 90 100 100 Code to threshold measurements: Ear COLOR AC (earphone) BC (bone vibrator) right red O (I left blue I '> Test-retest reliabilityJ'n-gL Audiological impression ‘1 APPENDIX F INSTRUC’IIONS 112 APPENDIX F INSTRUCTIONS You, the examiner, will look at speech spectrograms and listen to tapes containing speech samples recorded from people in the Lansing community. There are 192 discrimination type trials of speaker identi- fication (where one Unknown speaker is compared to one Known speaker). The speech samples in each trial consisted of the following sentences: "Julie ate clam.chowder for lunch" - (sentence #1) "The daytime flight is very boring" - (sentence #2) "I'll speak with you later Mbnday night" - (sentence #3) "Vision my boy sharing hotdogs" - (sentence #4) In each trial, the speech samples of the Known speaker contain two different repetitions of the sample sentences. Also, the Unknown speaker contains only one repetition of the sample sentences. Your task is to determine in each trial whether the identity of the Unknown speaker is that of the Known speaker, based on observing and listening to speech samples of both speakers. You will choose one of five alterna- tive decisions for each trial. These alternatives are the following: 1. Positive Identification 2. Probable Identification 3. No Decision One way or the Other 4. Probable Elimination 5. Positive Elimination 113 Each trial must have a decision. Each decision will be marked on scoring sheets (provided with the tape and spectrogram materials) in one of five spaces which coincide with the numbers beside the previous listing of alternatives. Labeling of Spectrograms The Known samples are labeled with the word 'KNOWN,‘ stamped in ink on the front upper right corner of the spectrogram. The Unknown sam- ples are labeled with the word 'UNKNOWN,‘ stamped in ink on the front upper right corner of the spectrogram. All spectrograms are transcribed and targeted on the lower front margin in pencil employing the Interna- tional Phonetic Alphabet. The trial in which either the Known or Unknown speech samples occur is printed in colored ink on the front left border. Trials number 1 to 48 are marked in orange ink; trials 49 to 96 are marked in black ink; trials 97 to 144 are marked in blue ink, and trials 145 to 192 are marked in green ink. These numbers are also printed on labels affixed to the plastic envelope containing the speech samples (both Known and Unknown). These labels are color coded in the same colors previously stated, with the exception of the color black (trials 49 to 96) which has a white label. These labels will aid in locating the spec- trographic materials and replacing these materials in their respective envelopes. Spectrograms from the Known group are compared with spectro- grams from the unknown group, both of which have corresponding trial numbers. An example is to locate the materials for trial number 1 by selecting from the group of unknowns the label with the number '1' and simultaneously selecting the same number '1' from the group of Knowns. 