ICHIGAN STATE UNIVERSITY lWlHlll‘MllillT“)‘\|\\‘I|i\\|\|h|\.1\«W 3 1293 00083 71 8 I. C- TL'fuf -_ -‘K*II Michigan Seize Uri-{$12233 This is to certify that the dissertation entitled DETECTING DECEPTION I THE VOICE' AN YSIS OF TH FUNDAMENTAL RF :5ch SYLLABIC URATION AND ITUDE OF THE Fuww VOICE presented by MALCOLM E.‘ HALL- has been accepted towards fulfillment of the requirements for PH.D. «8min EDUCATION ADMINISTRATION Wa— Major professor Date 4’28“% MS U is an Affirmafiw Action/Equal Opportunity Imitation 0-12771 LASLJ RETURNING MATERIALS: Place in book drop to LIBRARIES remove this checkout from _ your record. ~FINES will 7 be charged if book is returned after the date vv-~**" stamped below. h5kxD395 Kkjj‘w ti: O I 0" A 032 “m “in” JufiafiiE ‘r 700? MWN; \ DETECTING DECEPTION IN THE VOICE: AN ANALYSIS OF THE FUNDAMENTAL FREQUENCY, SYLLABIC DURATION AND AMPLITUDE OF THE HUMAN VOICE BY Malcolm Eldon Hall A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Education Administration 1986 {O :3/9 /§/ f MALCOLM ELDON HALL 1986 ABSTRACT DETECTING DECEPTION IN THE VOICE: AN ANALYSIS OF THE FUNDAMENTAL FREQUENCY, SYLLABIC DURATION AND AMPLITUDE OF THE HUMAN VOICE BY Malcolm Eldon Hall The purpose of this study was to determine whether or not changes in the fundamental frequency (voice pitch), amplitude (correlated to intensity measured in dBSPL) and syllabic duration (measured in ms for prolongation of vocal utterances) of the human voice are useful in detecting deception. IAmdio tape recordings of actual polygraph examinations recorded by police polygraph examiners were subjected to analysis with a PM Analyzer, an audio-microprocessor that produces an oscillographic envelope display and fundamental frequency of sound, to obtain measurements of the fundamental frequency, amplitude and syllabic duration of the subjects tested. Digital readings of the above mentioned speech parameters from the audio recordings were statistically analyzed and tested for significance. Malcolm E. Hall The recordings of 47 confession-verified deceptive subjects, 33 verified truthful subjects, 24 nonverified deceptive sub- jects and 21 nonverified truthful subjects were analyzed. The vocal data from the primary relevant and primary control questions from the list of test questions in each of two tests were extracted from the list of test questions. IAccording tOIthe polygraphers control question theorem, when a deceptive person responds to a relevant question, the physiological response is greater than the response to a controllquestion and a truthful persons physiological re- sponse to a control question is greater than the response to a relevant question. Therefore one would expect that the fundamental frequency, amplitude and syllabic duration of a deceptive person, when responding to a relevant question would be greater then when responding to a control question and the same speech parameters of a truthful person, when responding to a control question would be greater then‘when responding to a relevant question. Analysis of the data supported the interaction effect pre- dicted for the fundamental frequency changes of verified subjects for test one (P=0.0008). The predicted interaction effect for the fundamental frequency changes of verified subjects for test two and the combined measurements of tests Malcolm E. Hall Tone and two, did not prove significant. However the change in the fundamental frequency‘was in the predicted direction in both instances. The predicted interaction effect for the fundamental frequency changes of nonverified subjects for test one did not prove significant(P=0.755); however the data supported the predicted interaction of nonverified subjects on test two and.the combined measurements of tests one and two(P=0.0l7 and P=0.015), respectively. Analysis of the data did not support the interaction effect predicted for the amplitude (correlated to intensity measured in dBSPL) and syllabic duration (prolongation of vocal utter- ances, measured in ms) for either verified or nonverified subjects. ACKNOWLEDGMENTS First and foremost I wish to acknowledge with great appreciation Mr. Joseph Kochanski, Director of Forensic Science Programs and Financial Management, at the National Institute of Justice, Washington, D.C., for his attentive assistance, enabling me to acquire a fellowship grant, that without, this study could not have been undertaken. Secondly, it is with sincere heartfelt appreciation that I thank the following Michigan State Police Polygraphers who gave voluntarily an affectual effort to deviate from their normal practice to acquire tape recordings of polygraph sessions; stressing the importance of the study, convincingly to those suspects of crime; enduring the additional paperwork of logging information, acquiring signatures on permission forms and mailing or sometimes personally delivering the recorded data: Charles R. Allen, Jackson Terry L. Anderson, Paw Paw Robin M. Bratton, Flint John G. Hulsing, Grand Rapids Christopher J. Lanfear, Madison Heights Carl G. Lundgren, Wyandotte ii Theodore J. Monfette, Northville John J. Palmatier, Lansing James R. Ward, Jr., Grand Rapids John H. Wojnaroski, III, Pontiac The design of an appropriate program to accomplish my educational goal was directed by Dr. Max Raines. Dr. Raines, Chairman of my committee, through friendship and personal caring provided a guiding light into educational experiences that may have remained darkened. Two committee members who played a critical part in my program were Dr. Norman Bell and Dr. Joseph Levine. Their contributions in statistics and adult continuing education, respectively, aided me with my dissertation and future vocational growth. The fourth member of my committee, Dr. Frank Horvath, from the college of Social Science and School of Criminal Justice, directed my study and provided me with insight into polygraph through his vast experience and knowledge in this field. It is with a close personal friendship, that I express my gratitude to him. My family has always been a never-ending source of love and support. I thank my lovely wife, Gwen and my two sons, .Malcolm (Mackey) and Trevor, whom I am most proud, for being there during the trying times and putting up with my quick tempered rhetoric and overall neglect. iii TABLE OF CONTENTS LIST OF TABLES.00....OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOVii LIST OF FIGURESOOOOOOOOOOOOO0.0.0.0000...OOOOOOOOOOOOOOOXii Chapter I. II. RATIONALE OF THE STUDYOOOOOOOOOOOOOOO0....00.0.000001 Background..........................................l Statement of the Problem............................6 Importance of the Study.............................7 Purpose of the Study................................9 Definition of Terms.................................9 Research Questions.................................l4 Hypotheses to be Tested............................15 REVIEW OF RELATED LITERATUREOOOOIOO0.00.00.00.0000017 Polygraph..........................................l7 Field Procedures in Polygraph Testing..............19 Pre-Test Interview.................................19 Actual Testing.....................................21 Post TestOOOOOOOOOOOOOOOOOOOO0.00.00.00.0000000000021 iv Table of Contents Continued: Acoustics of Speech Production.....................22 PSYChOJ-ogical motionOOOOOOOO0.0.0000...00.0.0.0...24 III. RESEARCH METHODS AND PROCEDURES....................27 Population and sample. C O O O O O O O O O C O O O O O O O O O O O O O O O O O O 27 Instrumentation and Data Collection................29 oueStions for Analyses. O O O O O O O O O O O O O O O O O O O O O O O O O O O O 32 Data AnaIYSiSOOOOOOOOOOOOOOOOOOOOOOOOOOO0.0.0.0....34 Iv. FINDINGSOOOOOOOOOOO0.0.0..0...0.00.00.00.000000000036 Fundamental Frequency..............................36 AmplitUdeOOOOOOOOOIOOOOOOOOOOOOO0.0.0.0.0000000000043 SYllabi-c DurationO0.0...OO...0.0.0.000000000000000047 V. CONCLUSION AND IMPLICATIONS.0.0.0.000...O0.0.0.0...53 ImplicationSOOOOOOOO0....00....0.00.00.00.00000000058 APPENDICES A. Multivariate Analysis of Variance Tables, Verified Subjects on Test One......................59 B. Multivariate Analysis of Variance Tables, Verified Subjects on Test Two......................61 C. Multivariate Analysis of Variance Tables, Nonverified Subjects on Test One...................63 Table of Contents Continued: D. Multivariate Analysis of Variance Tables, Nonverified Subjects on Test Two...................65 E. Multivariate Analysis of Variance Tables, Verified Subjects, Combined Tests, One and Two.....67 F. Multivariate Analysis of Variance Tables, Nonverified Subjects, Combined Tests, One and Two..69 G. Fundamental Frequency Means and Standard Deviations for Verified and Nonverified Subjects using Know- ledge-Relevant Type Questions......................7l REFERENCESOOO0.0...0..0...0...OOOOOOOOOOOOOOOOOOOOO0.0.0.74 vi LIST OF TABLES Permission to Tape Record Polygraph Session...........28 Fundamental Frequency Means and Standard Deviations for Verified Subjects on Test One.....................38 Fundamental Frequency Means and Standard Deviations for Verified Subjects on Test Two.....................39 Fundamental Frequency Means and Standard Deviations for Verified Subjects on Combined Tests, One and Two..40 Fundamental Frequency Means and Standard Deviations for Nonverified Subjects on Test One..................41 Fundamental Frequency Means and Standard Deviations for Nonverified Subjects on Test Two..................42 Fundamental Frequency Means and Standard Deviations for Nonverified Subjects on Combined Tests, One and Two...42 Amplitude Means and Standard Deviations for Verified Subjects on Test One..................................44 Amplitude Means and Standard Deviations for Verified Subjects on Test Two..................................44 Amplitude Means and Standard Deviations for Verified Subjects on Combined Tests, One and Two...............45 vii List of Tables Continued: ll Amplitude Means and Standard Deviations for Non- 12 13 14 15 16 17 18 19 20 21 22 verified Subjects on Test One.........................46 Amplitude Means and Standard Deviations for Non- verified Subjects on Test Two.........................46 Amplitude Means and Standard Deviations for Non- Verified Duration Subjects Duration Subjects Duration Subjects Duration verified Duration verified Duration Verified Subjects on Combined Tests, One and Two......47 Means and