114 Labeling of Tapes Each 7" reel tape is labeled on the reel and on the storage con- tainer with the trials which appear in that tape. The recording of speech samples follows the following format: 1. 2. 10. The trial number is announced verbally by the words "Trial number ..." (and the actual number). The end of the announcement begins a three-second pause before the samples begin. The first sentence, spoken by the Known speaker, is heard first. This sentence is one of two repetitions spoken by the Known. This is followed by a 0.5 second pause of silence. The first sentence, spoken by the unknown speaker, is presented. Only one repetition is presented. This is followed by a 0.5 second pause of silence. The first sentence, spoken by the Known speaker, is presented. This is the second of two repetitions spoken by the Known speaker. This is followed by a 1.0 second pause of silence. The second sentence is presented, following the previous cycle of Known - unknown - Known with the same intervals of silence between each. This pattern is continued until the third and fourth sentences are completed. The end of a trial is signaled by a three-second gap of silence. There are five seven-inch reels of tape, 1/2 hour duration each. You may play a trial segment as many times as you like. Each trial must be examined one at a time. No prior trials are to be listened to after a decision is made. Trials are not to be skipped and returned to at a later time. Time Limits Each trial is completed before proceeding to the next. There are no limits to the amount of time to complete one trial. 115 Environment You may work under those conditions which are most comfortable. But be sure to have adequate light and sufficient space to observe. Take breaks whenever fatigue arises. Tape Recorder Settingg_ Equalize for 7.5 ips and adjust the volume for comfortable levels. APPENDIX G SCORE SHEETS .2. .v. ... .fl. .ufig nflud .(. .—. ...... “H. a A ppendlx G .N... Kauai: ... 7e aawasuwnsova. 13.3.1.3..3 1d noncnorauou... .41.. inast s. ...: :.u‘aswaasl: 9.2.51.2..3TT In: non—snoweaala. 3.. . has... :2 ... TS Susannaaowa. ...o.nvo;.av ... 1. neweuvnehere. Quasar. an. nasal-1. anwnunwafinnin. Go. Ignaz: .. 317.8 .~u~x~uxa~u\.. ...vaa.~.n:_a:is :...n_u.u—e.. :31... say—...: 3 2:1... .nevoaoe..nowa. q _ : D E 35.09 ><¢ J .0. plans: w .. aw..\aua.. .N. \4Nu. «Ne... ..N.... 3:... .> .253, CC... .>..>n>w>4>ea.;~»..>s>. .x tuxe..xw« x.xuxvxexvxaxrv~xnr :1 ,4. .34 5.32.». 25...... aluginkés s .2: .... « 2n>a.ze>n. >. 24>!)4>uz..>+>fl>v). :..")...zas... 3.22.3.5“.......Dizaas... ...»..p: :..—34.3.5222»: . 13.13:. 01.23.... 31. .no... are... 4.: .....y...~.e_. ..u....¢o.; .8..: at. : 4:..:. .3 33.3.3.3... 7.34::.3~.....3...~.5..3. ... .95.... ans... .as.. an; at I:- u‘ SUCH 3 Us ... 84' ...-on is 'h i 8" .8.‘ a-li ... ! fig 3 '1. ‘6) .2 He‘s n1:- azu a.» "by .‘n :..-u 41..- .5. «5. 5. a3 .4. .60 4. 0‘0 .1: n3 .3“ ‘1‘ :1» a): .... 2.6 .1. n'u a... «in .c. a n n ‘5‘ 2...“ 95¢ n5..— ~Ja ~40 nude a‘u nzu 030 2!: «33 n2. .2. ...; 4,.» .. . ale "1. «In a a n. a :3. as. 73. ~50 .4. ~10 .... «do . J. .13 . 