Standard Deviations for Verified on Test One..................................49 Means and Standard Deviations for Verified on Test Two..................................49 Means and Standard Deviations for Verified on Combined Tests, One and Two...............50 Means and Standard Deviations for Non- Subjects on Test One.........................Sl Means and Standard Deviations for Non- Subjects on Test Two.........................51 Means and Standard Deviations for Non- Subjects on Combined Tests, One and Two......52 Multivariate Analysis of Variance for Fundamental Frequency on Verified Subjects, Test One..............59 Multivariate Analysis of Variance for Amplitude on Verified Subjects, Test One...........................59 Multivariate Analysis of Variance for Duration on Verified subjects, Test one...0.0.0.0000...0.0.00.0...60 viii List of Tables Continued: 23 24 25 26 27 28 29 30 31 32 33 Multivariate Analysis of Variance for Fundamental Frequency on Verified Subjects, Test Two..............61 Multivariate Analysis of Variance for Amplitude on Verified Subjects, Test Two...........................61 Multivariate Analysis of Variance for Duration on Verified Subjects, Test Two...........................62 Multivariate Analysis of Variance for Fundamental Frequency on Nonverified Subjects, Test One...........63 Multivariate Analysis of Variance for Amplitude on Nonverified Subjects, Test One........................63 Multivariate Analysis of Variance for Duration on Nonverified Subjects, Test One........................64 Multivariate Analysis of Variance for Fundamental Frequency on Nonverified Subjects, Test Two...........65 Multivariate Analysis of Variance for Amplitude on Nonverified Subjects, Test Two........................65 Multivariate Analysis of Variance for Duration on Nonverified Subjects, Test Two........................66 Multivariate Analysis of Variance for Fundamental Frequency on Verified Subjects, Combined Tests, One and Two...............................................67 Multivariate Analysis of Variance for Amplitude on Verified Subjects, Combined Tests, One and Two........67 ix List of Tables Continued: 34 35 36 37 38 39 40 41 42 Multivariate Analysis of Variance for Duration on Verified Subjects, Combined Tests, One and Two........68 Multivariate Analysis of Variance for Fundamental Frequency on Nonverified Subjects, Combined Tests, One and Two...............................................69 Multivariate Analysis of Variance for Amplitude on Nonverified Subjects, Combined Tests, One and Two.....69 Multivariate Analysis of Variance for Duration on Nonverified Subjects, Combined Tests, One and Two.....70 Fundamental Frequency Means and Standard Deviation for Verified Subjects on Test One/Knowledge-Relevant Type Question.........................................71 Fundamental Frequency Means and Standard Deviation for Verified Subjects on Test Two/Knowledge-Relevant Type Question.........................................71 Fundamental Frequency Means and Standard Deviation for Nonverified Subjects on Test Two/Knowledge- Relevant Type Question................................7l Amplitude Means and Standard Deviations for Verified Subjects on Test One/Knowledge-Relevant Type Question.72 Amplitude Means and Standard Deviations for Verified Subjects on Test Two/Knowledge-Relevant Type Question.72 List of Tables Continued: 43 Amplitude Means and Standard Deviations for Non- 44 45 46 Verified Subjects on Test Two/Knowledge-Relevant Type Question.........................................72 Syllabic Duration Means and Standard Deviations for Verified Subjects, on Test One/Knowledge-Relevant Type Question.........................................73 Syllabic Duration Means and Standard Deviations for Verified Subjects, on Test Two/Knowledge-Relevant Type Question.........................................73 Syllabic Duration Means and Standard Deviations for Nonverified Subjects, on Test Two/Knowledge-Relevant Type QueStiononoone...ooooooooooooooooooooooo00000000073 xi LIST OF FIGURES 1 PM Analyzer graph...0.0..O0....0..0.0.0.0000000000000012 xii I. RATIONALE OF THE STUDY BACKGROUND Among even the best of liars, insofar as outwardly observable symptoms are concerned, there are a considerable number who experience certain non-observable internal sensations of uneasiness or fear of detection, and this is particularly true regarding matters of a serious nature, as in the case of a criminal offense (Reid and Inbau, 1966). It is commonly assumed that the fear of detection, or the fear of the consequences of detection, creates stress within an individual and that this stress may be recognized by obtaining recordings of non-visible physiological phenomena, such as changes in blood pressure, pulse and respiration. Since 1945, the polygraph has been the main scientific tool used for detecting deception. The reliability and validity of polygraph has been historically disputed; however, in the past decade a number of studies indicate a high degree of validity and reliability in analysis of the physiological data obtained during polygraphic examinations (Barland, 1972: Bersh, 1969; Edel and Jacoby, 1975: Horvath and Reid, 1971; Kubis, 1962: Wicklander and Hunter, 1975). In the field of lie detection, there has been a more recent publicized method of detecting deception, the so-called voice stress analyzer. Voice stress analyzers are described by their marketers as capable of detecting lies with an accuracy that equals or exceeds that of polygraph (Horvath, 1982). The basis of conventional voice stress technology is a physiological phenomenon called a micro-tremor, which is attenuated when a subject is under stress. This tremor, believed to be caused by the central nervous system, consists of a low intensity periodic frequency wave which can be detected electronically in voluntary muscles of the body, such as the glottis (Horsley, V. and Schaefer, E.A., 1886: Schaefer, E.A., 1886: Bigland, B. and Lippold, O.C.J., 1954: Marshall, J. and Walsh, E.G., 1956; Van Buskirk, C. and Fink, R.A., 1962). This reaction is purported to affect the speech mechanisms, the result being a diminished or complete masking of a constant and predominant frequency (subperturbation of the voluntary muscles of the larynx) of about ten cycles per second. This absence is detected from the normal vibrations present and is portrayed by some voice stress instruments as a graphic pattern, thereby used to detect stress (Dektor Counterintelligence and Security, Inc., 1977). Inbar and Eden (1976) carried out a study to verify the phenomenon reported by Dektor. They recorded temporal measurements of frequency changes in the human voice and a tremor in the muscles of the vocal area, suggesting that these phenomena are correlated. Surface electromyographic (EMG) recordings were used to estimate tension changes of the muscles in the vocal area. EMG signals were monitored from various places along the vocal tract. Three male and three female subjects were tested and in all cases, the EMG tremor consistently preceded the voice tremor by a few milliseconds. Inbar and Eden believe that their results indicate the voice tremor is induced by the central nervous system (CNS). This was supported by the evidence that the oscillations are random in nature and that the EMG tremor always leads the voice tremor by approximately the same amount of time for a particular vowel. Inbar and Eden admit that their methods need refinement, becauselof the noise present in the recorded EMG, masking tension changes. Kreifeldt and Yao (1974) also reported a major source of difficulty in reading EMG results of a. proportional nature and indicated that the difficulty could be traced to the relatively large amount of noise present in the normally processed EMG output. Olof Lippold, et al., (1957 a and b, 1969 and 1971) examined the physiological tremor described by many of his prede- cessors (Halliday and Redfearn, 1956: Horsley and Schaefer, 1886; Jasper and Andrews, 1938; Marshall and Walsh, 1956: Schaefer, 1886: VanBuskirk and Fink, 1962). In examining a large number of normal human subjects, most of them had some tremor superimposed on their muscular activity. The tremors varied not only from one person to another, but also in the same individual from time to time. It was deduced by Lippold that the central nervous system apparently brings about variations in tremor from minute to minute. Lippold found that a rhythmical tendency existed at a frequency of about ten cycles per second and this occurred in “5193; and probably all muscles in normal individuals.” Most of Lippold's experimentation dealt with outstretched finger muscles. Marshall and Walsh (1956) tested different parts of the body and found the tremor to be similar (about ten cycles per second) at the knee, shoulder, biceps, wrist, gastrocnemius and quadriceps. VanBuskirk and Fink (1962), however, be- lieved the tremor to be a Ballistocardiographic effect rather than a central nervous system function, since they found the tremor existed even though the spinal cord was removed. The tremor amplitude was reduced but the frequency remained the same. These aforementioned studies corroborate the existence of a micro-tremor existing in the voluntary muscles of the human body, at least those muscles of the body examined. Two studies were performed to find out if the tremor existed within the larynx and musculature of speech mechanisms (Shipp and McGlone, 1973 and 1975). It was determined that there wasn't any EMG evidence to suggest that low frequency tremors occurred in the muscles of the larynx which produced the voiced components of speech. McGlone and Hollien (1976) examined the sub-fundamental frequency range produced in unstressed and stressed speech situations. A fast Fourier Transforms Analysis of the EMG recordings by computer was used to check for low frequency vibrations. The results supported the position that tremors do exist in the extrem- ities of the body when in certain states of contraction, but were not found in fast moving central muscles, like the vocal cords. If Shipp and McGlone (1973 and 1975), and McGlone and Hollien (1976) are correct in their findings, it presents a problem with the present day theory of a micro-tremor phenomenon, as the basis of voice stress technology. In addition to this, the effectiveness of voice stress instru- mentation in the detection of deception is a matter of sci- entific controversy (Horvath, 1982). What then, if any- thing, do we know about the cause and effect of emotional stress and vocal responses of a speaker under emotional stress? Speech communication scientists have been studying emotional stress and its physiological effects upon the individual and the speech problems derived from this psychological stress. Studies have shown that psychological stress causes a change in the speaker's fundamental frequency (voice pitch), syl- labic duration (prolongation of vowel sounds), and the amplitude (intensity) of the voice (Alpert, et al., 1963: Fry, 1958: Hall and Hutchinson, 1976: Inbar, et al., 1977: JLieberman, 1959 and.l96l; Lieberman and Michaela, 1962; McGlone and Hollien, 1976: Rubenstein, 1966; Simonov and Frolov, 1973). The change in the fundmental frequency is reported by these researchers as the most dominant and predictable vocal change. STATEMENT OF THE PROBLEM Studies related to voice stress analysis, when correlated with deception, have shown that an increase in the funda— mental frequency occurs when one tries to deceive (Ekman, et al., 1976: Streeter, et al., 1977). Other changes occur in the amplitude (Friedhoff, et al., 1962; Alpert, et al., 1963) when this stress is correlated with deception. The fundamental frequency and the amplitude of the voice is easily determined by vocal analysis and recently a micro- processor controlled device, the PM Analyzer, has been developed that can automatically accomodate sudden and frequent shifts in the fundamental frequency and amplitude of the voice, caused by various emotional or psychological stimuli. The PM Analyzer has been designed for use in a wide range of speech and language applications. This new microprocessor has not yet been tested for the detection of stress and its relationship to deception. IMPORTANCE OF THE STUDY The available scientific literature (Friedhoff, et al., 1962; Alpert, et al., 1963; Ekman, et al., 1976: Streeter, et al., 1977) shows not one of them has involved testing subjects under field conditions. This is a concern because there are obvious differences between field and laboratory subjects, for examples, the attitudes toward the examiner, the testing situation, the purpose of the examination, expectations as to the outcome of the examination, the amount of examiner-subject interaction expected and occur- ring, different motives for volunteering to take the exam- ination, different modes of compliance, different degrees of resentment toward the test, different levels and types of examiner-subject rapport, may all be affecting the stress levels of the individual and thereby effecting the results of the test data. In addition to the short term stress applied to the subject during the testing procedure, there is a long term stress acting upon the non-experimental subject which is almost always absent in the laboratory situation. The experimental populations are highly homog- enous, whereas in field (forensic) populations there are variances in intelligence and educational levels, emotional and mental stability, previous experience with lie detection devices and beliefs regarding the efficacy of the polygraph technique. The present study is important, because to date, no empir— ical research involving fundamental frequency, amplitude and syllabic duration of the human voice has been used to detect deception in actual criminal investigations. Examining the high stress or actual field experiences will allow for results applicable to law enforcement. PURPOSE OF THE STUDY The purpose of this study was to determine whether or not changes in the fundamental frequency (voice pitch), ampli- tude (relative intensity) and syllabic duration (measured in millisecond (ms) for prolongation of vocal utterances) of the human voice are useful in detecting deception. .As previously indicated, studies have shown that people under stress, other than sorrow, have an increase in fundamental frequency and amplitude and a prolongation in syllabic duration (Alpert, et al., 1963; Fry, 1958; Hall and Hutchinson, 1976; Inbar, et al., 1977: Lieberman, 1959 and 1961: Lieberman and Michaels, 1962; McGlone and Hollien, 1976: Rubenstein, 1966: Simonov and Frolov, 1973). There- fore, in this study, vocal changes of persons under stress, specifically those being questioned during polygraph examin- ations for alleged involvement in criminal cases, were examined. DEFINITION OF TERMS The following definitions are provided to create a common basis for understanding the terms and phrases used in this study: 10 fundamental frequency- The frequency of the vocal fold vibration directly determines the lowest frequency (fundamental) of the sound which is produced. The human listener perceives the lowest frequency as the speaker's pitch. The pitch of the human voice is the subjective, psychological perception of sound frequency. Pitch canonly be measured by asking listeners to make judgements, whereas, frequency is a physical parameter which can be measured by instruments. amplitude- A waveform is an abstract representation of the displacement from rest of vibrating particles, through time. The amplitude of this displacement is correlated with intensity or power of the sound. Like frequency, the amplitude or correlated intensity of sound is a physical property of the acoustic signal which can be measured by an instrument. The intensity is directly related to its loudness: as the intensity increases, the sound is judged by listeners to be louder. Loud- ness then, is the subjective, psychological perception of intensity. 11 decibel- sound pressure or intensity is measured in decibel (dBSPL-sound pressure level and dBSIL-sound intensity level): it is one-tenth of a Bel, so named in honor of Alexander Graham Bell, the U.S.A. inventor of the telephone and educator of the deaf. The Bel is a logarithmic ratio between the sound pressure or the alternative intensity of a sound and a given reference, usually the threshold of hearing of a pure tone of l KHz. syllabic duration- duration of uttered phonemes by an individual, either in isolation or in conjunction with other consonants and syllables, measured in milli- second (ms), thousandth of a second. PM Analyzer- a microprocessor that analyzes speech sounds at real-time, displaying the fundamental frequency on the upper half of a split screen video monitor (interfaced with the PM Analyzer). The intensity is displayed on the lower half of the split screen with a second trac- ing representing the envelope of a speech wave (oscill- ographic display); this display permits the user to segment the speech temporarily on the screen, immedi— ately. The measurement of this envelope of speech - 12 length, represents the syllabic duration according to a graphic scale. A B ’k fundamental -~ r-- frequency mean 8 SS 0M2 lulll z 6/:33Hz -and std. dev. envelope envelope opening , ' closure -oscillographic ’ scale . ‘\..1 Rl=28 43cs 12am M:27.sus (:6.0da syllabic duration amplitude mean and std. dev. Figure 1. PM Analyzer Graph. irrelevant questions- are questions asked by the polygrapher of the subject dealing with nonissue matters such as, “are you 21 years of age and did you drive here today?“ These questions are asked during the testing in order to establish a so-called normal or baseline pattern. relevant questions- are questions asked by the polygrapher dealing with the primary issue, such as “did you steal the money?“ and ”did you use a gun to kill John Doe?“ control questions- are questions asked by the polygrapher to the subject dealing with matters similar to, but of 13 presumed lesser significance than the offense being investigated, such as, ”other than what you told me today, have you ever stolen anything else? verified truth- a polygrapher, subsequent to testing the truthful subject, verifies his findings by obtaining a confession from another subject, or by receiving infor- mation from another criminal justice practitioner that a person has confessed to the crime. verified deception- a polygrapher verifies his findings by obtaining a confession from the deceptive subject. nonverified truth- a polygrapher concludes that the subject tested is truthful to the relevant issue; however no confession is obtained from any other subject, thereby not verifying the truthfulness of the subject tested. nonverified deception- a polygrapher concludes that the subject tested is deceptive to the relevant issue, however, he/she does not confess to the crime or any involvement therein. polygraphers control question theorem- when a deceptive person responds to a relevant question, the physio- 14 logical response (e.g.cardiovascular, pneumogastric and galvanic) is greater than the physiological response to a control question of presumed lesser significance: whereas truthful people respond with greater physio- logical response to a control question than to a rele- vant question. Fourier analysis- a mathematical algorithm to analyze (break down) a complex wave into its component simple waves (pure tones of different frequencies and intensities). This analysis was discovered by J.B. Fourier, in France during the first quarter of the 19th century. MANOVA- a multivariate analysis of variance. It is included in the Statistical Package for Social Sciences (SPSS). RESEARCH QUESTIONS The researcm.questions in the present study are adapted from the review of literature related to polygraphy and speech sciences. This study addresses whether or not changes in the vocal parameters of the human voice are useful in detecting deception. This study will examine whether or not the fundamental 15 frequency (voice pitch) changes, due to stress related to deception, when interviewed by a polygrapher for an alleged involvement in a crime. Furthermore, this study will exam- ine two other vocal parameters, that of syllabic duration or prolongation of the vocal utterance and that of the vocal amplitude or relative intensity of the voice to determine whether or not these vocal parameters.change due to stress related to deception. HYPOTHESES TO BE TESTED Based upon the research questions, the following testable hypotheses are hereby presented: Hypothesis No. 1 Given the polygraphers control question theorem, the fundamental frequency (pitch measured in cycle/saHz.) of a deceptive person, when responding to a relevant question, will be greater then when responding to a control question and the fundamental frequency of a truthful person, when responding to a control question, will be greater then when responding to a relevant question. 