3 :..... ... _ ... .. ~13 ....o .. . 7L. u n w u . 3 -Se 7:. a5. .1. 7.: nu: ale ...: as; ".... ~15 .v . $3 ...... .2. :.....IL .. . .1. n A! a .q. .5 e... ...... .5. ... an: ... . an: .1.“ a .3 :..—u .y..§u .24 n1: ... . ..‘u .3 e «In ..2. .5u 2 . an ... . o.¢ 7 . Tau 2.... an . 4. Sn "9 .9“; .7. «a... I.“ la a- a «‘0 n n n n ~5. n5. «.... gu ... Tow C... n.u as... S. ...: ~39 ...... Qty «2. «2.. .. . 7v. .1. a. u w n u u .5. at; ...: 9.5 .... at. .4. n‘u .3. .13 .3. 244 .7. a): .13 n3 .1. ”I. .1. ~15 9 "an .6. neon n3. «Q63 an... nun: .4- =4. “2‘ a...“ .5. and. 3...: «As. as; age. «.... :4. p3. n~éc «It. as. «4— .....a .4.- ng .5. ale .3. R3 a3. 02. .33 «all «cl. a u n’u «5. «Cu .4. 2.: «3.. as 53.4 awn «We a: :3. ~43 .4. as «.0.: a): :... «sun as . as. ... . as: ale. ans-u .1 g 92.- .. ~14 as 9 M 45¢ 33 a... .4. n1; .. 3a . .. HWa .v“. .5. ..r... He.‘ see . n1: .1 C...- «a «I. a! . m .9. go Tn... add _I. n13 .3. n2» :.. «we a. _ a at: as: n‘. a... 3.. 3.. halo 2‘. «an. a}. a 9 HS» aloe As... used n.3g n4u «1: as.” RSI «3. «13¢ ‘24 , 7.. 73.. an. n.)- .2. note :..-u a... .14- ~¢.. fl‘u «Sn «.0 4 .5so field 3.5 7s. ....5 «Zn ~.~ a «3. .3.— «iv. ..34 Qaed .15. ...-u .1.— a1: ash n‘u n3 .4. n1.» «3 new «5.- n‘u «3. «13 «I. H5. ‘Jo n1: «2. rel. as. a}; .3. nfi‘a 2.... «e. a]. .. _ . 04.3533 .933 68¢ 1...; 3.352 .33» -252 52339. aura. Susie «as: >30 ..ozaa an: ..o. .5- ..e. no. .3. «13 .2. a): ..-. «In .. ... an... «I: .4. 2e- :3. H15 :3 «a .1. 6'.- » m «.5. n5u «on. n‘- ‘3. at: 2;. 3a .9. . ..lu .. m .5. . at: ql. al- ..5. n3. ..l. nqeu .. . a... m. m e! :... n5» . ... «no. ...: n3. _ 3 424 4 .. o u w cl 2... a a. 1 . n‘u :... 1.: n1; n}; a. . Helm m. n ...J as. .... Tan 1:. '3» .2. in .\ a Halo m a :..: :35 .... wen u). .1: 2... 21: a.-. noun u n . .3- .‘u «3‘ 333‘ as. ...... at. .4. n3: n3. sale ~2- niu at. ... ... a! n5.— n4. «4. n3 2:: Q3 .73 noun a?— u n as e- 35- ..5. a‘. ad.- nzu .1: ~93 as: 8“ «la n a el cl «‘3 :3. Neon use; n1: .1..— nseu had. «In. a... 7 1 a el en fit: n‘é .... :3 Q15 2.: 62. ~2— nle . . a . ‘ 1. 2 1| el 059 «:3 n3. .3. :... «3. ~23 no; a,“ a? a I. .. ...... 1| el «:5 .5. «In a}. at: 2.3 3n n): his a. a m .. iv .5. 7.3 a... an «3. n1: :1. 5" .10 3 es el 2 el as 3!; an 23 :..: filed 23 a.» 01: 2,3 noun ...3 file n3 «la n. .u n5..- ~40 «1: as“ «'8 an 0‘0 «J- not.- no... gala ‘5‘ 33. a1)“ 9.1.— n-. 4 nPJu ad. :3— as: ash a... :1: an.. Aal- n5- anew :Ju a): n1u n5. ale .3. .2. a... n...- as.- Give :4» a... «1.3 n...— .1. :..- «In ...... n2“ 3.. . was. Fannie-.52. :5. ~Su .... «new 13— Halve as c Ag...- wne Hid W . 6 el el ....u ~50 a... a3. :...— ~30 1... a... a... ~11u 9 7 5 a Cl el ...: as. a... n40 :5. n2: . .. .Zu .... 7.3 a s a 1| 1| .c.— ~20 .1. w‘. .1... n3.- :.— n2¢ .. . ROI. 7 5 a el 1.. . ... n.)— _l. n..— 4... "and. .7: a): .. . 51... a s a «I el .... ..3. p... “IO. _ no: a. .4 can as _ Hal.— 5 3 5 6 e' 0' .... at; .1. ad. .3— an: .s: .2; .. . sin a 2 6 el el 3... as ..3. .44 ...: 2.3 :.. ~13 .1.— .‘u 3 es 5 0 cl 1 a; 3:0» ”Mummies-ton” ‘UOI "1 0-792! .... .... 