16 Hypothesis No. 2 Given the polygraphers control question theorem, the amplitude (correlated to intensity measured in dBSPL) of a deceptive person, when responding to a relevant question, will be greater then when responding to a control question and the amplitude of a truthful person, when responding to a control question, will be greater then when responding to a relevant question. Hypothesis No. 3 Given the polygraphers control question theorem, the syllabic duration (prolongation of a vocal utterance measured in ms) of a deceptive person, when responding to a relevant question, will be greater then when responding to a control question and the syllabic duration of a truthful person, when responding to a control question, will be greater then when responding to a relevant question. II. REVIEW OF RELATED LITERATURE A review of literature provides insight regarding previous methods of detecting deception, present day methods of detecting deception and experimental advances toward in- creasing ones knowledge of the present state of the art. This review further presents an outlook to future methods of detecting deception by using physiological responses of the human voice. The review of related literature is organized under three major headings: Polygraph, Acoustics of Speech Production and Psychological Emotion. POLYGRAPH The first attempt to use a scientific instrument in an effort to detect deception occurred about 1895, when Cesare Lombroso published an account of several experiments. En- couraged by the reported successes of his predecessors, John A. Larson, in 1921, constructed an instrument capable of continuously recording all three phenomena: blood pressure, pulse and respiration. In 1926, Leonarde Keeler constructed a more satisfactory instrument than the one used by Larson and is generally credited with developing the prototype of 17 18 the polygraph instrument now used in most field settings. In 1949, Keeler made some additional changes, adding a galvanometer for recording what is known as the galvanic skin reflex or electrodermal response (G.S.R.). Horvath (1976, p.109) has indicated that there has been success at detecting deception by other measurements of physiological activity: “such as hand.tremors, electroencephalic activity, pupil dilation, oculomotor activity, voice modulation and oxygen- ation of the vascular system. What is now agreed upon by field examiners is that any attempt at detecting deception must be made with an instrument that records both cardiovascular and respira- tory activity; It is in fact illegal in some states for a detectionlof deception examiner to use an instrument not capa- ble of recording these two parameters, although others, particularly electro- dermal activity, are also commonly recorded in conjunction with them." The measure which has shown the greatest success in discrim- inating between truthfulness and deception in laboratory studies has been electrodermal activity, including the skin resistance response (SRR), (Ellson, et al., 1952; Thackray and Orne, 1968) and the skin potential response (SPR), (Lindsley, 1955: Thackray and Orne, 1968). Yet the most influential field examiners consider electrodermal activity 19 to be the least effective of the measures used in the field (Arthur, 1971: Lee, 1953: Marston, 1938; Reid and Inbau, 1966). Field Procedures in Polygraph Testing The polygraph examination consists of three stages: the pre-test interview, the testing stage and the post-test interrogation, when appropriate. However, no test should ever be conducted without an interview with the investi- gators. Unless a polygrapher is fully informed about the case, he/she will not be in a position to conduct an adequate pre-test interview with the subject. Pretest Interview While there are differences between pre-test interviews from one polygrapher to another, the differences lie in the nature of the objectives formulated during the interview with the investigators and the knowledge the polygrapher gains by learning the case factsland background of the subject. There are irrelevant questions, those used for establishing a normal or truth-telling response dealing with such matters as: are you 21 years of age and did you drive here today? There are relevant questions, those which 20 pertain to the matter under investigation: such as, did you shoot Mary Smith and did you use a gun to kill Mary Smith? Control questions, are those resulting from past behavior of the subject and what was learned about the subject during the pre-test interview. In general, they deal with matters similar to, but of presumed lesser significance than the offense being investigated. An example might be: did you ever steal anything else, other than what you told me today, or other than what you told me today, have you ever lied to another to avoid getting caught? The polygrapher develops these questions in such a way that the subject will answer no, but will, in all probability be lying or at least will have some doubt or concern about the truthfulness or accu- racy of his answer (Horvath, 1976). The control questions work to the advantage of the innocent subject by diverting him/her from the relevant questions and focusing his/her concerns on the control questions. On the other hand, the guilty subject will be more concerned about the relevant questions than the lies that he makes about inconsequen- tial or irrelevant issues. The irrelevant, relevant and control questions are estab— lished during this pretest interview. With the completion of the review of all questions, the pre-test interview draws 21 to a close and the transition is made to the data acquisi- tion phase of the examination. Actual Testing In polygraphic testing, the polygrapher asks the subject the previously reviewed irrelevant, relevant and control ques- tions in a series of polygraph tests. Each test generally consists of about ten or eleven questions: four irrelevant, two control and four or five relevant questions and will usually last about three minutes. A complete examination consists of several of these tests administered before a determination of deception is made. It is often claimed that response data contained in the first two tests are sufficient to indicate the subject's truthfulness or decep- tion. Post Test If the testing reveals no indication of deception, the sub- ject is thanked for his/her cooperation and released. De- pending upon the polygrapher and the situation, the subject may be told that the results will be given to the investi- gators when they become available. The latter maneuver per- mits more control over the subsequent investigation. If the 22 examination indicates that the subject was lying to one or more of the relevant questions, he/she is usually interro- gated in order to seek an admission of involvement in the issue under investigation. ACOUSTICS OF SPEECH PRODUCTION Speech sounds are produced by resonance of glottal vibra- tions and/or friction noise in the vocal tract and nasal cavities. The vocal cords vibrate medially and the vibra- tory cycle of being forced apart and then returning to the medial position creates a complex wave. The frequency vibration of a simple wave is measured in Hertz (Hz) in honor of Dr. Heinrich R. Hertz, the physicist who investigated electromagnetic vibrations. The lowest frequency (fundamental) as well as the frequency of a peri- odic speech wave uttered by a speaker is perceived as the “pitch" of the speaker's voice. The fundamental frequency of a speaker's voice is easily determined by voice analysis equipment (e.g. sound spectro- graph, pitch meters and the PM Analyzer), it is measured in Hz. Since Frequency is perceived as pitch, the higher the frequency of a sound the higher the perceived pitch. 23 The frequency range of speech sounds (complex waves) goes from approximately 100 to 7000 Hz. Most spectral infor- mation regarding speech, including talker-dependent fea- tures, is found within a range of 100 to 4000 Hz. A human ear in optimal conditions can perceive vibration from 20 to 20,000 Hz. as sound, providing that the amplitude (inten- sity) is above threshold. Amplitude is the maximum displacement of equilibrium within the elastic medium. Amplitude (intensity) is perceived as loudness, usually measured in decibel (dB), which is a log- arithmic ratio between the intensity (or alternative pres- sure) of the sound being measured and a given reference, usually the threshold of hearing at 1000 Hz. When speaking, an individual tends to establish his/her own intensity level of response and when an emotional stimuli is introduced, there arises a tendency to depart from a pre-set level of response and the stress responses tend to be less like each other. Some individuals respond by increasing the intensity of the voice, while others decrease their inten- sity ( Friedhoff, et al., 1962). Even when a lie was con- doned by an experimenter, the emotional stimuli was still sufficient to produce measureable changes; however, "these cues may be more relevant for the trained clinician in 24 evaluating the emotional state of his patient than the usually measured silent systems such as skin resistance and blood pressure.‘I (Alpert, et al., 1963, p.365). PSYCHOLOGICAL EMOTION The third area of study involves emotional or psycho- logical stimuli. It has been found that the effects of emotion on acoustic characteristics of speech show that average values and ranges of fundamental frequency differ from one emotion to another. In a study by Williams and Stevens (1972), they reported that the lowest fundamental frequency was obtained for the emotion of sorrow (usually lower than that for neutral situations), and the highest was for anger. Lehiste and Peterson (1959) found that funda- mental frequency and syllabic duration were among the most commonly cited parameters that influence a listener's judgement about stress. Coker, et al.,(1973, p.440) reported that: “Newly introduced improbable words (ambiguous speech or that which is indistinct) that are difficult to guess from context, produce high stress within the speaker. The first occurrence of an unlikely noun, verb or other substantive is pronounced slowly and distinctively with almost emphatic pitch values.. subsequent repetitions of the word, showed pitch and other attributes of stress were diminished.“ 25 Malcolm Brenner (1974) reported that stress does occur with- in an individual and is ever increasing as a power function of audience size. In other words, as the number of listen- ing people increases, a speaker's psychological stress increases. In another study performed by Hall and Hutch- inson (1976) there was found an increase in fundamental frequency and a prolongation of syllabic utterances for each speaker tested, when they read a slurvian passage (phonetic sequences involving improbable word combinations). Simonov and Frolov (1973) investigated voice frequency as related to physical and emotional well-being of pilots and cosmonauts, as well as actors. Their method of study using a one-third octave spectral analyzer permitted them to differentiate adequately the degree of emotional stress in 85 percent of all the cases, without singling out separate sounds, which is very important in real conditions, espe- cially in the presence of ambient distortions in EMG output. Body movements and voice pitch were studied by Ekman, et al., (1976) using sixteen student nurses in an interview process, where each subject watched a short film and then was asked questions concerning her feelings about it. One film was a pleasant feeling nature film and the other was a 26 film showing amputations and burns that intended to elicit a strong unpleasant effect. However, they were instructed to conceal their negative feelings and convince the interviewer they had seen another pleasant film. In deception regarding the second film being a pleasant effect, there was a sig- nificant decrease in illustrators (p<0.05), a trend for an increase in shrugs (p<0.10), an increase in pitch (p<0.05), and no change in adaptors or total hand activity. Low pitch was associated with observers' judgements that a person was sociable, calm and relaxed, whereas, negatively correlated with illustrators (Rho = -0.61, p<0.01), pitch became higher in deceptions. A study of pitch changes during attempted deceptions was conducted by Streeter, et al., (1977) using 32 pairs of male college undergraduates. The students were paired as inter- viewer and interviewee unknown to each other. The inter- viewee was told to lie to certain questions asked. The interviewer did not know which questions the interviewee lied about. On the average, fundamental frequency was 3.3 Hz. higher when the subjects lied than when they told the truth. III. RESEARCH METHODS AND PROCEDURES “This chapter is divided into four categories: population and sample, instrumentation and data collection, types of questions for analysis and data analysis. POPULATION AND SAMPLE The population for this study consisted of general American mid-western speaking individuals from southern-lower Michigan. The individuals recorded were male and female subjects being tested by ten licensed polygraph examiners of the Michigan State Police Forensic Science Division. The individuals were tested for their alleged involvement in crimes of the State and each individual volunteered to be audio tape recorded during their polygraph examination. After the initial introduction of the goals and objectives of the study, it was explained to each polygrapher the importance of the subjects being tested, for them to give their complete cooperation and for the polygrapher to keep 'the identity of the subject tested.in.strict confidence by omitting his/her identity from the audio tape recordings. 'The “Permission to Tape Record Polygraph Sessions" form was explained and provided to the polygraphers to present before: each individual for their perusal and signing. 27 28 Table 1. Permission to Tape Record Polygraph Session fia— I hereby give the Michigan State Police permission to tape record the pre-test interview and actual polygraph examination. All tape recordings are to be used for research purposes and will be identified by number with names being kept in strict confidence. Participation is strictly voluntary. The tape recordings will have no bearing on the polygraph examination being conducted, but will be used only for testing present and future voice analyzers for detecting deception. Signed and Date Witness: The objective was to obtain a minimum of 25 verified truthful and 25 verified deceitful audio tape recordings of polygraph subjects being tested. Many more recordings were obtained initially by the polygraphers and submitted to the researcher because some verifications could only be made after other tests were run by the examiner or after time had passed in the investigation. Therefore, non-verified audio tape recorded polygraph examinations were also received by the researcher for analysis. For the purpose of this study, verified deceptive audio tape recorded polygraph examinations are only those examinations where, subsequent to polygraph testing, the subject being tested confesses to the polygrapher or confesses to another 29 criminal justice practitioner, as having committed or being involved in the crime that he initially denied. A verified deceitful examination would not exist , if a judge or jury found the subject guilty, while the individual maintained his/her innocence, or if some plea bargaining was accomplished by the prosecutor. A verified truthful subject is one being verified by the polygrapher after receiving a confession from another subject, or by the polygrapher receiving information from another criminal justice practitioner that another person had confessed to the crime subsequent to the first subject being tested as truthful. INSTRUMENTATION AND DATA COLLECTION Pearlcorder, Model MES remote condenser microphones (Lavaliere type) were placed two to three inches from each individual's mouth during the polygraph examination in order to assure quality voice recordings. The microphones were ultra light and compact so that they could easily be attached to the upper pneumographic tube placed around the individuals' upper chest. With the subjects eyes closed and head slightly bent downward during the testing period, the sensitive microphones had no difficulty transducing the subjects' voice to the tape recorder. 30 The tape recorders used were Sony brand, portables, using standard size cassettes (Sony LNC 90) at a taping speed of 1-7/8 ips. This size tape provided 45 minutes recording time per side, which was sufficient to record both control question tests, which usually lasts less than ten minutes. A preliminary analysis was performed in order to ascertain whether or not the recordings were of sufficient quality to be used on the PM Analyzer. Some recordings were found to be of insufficient quality due to ambient noise (e.g. air conditioner/heat duct noise and/or motor traffic noise outside the building bleeding through the windows and doors) and recording techniques (e.g. improper grounding causing a 60 cycle hum masking the speech signal, or the polygrapher forgetting to turn the microphone on, and/or putting the recorder into play mode instead of the record mode). A total of 125 subjects (47 confession-verified deceptive, 33 verified truthful, 24 nonverified deceptive and 21 non- verified truthful) were utilized for analysis, as to their truthfulness or deception. The reason for having more sub- jects than proposed was that verifications came in slowly and voice analysis was performed on all quality audio tape recordings as they were received from the polygraphers, contemplating subsequent verification. The audio tape 31 recordings that were determined to be of sufficient quality were transfered to the PM Analyzer by hardwire for proper transfer of the speech signal without ambient noise interfering. The PM Analyzer displays (Figure 1, page 15) simultaneously the fundamental frequency on the upper half of a split- screen video monitor (interfaced with the PM Analyzer) and the intensity on the lower half of the split-screen video monitor, along with a second tracing representing the envel- ope of speech (oscillographic display) that permits the user to segment the speech on the screen immediately. Two cur- sors move from point A (opening of the envelope of speech) to point B (closing of the envelope of speech). The ana- lyzer instantly reads on-screen the mean values of the fundamental frequency and the amplitude (intensity) of the utterance. There is also a digital display of the standard deviation of both frequency and amplitude. The slow sweep of the screen display allows for measurement of phonemic sounds (e.g. syllables and consonants) over-time in milli- seconds (msec.) and depict a speakers' prolongation of speech sounds. Since the PM Analyzer analyzes at real-time, it was decided to interface the video output with a thermal printer, for 32 reasons of efficiency and expediency. A Mitsubishi PSOU Video Printer was used for this study because it produces hard copies within 15 seconds. The reproduction is an exact image of the video screen. These reproductions are on a continuous roll of paper, so that many polygraph test ques- tion responses can be examined and the records kept on each test with little difficulty. QUESTIONS FOR ANALYSES Prior to statistical analysis, contact was made with each polygrapher to ascertain which relevant question was the primary question covering the issue. The primary relevant question was the question that the verified deceitful sub- ject confessed to and/or the verified truthful subject was found to have no knowledge about or to have committed. A primary relevant question on a crime of auto theft would be, 'did you steal the auto?“ or a primary knowledge relevant question for the same test would be, "do you know who stole the auto?" The primary control question that was compared to the primary relevant question was also determined from each polygrapher. The primary control question would be similar to the primary relevant question, only with lesser 33 significance. In the above example of an auto theft, a primary control question might be: “other than what you told me today about your past behavior, have you ever stolen any- thing else in your life?