3. .... .... .n. .3. .2. .... .... «I. .<. ... a... H :..... B :..: 5.9.2.1; .... T... noumnawnaava. .»H. .3321. a. T4... «cacao-caulz. .xv:a~u~ Z I. ...... gulls-«3:1. .Su....n..~...«a. .... In: navalcu..30fla. 3.. Sign: .. . TS sonata-3.3!... 25.3.7.1... 1. avwvgvnquvv: ...-Jamaal” ...nnurl. «nununva:n~«. Svaanunan .. 3......- .«vmaahxnuua. .xv:iu.aL.A:}a i.2.a.i.-.v... .93.. Aouaao 33.15 «00:40.1...9”... a. _ . H r h... E 3 8.2.3 5.- J .3 .22... 117 2.. .. ... .... .3 :1 .... ... .u .3 .u u... x..—N.N\IV., . .a>.... «>v.. .4XI...X~. ... at! S 431; >«>u..) a)...» ..Aq.....n:u.; .Abi. 4.... .. «$1.. «nah . 2......11; 3.53.3.3... :..... at... 2.3.3.53: .22.}...31 .... . .4» ... «2.. I .n... .9... ..4! ...t..& «navqig. .. ... . a... Z 2...: .22.. $461.50.... ......A.... :3 .nwu. ..nc...ac.... .49....3...) 24.....nclz ...u. .p n>x> 4>n> are.» ..>u... w... 3...... :3. ..Khr :..: i1...23..35n...f§n.> "3...... ....) n>v>a>u 2.54) as; .. ......) .Dsra... 2.2.1: 2...... ...—:..»... 2:2»: 0.1.3.1... .am.... in. .. ..mv: . ..H: .31: .31. 3.9.. 2...... :...3133A9-2 so .3 a0»... ...... 4.2.1:... Ce.....a... T3...3:3..31343u..a03.. 2.3.3.4... AZu._421._421J. 7.3.... 3.3.31.2 4:. 2.3:... ..., .1.” :..: :.n ...... .. .....41812fix2“.4xetaxv¢. .«thquqaquq..«uqa«v.. ....I«V.d.I.A-9.4.!.. :......zu..a:u..4:....zv.; O!«.u°i3~¢i..xm.w 5 do. ... ..U.A.I.4«u.aur.aav. a... .nun .30» .au: .43. .. :.vanowaauzausaov... ui..Au~un0n ..«Uvuuuau. avzpagaucraacvanana. I: ““1” 0732' APPENDIX H RAW SCORES 1W ‘ - 4.0 Examiners ABCD Identification 4 Examiners A B C D Bandwidth (Hz - KHz) gflQ' 2.9 239' 2.5 230' :.Q 118 TABLE 2.]. A300 3 Examiners 3 Examiners A B C D i 3 3 2 1 1.11211 '1'112' 111113 22 '1 '0' 0' 111 121.2 1222 111111211 122221211 131111114.- 1.2 i ...-all.- 132 '31 133 Olllll 333 233 332 111111.214! 333233333 232231322 123212 121212 232333 232333 143212 21- '1 3" 1221 3" all 14.. 12 1'2 22 Ala! 32 all! 22 20.- 3222 4.1.2 129.- 333 Examiners‘Responses Within Each Frequency .Bandwidth for Each Talker in Match Trials of Voice J A t Known Talkers CFZ 002 “me 0112 caz JL2 one .1112 use 392 JWZ M32 4 . Probable Elimination 5 8 Positive Elimination GO 8 Trainee 1: Positive identification 2 8 Probable identification 3 9 No-Decision A-B = Professional HS man: 22 Examiners‘ Responses Within Each Frequency Bandwidth for Each Talker in No-Match Trials of Voice identification Known Bandwidth (Hz-KHz) Talkers M 240 - 2.5 240- 39 240-40 Examiners Examiners Examiners Examiners Male ABCQ AEQQ_.A_B_9_D_ JAI GT1 MWI 6M1 BFl 081 CM JWi PM PM WM R81 Egmaie CFZ CCZ Twz CH2 0J2 JLZ DMZ TMZ M82 862 JWZ M62 huupuuupuuuu nuauuuuu3®bu> uuuuubunuuuu unmanaauauuum ouunuumpuumo uuumuuuuhaaéo abhauuabuuuu uuwuuuuuuuu¢o phanupuuuuu» haveuunuhuhb uuuomuuuhhun ouwuubuhuumu buehubauuauu Afiflfibhbubfihh puguuuuupuua bumubuubhbhu oopuuuhuuuuu a&uoauuuu¢uu cououuuobuuh #buuuuuuhhuo buuauu¢oumha uwuuohbmhuua uauauaouhaua uwNauuuuuuub nouuuaubhhuh #wwachuuAbuu A¢uu¢¢uh¢wh& ##uuuuuebhaé ouuouoaupuug u¢uuu¥u¢33ua nougouuuumuu uuuuuuuuuuuu is Positive identification 4. Probable