‘I This question for a deceptive person, theoretically would have lesser significance than the actual auto theft issue, thereby providing less stress in his/her voice; whereas the truthful subject, theoreti- cally will have enough concern about his/her past behavior, so that his/her answer will provide greater stress in their voice for the less significant question. Michigan law requires at least two control question tests be given each individual examined by polygraph. Other tests may be given during the polygraph examination. However the number of tests given will depend upon the factors mentioned in the first chapter, where the differences between field and laboratory examinations were discussed. The individuals interaction with the examiner, modes of compliance, resent- ment towards being tested and/or polygraph technique in general. For the purpose of analysis, two control tests per individ- ual were used with one primary relevant question response compared to one primary control question response. 34 Each individual tested was asked by the polygrapher to respond with a one word "yes' or a one word “no." They were asked not to nod their head, repeat “uh, uh,” "no I didn't," "yes I did,“ or any like response. Each subjects' verbal response to the primary relevant, primary control and relevant knowledge questions asked in test numbers one and two, produced individual data for each of the three vocal parameters (fundamental frequency, amplitude and syllabic duration). The knowledge-relevant type question responses by the individuals were determined to be predominantly located in the category of verified deceptive and nonverified truthful subjects tested; however not enough of these type questions were asked each subject in order to provide sufficient data for testing. The knowledge-relevant question means and standard deviations are located in Appendix G. DATA ANALYS I S The fundamental frequency of the individuals' voice was considered the measurement recorded in cycles per second, commonly called Hertz (Hz). The Amplitude of the individ- uals' voice likewise was considered the measurement recorded in decibel (dB), a logarithmic ratio between the 35 intensity of the sound uttered and a given reference, usu- ally the threshold of hearing. Syllabic duration of the individuals' voice, the third parameter measured was consid- ered the individuals' prolonged voiced sound as a result of the response to the questions asked by the polygrapher and was measured in milliseconds (ms), thousandth of a second. These measurements of the human voice for each primary relevant, primary control and relevant knowledge question became the raw data used for the statistical analysis. A computer Statistical Package for Social Sciences (SPSS) was used to analyze the raw data collected. Arithmetic means and the standard deviations were calculated for each vocal parameter for each relevant question and each control question response on each of two control question tests for all verified and nonverified subjects. A multi- variate analysis of variance (MANOVA) with repeated measures of a proportional design was computed on each of the vocal parameters for each of the two tests and the combined mea- surements of both tests. The level of significance estab- lished for this study is 0.05. IV. FINDINGS The data analyses are presented in this chapter and analyzed according to the procedures described in Chapter III. The analyses are organized into two categories: verified and nonverified subjects. Each of the hypotheses presented in Chapter I is examined, using each vocal parameter (funda- mental frequency, amplitude and syllabic duration) for each test. The polygraphers control question theorem, states that a deceptive person's physiological response to a relevant question is greater than his/her response to a control question and a truthful persons physiological response to a control question is greater than his/her response to a rele- vant question. Therefore, it would be expected that the fundamental frequency, the amplitude and the syllabic duration of deceptive persons should show an increase when responding to relevant questions, greater then when re- sponding to control questions. The same vocal parameters of truthful persons should show an increase when responding to control questions, greater then when responding to relevant questions. FUNDAMENTAL FREQUENCY The first hypothesis is that the fundamental frequency 36 37 (pitch measured in cycle/s=Hz) of a deceptive person, when responding to a relevant question, will be greater then when responding to a control question and the fundamental fre- quency of a truthful person, when responding to a control question, will be greater then when responding to a relevant question. The rationale for this hypothesis is that'studies show that a person under emotional stress has an increase in his/her fundamental frequency; therefore, a deceptive person lying to a relevant question under emotional stress due to a fear of losing his/her freedom, will have an increase in his/her fundamental frequency. Verified Subjects The first hypothesis was tested by using the fundamental frequency means of 33 verified truthful and 47 verified deceptive subjects' vocal responses to relevant and control questions. If the hypothesis is true, it would be expected that the interaction between the types of questions (control and relevant) and the types of subjects (truthful and decep- tive) would prove statistically significant, thus supporting the hypothesis. The statistical results are shown in tables two through four with the complete MANOVA 38 tables located in appendices A, B and E. Table two shows the fundamental frequency means and standard deviations for verified subjects on test one, with the results of the interaction effect proving significant [F(1,77)= 11.16, P=0.0008]; therefore as predicted, decep- tive subjects responding to relevant questions showed a greater fundamental frequency then when responding to control questions and truthful subjects responding to con- trol questions showed a greater fundamental frequency then when responding to relevant questions. Table 2. Fundamental Frequency Means and Standard Deviations for Verified Subjects on Test One. Type of y'.__ Type of Question Examination Relevant Control Truthful . Mean 123.8 134.9 s.d 37.7 39.6 Deceptive f Mean 112.8 111.5 s.d. 27.9 25.7 Question Type: §(1,77)= 4.44425,P=0.038 Exam Type: F(l,77)= 5.82127,P=0.018 Quest Type/Exam Type: F(1,77)= 11.16292, P= 0.0008 Table three shows the fundamental frequency means and standard deviations for verified subjects on test two with the results of the interaction effect [F(1,77)= 2.135, 39 0.148], not proving significant. Although the interaction effect was not statistically significant, the subjects' responses were in the predicted direction. 'Lable 3. Fundamental Frequency Means and Standard Deviations for Verified Subjects on Test Two. Type of I; ‘ Type ofguestion Examination Relevant Control Truthful Mean 126.4 128.3 s.d. 39.7 39.6 Deceptive . Mean 113.5 112.8 s.d. 27.2 27.0 Question Type: FI1,77)= 0.23353, P20.630 Examination Type: F(1,77)= 3.65488, P80.060 Quest. Type/Exam Type: F(1,77)= 2.13548, P= 0.148 Table four shows the fundamental frequency means and standard deviations for verified subjects on the combined tests, one and two, with the results of the interaction effect [F(1,78)= 3.065, P8 0.084], not proving significant. However, the results were in the predicted direction and approached statistical significance. 40 Table 4. Fundamental Frequency Means and Standard Deviations for Verified Subjects on Combined Tests, One and Two. Type of Type of Qpestion Examination Relevant Control Truthful Mean 125.1 132.0 s.d. 39.0 40.0 Deceptive ‘ Mean 113.1 113.0 s.d. 28.0 27.3 Question Type: F(1,78)= 4.45956, p- 0.038 Examination Type: F(l,78)= 5.51173, P8 0.021 Qpest. Type/Exam Type: F(1,78)- 3.06490, P= 0.084 Nonverified Subjects The first hypothesis was also tested by using the funda- mental frequency of 21 nonverified truthful and 24 nonveri- fied deceptive subjects vocal responses to relevant and control questions. If the hypothesis is true, it would be expected that.the interaction between the types of questions and the types of examined subjects would prove statistically' significant. The statistical results are shown in tables fiveethrough seven, with the complete MANOVA tables located in appendices C,D and F. Table five shows the fundamental frequency means and standard deviations for nonverified subjects on test one, 41 with the results of the interaction effect [F(1,43)= 0.09887, P= 0.755], not proving significant, although they were in the predicted direction. Table 5. Fundamental Frequency Means and Standard Deviations for Nonverified Subjects, Test One. Type of ' Type of Question Examination Relevant Control Truthful - . Mean 134.8 135.8 s.d. 34.3 29.8 Deceptive .1 Mean 133.9 133.5 s.d. 35.3 39.7 Question Type: FTI,43>= 0.01347} pa 0.900 Examination Type: F(1,43)= 0.02292, P= 0.880 Quest. Type/Exam Type: F(1L43)= 0.09887L7P8 0.755 Table six shows the fundamental frequency means and standard deviations for nonverified subjects on test two, with the results of the interaction effect [F(l,43)= 6.13, P3 0.017], proving significant. The interaction effect supported the control question theory providing evidence that truthful subjects showed a greater fundamental frequency when respon- ding to control rather than relevant questions and deceptive subjects showed a greater fundamental frequency when responding to relevant rather than control questions. 42 Table 6. Fundamental Frequency Means and Standard Deviations for Nonverified Subjects, Test Two. Type of TT Type 0?:Question Examination Relevant Control Truthful _ Mean 132.6 138.3 s.d. 35.9 33.5 Deceptive ,, Mean 132.2 126.0 s.d. 39.9 36.8 Question Type: F(l,43)= 0.07201, pa 0.790 Examination Type: F(l,43)= 0.35496, P8 0.554 Quest. Type/Exam Type: F(l,43)= 6.13417, P8 0.017 Table seven shows the fundamental frequency means and standard deviations for nonverified subjects on the combined 'tests, one and two, with the results of the interaction effect [F(l,43)= 6.447, P: 0.015], proving significant. Table 7. Fundamental Frequency Means and Standard Deviations for Nonverified Subjects, Combined Tests, One and Two. Type of Type of Question Examination Relevant Control Truthful Mean 134.0 137.1 s.d. 35.1 31.6 Deceptive _ Mean 133.1 130.0 s.d. 38.0 38.2 Question Type: F(l,43)= 0.02361, p= 0:579 Examination Type: F(l,43)= 0.14914, P= 0.701 Quest. Type/Exam Type: F(l,43)= 6.44713, P= 0.015 43 AMPLITUDE The second hypothesis states that the amplitude (correlated to intensity measured in dBSPL) of a deceptive person, when responding to a relevant question, will be greater then when responding to a control question and the amplitude of a truthful person, when responding to a control question, will be greater then when responding to a relevant question. The rationale for this hypothesis is that studies show that people under emotional stress have an increase in their armr- 1itude; therefore, a deceptive person lying to a relevant question under emotional stress due to a fear of losing his/her freedom, will have an increase in his/her amplitude on the relevant question as opposed to a control question. Verified Subjects The second hypothesis was tested by using the amplitude means of 33 verified truthful and 47 verified deceptive subjects' vocal responses to relevant and control questions. If the hypothesis is true, it would be expected that the interaction between the types of questions (control and relevant) and the types of examined subjects (truthful and deceptive) would prove statistically significant, thus 44 supporting the hypothesis; however, the interaction effect, though in the predicted direction, did not prove significant for the verified subjects. Since the amplitude did not prove significant the statistical results will not be described in detail, but are shown in tables eight through ten, with the complete MANOVA tables located in appendices A, B and E. Table 8. Amplitude Means and Standard Deviations for Verified Subjects on Test One. Type of T Type of Question Examination Relevant Control Truthful . _ Mean 24.4 24.5 s.d. 3.5 3.4 Deceptive ‘,_ Mean 25.8 25.7 s.d. 3.3 3.0 Question Type: r<1,77)= 0.04261, 9:0.837 Examination Type: F(l,77)= 3.15684, P=0.080 Question Type/Exam Type: F(l,77)= 0.08427,P= 0.772 Table 9. Amplitude Means and Standard Deviations for Verified Subjects on Test Two. Type of F _ Type of Question Examination Relevant Control Truthful ‘ M Mean 24.4 24.5 s.d. 3.8 3.6 Deceptive . Mean 25.6 25.5 s.d. 3.5 3.2 Question Type: F(1,77)# 0.00023, P=0.988 Examination Type: F(1,77)= 1.92493, P=0.169 Quest. Type/Exam Type: F(l,77)= 0.21716, P= 0.643 45 Table 10. Amplitude Means and Standard Deviations for Verified Subjects on Combined Tests, One and Two. Type of Type of Questions Examination Relevant Control Truthful (H Mean 24.4 25.0 s.d. 3.6 3.5 Deceptive Mean 25.7 25.6 s.d. 3.4 3.1 Question Type: F(l,78)= 0.49790, p= 0.483 Examination Type: F(l,78)= 0.80540, P=0.372 Quest. Type/Exam Type: F(l,78)= 0.13048, P= 0.719 Nonverified Subjects The second hypothesis was also tested by using the amplitude of 21 nonverified truthful and 24 nonverified deceptive subjects vocal responses to relevant and control questions. If the hypothesis is true, it would be expected that the interaction between the types of questions and the types of examined subjects would prove statistically significant. However, the interaction effect though in the predicted direction with the exception of the nonverified truthful subjects on test two (Table 12), did not prove significant for the nonverified subjects; statistical results will not be described in detail, but are shown in tables eleven through thirteen, with the complete MANOVA tables located in appendices C, D and F. 46 Table 11. Amplitude Means and Standard Deviations for Nonverified Subjects, Test One. Type of Type of Question Examination Relevant Control Truthful ' _ Mean 24.7 25.1 s.d. 4.9 4.3 Deceptive ' . Mean 25.3 24.7 s.d. 3.3 3.5 Question Type: FT1,43)- 0.50700, pa 0.651 Examination Type: F(l,43)= 0.00579, P: 0.940 Quest. Type/Exam Type: F(l,43)= 2.26013, P= 0.140 Table 12. Amplitude Means and Standard Deviations for Nonverified Subjects, Test Two. Type of . Type of Question Examination Relevant Control Truthful Mean 24.1 23.9 s.d. 5.3 4.8 Deceptive Mean 25.3 24.9 s.d. 3.8 3.9 Question Type: F(l,43)= 0.84899, P= 0.362 Examination Type: F(l,43)= 0.64366, P= 0.427 Quest. Type/Exam Type: F(l,43)= 0.07964, P= 0.779 47 Table 13. Amplitude Means and Standard Deviations for Non- Verified Subjects on Combined Tests, One and Two. Type of > Type of Questions Examination Relevant Control Truthful "_Q . Mean 24.4 25.0 s.d. 5.1 5.0 Deceptive -,7 Mean 25.3 25.0 s.d. 4.0 4.0 Question Type: F(l,43)= 1.08196, P= 0.304 Examination Type: F(l,43)= 0.22005, P- 0.641 Quest. Type/Exam Type: F(l,43)= 1.77549, P= 0.190 SYLLABIC DURATION The third hypothesis states that.the syllabic duration (prolongation of a vocal utterance measured in ms) of a deceptive person, when responding to a relevant question, will be greater then when responding to a control question and the syllabic duration of a truthful person, when re- sponding to a control question, will be greater then when responding to a relevant question. The rationale for this hypothesis is that studies show that people under emotional stress have an increase in their syl- labic duration; therefore, a deceptive person lying to a relevant question under emotional stress due to fear of 48 losing his/her freedom, will have an increase in his/her syllabic duration on the relevant question as opposed to a control question. Verified Subjects This hypothesis was tested by using the syllabic duration means of 33 verified truthful and 47 verified deceptive subjects' vocal responses to relevant and control questions. If the hypothesis is true, it would be expected that the interaction between the types of questions (control and relevant) and the types of examined subjects (truthful and deceptive) would prove statistically significant. The interaction effect did not prove significant for the veri- fied subjects. However, the main effect for the type of examined subjects (truthful and deceptive) on tests one, two and the combined measurements of tests one and two showed that the verified deceptive subjects had a significantly greater syllabic duration than the verified truthful sub- jects. This does not reflect the control question theory and no predictions were made relative to the main effects. Statistical results of the sample means and the effects of the interaction between the types of questions and the types of examinations will not be described in further detail, but 49 are shown in tables fourteen through sixteen, with the complete MANOVA tables located in appendices A, B and E. Table 14. Duration Means and Standard Deviations for Verified Subjects on Test One. Type of " Type of Question Examination Relevant Control Truthful . Mean 37.9‘ 36.7 s.d. 8.3 7.8 Deceptive . Mean 42.7 42.0 s.d. 10.7 9.2 Question Type: F(l,77)= 1.23721, P=0.270 Examination Type: F(l,77)= 6.88489, P=0.011 Quest. Type/Exam Type: F(l,77)= 0.02038, P= 0.887 Table 15. Duration Means and Standard Deviations for Verified Subjects, Test Two. Type of n _ Type of Question Examination Relevant Control Truthful Mean 37.1 38.1 s.d. 8.5 9.6 Deceptive ‘ Mean 42.7 42.4 s.d. 11.3 10.3 Question TypeiTF(1,77is 0.09394, p=0.7§0 Examination Type: F(l,77)= 5.34759, P=0.023 Quest. Type/Exam Type: F(l,77)= 0.55133, P= 0.460 50 Table 16. Duration Means and Standard Deviation for Verified Subjects on Combined Tests, One and Two. Type of . Type of Questions Examination Relevant Control Truthful 1 Mean 38.0 37.4 s.d. 8.4 8.7 Deceptive ‘ . \ Mean 43.0 42.2 s.d. 11.0 9.8 Question Type: F(l,78)= 0.00884, 9- 0.925 Examination Type: F(l,78)= 4.39024, P= 0.039 Quest. TypeZExam Type: F(l,78)= 0.00065, P= 0.980 Nonverified Supjects ‘The third hypothesis was also tested by using the syllabic duration of 21 nonverified truthful and 24 nonverified deceptive subjects vocal responses toirelevant and control questions. If the hypothesis is true, it would be expected that the interaction between the types of questions and the types of examined subjects would prove statistically signif- icant. The interaction effect did not prove significant for nonverified subjects; therefore, the statistical results will not be described in detail, but are shown in tables seventeen through nineteen, with the complete MANOVA tables located in appendices C, D and F. 51 Table 17. Duration Means and Standard Deviations for Nonverified Subjects, Test One. Type of i _ Type of Question Examination Relevant Control Truthful Mean 37.1 36.4 s.d. 8.1 8.4 Deceptive Mean 39.5 38.0 s.d. 7.8 8.1 Question Type: F(l,43)8 1:93654, P8 0.169 Examination Type: F(l,43)8 0.77840, P8 0.383 Quest. Type/Exam Type: F(l,43)8 0.25740, P8 0.615 Table 18. Duration Means and Standard Deviations for Nonverified Subjects, Test Two. Type of _‘ Type of Question Examination Relevant Control Truthful Mean 37.3 36.7 s.d. 7.8 8.8 Deceptive ,. Mean 40.8 39.0 s.d. 9.0 7.5 Question Type: §(1,43)-_1.41385, pa 0.241 Examination Type: F(l,43)8 1.71324, P8 0.198 Quest. Type/Exam Type: F(l,43)= 0.33414, P8 0.566 52 Table 19. Duration Means and Standard Deviations for Non- Verified Subjects on Combined Tests, One and Two. Type of , Type of Questions Examination Relevant Control Truthful . Mean 37.2 37.0 s.d. 8.0 9.0 Deceptive , . Mean 40.2 39.0 s.d. 8.4 8.0 Question Type: F(l,43)8 2.52427, 9: 0.100 Examination Type: F(l,43)8 1.34568, P8 0.252 Quest. Type/Exam Type: F(l,43)= 0.52128, P8 0.474 V. DISCUSSION AND IMPLICATIONS The purpose of this study was to determine whether or not changes in the fundamental frequency (pitch measured in cycle/s=Hz), amplitude (correlated to intensity measured in dBSPL) and syllabic duration (prolongation of vocal utter- .ances) of thezhuman voice are useful in detecting deception. More specifically, this study examined vocal changes of subjects being questioned during polygraph examinations for alleged involvement in criminal cases. The changes re- sulting from vocalized responses to relevant questions were compared to those resulting from responses to control ques- tions to determine if the changes were useful in discrimi- nating between truthful and deceptive persons as set forthI in the control question theory of polygraph testing. In this chapter the research findings as they relate to previously mentioned research will be discussed. ‘What these findings imply for future research and for educating and training present-day polygraphers, also will be discussed. Earlier studies involving human speech under emotional stress depicted changes in the vocal parameters (i.e. fundamental frequency, amplitude and syllabic duration) based upon listener perceived judgements and muscular and acoustical measurements involving stress caused by 53 54 electrical shock, actors portraying