Elimination 2 = Probable identification 5 8 Positive Elimination 3 a No-Decislon A-B~ = Professional C-D = Trainee LIST OF REFERENCES 120 LIST OF REFERENCES Boone, D. R. 1971. The voice and voice Therapy. Prentice-Hall, Englewood Cliffs, N. J. pp. 33-34. Bricker, P. and Pruzansky, S. 1966. "Effects of Stimulus Content and Duration on Talker Identification." Journal of the Acoustical Society of America. val. 40, pp. 1441-1449. Bricker, P. and Pruzansky, S. 1976. "Speaker Recognition." In N. Lass (ed.), Contemporary Issues in Experimental Phonetics. Academic Press, N.Y., pp. 295-326. Broad, W. J. 1979. "Experts Debate Authenticity of 'Shah' Tape." Science. vol. 203, pp. 852-854. Coleman, R. 1973. "Speaker Identification in the Absence of Inter- subject Differences in Glottal Source Characteristics." Journal of the Acoustical Society of America. v01. 53, pp. 1741-1743. Hall, M. 1976. "Spectrographic Analysis of Inter-Speaker and Intra- Speaker Variabilities of Professional Mimicry." Thesis, Michigan State university, pp. 26-27. Hazen, B. 1973. "Effects of Different Phonetic Contexts on Spectro- graphic Speaker Identification." Journal of the Acoustical Society of America. val. 54, pp. 650-660. Kersta, L. G. 1962. "Vbice Print Identification." Nature. Vol. 196, pp. 1253-1257. McGehee, F. 1944. "An Experimental Study of VOice Recognition." Journal of General Psychology: vol. 31, pp. 53-65. Pollack, 1., Pickett, J. and Sumby, W. 1954. "On the Identification of Speakers by voice." Journal of the Acoustical Society of America. Vol. 26, pp. 403-406. Poza, F. 1974. "Voiceprint Identification: Its Forensic Application." In The Proceedings of the 1974 Carnahan Crime Countermeasures Conference. university of Kentucky, Lexington, Kentucky. April 16-19. 121 Smrkovski, L. 1975. "Collaborative Study of Speaker Identification by the voiceprint Method." Journal of the Association of Official Analytical Chemists. Vol. 58, No. 3. Smrkovski, L. 1976. "Study of Speaker Identification by Aural and Visual Examination of Non-Contemporary Speech Samples." Journal of the Association of Official Analytical Chemists. Vol. 59, No. 4. Stevens, K. N., Williams, C. E., Carbonell, T. R. and Woods, B. 1968. "Speaker Authentication and Identification: A Comparison of Spectrographic and Auditory Presentation of Speech Material." Journal of the Acoustical Society of America. vol. 44, pp. 1596-1607. Tosi, 0. 1979. Voice Identification - Theory and Legal Applications. University Park Press, Baltimore, Maryland. pp. 6-7, 76-80. Tosi, 0. and Greenwald, M. 1978. "Vbice Identification by Subjective Methods of Minority Group Voices." Paper presented at the 7th Meeting of the International Association of Vbice Identifica- tion, New Orleans, Louisiana. Tosi, 0., Oyer, 8., Lashbrook, W., Pedrey, C., Nicol, T. and Nash, E. 1972. "Experiment on Vbice Identification." Journal of the Acoustical Society of America. vol. 51, No. 6, June. Young, M. and Campbell, R. 1967. "Effects of Context on Talker Identification." Journal of the Acoustical Society of America. Vol. 42, pp. 1250-1254.