emotional trauma and a few real-life stressful situations. Studies performed by Coker, et al., 1973; McGlone and Hollien, 1976; Simonov and Frolov, 1973; and Williams and Stevens, 1972 concerned themselves with the changes in fundamental frequency of the human voice due to psychological stress. Changes in ampli- tude along with the fundamental frequency was examined by Rubenstein in 1966 and changes in the syllabic duration along with the fundamental frequency was examined by Lehiste and Peterson in 1959 and by Hall and Hutchinson in 1976. Changes in all three vocal parameters (fundamental fre- quency, amplitude and syllabic duration) were studied by Fry, 1958; Lieberman, 1959 and 1961 and Lieberman and Michaels in 1962. Though all studies related changes in the vocal parameters to psychological stress, none of these studies related changes in the human voice to stress due to» deception. Relatively few studies to date involve changes in the fundamental frequency and amplitude of the human voice in detecting deception. Syllabic duration, until now, has never been used in studies for measuring stress related to deception. Studies to date (Alpert, et al., 1963; Ekman, et al., 1976; Friedhoff, et al., 1962 and Streeter, et al., 1977) have chosen a single vocal parameter, either the 55 fundamental frequency or amplitude to test for stress related to deception. Ekman, et al.,(l976) and Streeter, et al.,(l977) examined the fundamental frequency (pitch measured in cycle/s=Hz) and found that when subjects were deceptive there was an in- crease in their fundamental frequency. For example, the subjects tested in both of these studies were instructed to lie to some questions and to tell the truth to other ques- tions; the vocal responses of these individuals were measured and an increase of their fundamental frequency occurred when the individual's lied as opposed to telling the truth. The present study differed from these two studies in that the subjects tested chose of their own volition to be truthful or deceptive as opposed to being given instructions to lie or to tell the truth. Secondly, the subjects in the present study were suspected of commit- ting or being involved in committing a crime, therefore, they were under a presumed higher motivation to be truthful or deceptive as opposed to a presumed lower level of stress from less consequential laboratory situations. Finally, this present study differed from Ekman, et al., (1976) and Streeter, et al.,(l977) studies by measuring the changes in the fundamental frequency as a result of the control ques- tion theory. 56 Given these circumstances in which the present study was conducted the findings confirm and extend the knowledge gained through the Ekman, et al.,(l976) and Streeter, et al.,(l977) studies. That is, the fundamental frequency of the human voice is not only useful in detecting deception within the laboratory, but also in actual field situations using the control question theory. Like the present study, Alpert, et al.,(l963) and Friedhoff, et al.,(l962) examined the vocal intensity (amplitude, core: related to intensity and measured in dBSPL) of the human voice. In their studies, they found that even when a lie was condoned by the experimenter the vocal stress was suffi— cient to produce measureable changes. In the present study, the changes in amplitude did not reliably distinguish be- tween people who lied or told the truth. Perhaps the use of control question testing prevented significant changes in amplitude for depicting truth and deception. This blending of relevant question-amplitude response and control ques- tion-amplitude response perhaps dampened any significant measurable changes. Using criminal suspects in real-life circumstances may lead to very minute changes in vocal amp- litude (intensity) between truthful and deceptive utterances because the amplitude may become static or establish its own equilibrium; whereas in laboratory subjects the changes in 57 amplitude (intensity) may be more pronounced due to less consequential laboratory situations. The Streeter, et al.,(l977) study employing listeners to detect deception, indicated that the listeners perceptual judgements were based upon more than just pitch changes and the other acoustic cues may have been articulation rates (e.g. syllabic duration, prolongation of a vocal utterance, measured in ms). Lehiste and Peterson (1959) and Hall and Hutchinson (1976) both found.increases in syllabic duration due to psychological stress. Therefore, this third vocal parameter (syllabic duration) was examined to determine its usefulness in detecting deception. The findings of the present study do not support the use of the syllabic dura- tion as a means of detecting deception using the control question theory. However, the findings did show that the syllabic duration for deceptive subjects was significantly greater than the syllabic duration for truthful subjects, irrespective of the type of questions (control—relevant). This main effect was not anticipated, however the consis- tency suggests that this vocal parameter (syllabic duration) may hold promise in future research in detecting deception. Using confession-verified test subjects and the control question theory provided an objective means for testing the 58 validity of the human voice as an indicator of deception. Although it was expected to find the same results for non- verified subjects, the inconsistency between the findings of confession-verified subjects and the findings of nonverified subjects is hard to explain. This difference should be an avenue explored in future research. IMPLICATIONS Continued research is needed to establish the validity of the fundamental frequency as an indicator of truth and de- ception. If future research confirms these findings, funda- mental frequency changes may become an added dimension for improving efforts for detecting deception. Forensic science administrators of present-day polygraphers anticipate the reliability of their science increasing with new technology. Current studies in polygraph are under-way with heat response scaling of the GSR and its adaptation to color change; a digital readout from cardiographic responses and speculation towards blood pressure readings to further advance the polygraph technique. Fundamental frequency of the human voice could become another measurement for testing truth and deception in the criminal justice system. APPENDICES APPENDIX A MULTIVARIATE ANALYSIS OF VARIANCE TABLES VERIFIED SUBJECTS ON TEST ONE APPENDIX A Table 20. Multivariate Analysis of Variance for Fundamental Frequency on Verified Subjects, Test One. SOURCE OF F SIG. OF VARIATION df MS RATIO F Main Effect WiEEin Cells 77 1949.74726 Type of Exam 1 11350.00511 5.82127 0.018 Within Cells 77 133.34821 Type of Quest. 1 592.63291 4.44425 0.038 Interaction Type Exam and Type Question 1 1488.55517 11.16292 0.0008 Table 21. Multivariate Analysis of Variance for Amplitude on Verified Subjects, Test One. SOURCE OF F SIG. OF VARIATION df MS RATIO F Main Effect Within Cells 77 20.02570 Type of Exam 1 63.21800 3.15684 0.080 Within Cells 77 1.24931 Type of Quest. 1 .05323 0.04261 0.837 Interaction Type Exam and Type Question 1 .10527 0.08427 0.772 59 64 Table 28. Multivariate Analysis of Variance for Duration on Nonverified Subjects, Test One. SOURCE OF F SIG. OF VARIATION df MS RATIO F Main Effect Within Cells 43 116.82676 Type of Exam 1 90.93829 0.77840 0.383 Within Cells 43 13.63517 Type of Quest. 1 26.67778 1.95654 0.169 Interaction Type Exam and Type Question 1 3.50972 0.25740 0.615 APPENDIX B MULTIVARIATE ANALYSIS OF VARIANCE TABLES VERIFIED SUBJECTS ON TEST TWO APPENDIX B Table 23. Multivariate Analysis of Variance for Fundamental Frequency for Verified Subjects, Test Two. SOURCE OF F SIG. OF VARIATION df MS RATIO F Main Effect Within Cells 77 2134.66838 Type of Exam 1 7801.96529 3.65488 0.060 Within Cells 77 29.51404 Type of Quest. 1 6.89241 0.23353 0.630 Interaction Type Exam and Type Question 1 63.02657 2.13548 0.148 Table 24. Multivariate Analysis of Variance for Amplitude on Verified Subjects, Test Two. SOURCE OF F SIG. OF VARIATION df MS RATIO F Main Effect Within Cells 77 23.56048 Type of Exam 1 45.35224 1.92493 0.169 Within Cells 77 1.09030 Type of Quest. 1 0.00025 0.00023 0.988 Interaction Type Exam and Type Question 1 0.23677 0.21716 0.643 61 62 Table 25. Multivariate Analysis of Variance for Duration on Verified Subjects, Test Two. SOURCE OF F SIG. OF VARIATION df MS RATIO F Main Effect Within Cells 77 177.34996 Type of Exam 1 948.39447 5.34759 0.023 Within Cells 77 26.38857 Type of Quest. 1 2.53165 0.09594 0.758 Interaction Type Exam and Type Question 1 14.54872 0.55133 0.460 APPENDIX D Table 29. Multivariate Analysis of Variance for Fundamental Frequency on Nonverified Subjects, Test Two. SOURCE OF F SIG. OF VARIATION df MS RATIO F Main Effect Within Cells 43 2562.26640 Type of Exam 1 909.50020 0.35496 0.554 Within Cells 43 129.77028 Type of Quest. 1 9.34444 0.07201 0.790 Interaction Type Exam and Type Question 1 796.03353 6.13417 0.017 Table 30. Multivariate Analysis of Variance for Amplitude on Nonverified Subjects, Test Two. SOURCE OF F SIG. OF VARIATION df MS RATIO F Main Effect Within Cells 43 37.33143 Type of Exam 1 24.02857 0.64366 0.427 Within Cells 43 2.45637 Type of Quest. 1 2.08544 0.84899 0.362 Interaction Type Exam and Type Question 1 0.19563 0.07964 0.779 65 66 Table 31. Multivariate Analysis of Variance for Duration on Nonverified Subjects, Test Two. SOURCE OF F SIG. OF VARIATION df MS RATIO F Main Effect Within Cells 43 111.45245 Type of Exam 1 190.94464 1.71324 0.198 Within Cells 43 26.43695 Type of Quest. 1 37.37778 1.41385 0.241 Interaction Type Exam and Type Question 1 8.83353 0.33414 0.566 APPENDIX E MULTIVARIATE ANALYSIS OF VARIANCE TABLES VERIFIED SUBJECTS ON COMBINED TESTS ONE AND TWO APPENDIX E Table 32. Multivariate Analysis of Variance for Fundamental Frequency on Verified Subjects, Combined Tests One and Two. SOURCE OF F SIG. OF VARIATION df MS RATIO F Main Effect Within Cells 78 2064.22915 Type of Exam 1 11377.47621 5.51173 0.021 Within Cells 78 96.20348 Type of Quest. 1 429.02500 4.45956 0.038 Interaction Type Exam and Type Question 1 294.85379 3.06490 0.084 Table 33. Multivariate Analysis of Variance for Amplitude on Verified Subjects, Combined Tests One and Two. SOURCE OF F SIG. OF VARIATION df MS RATIO F Main Effect Within Cells 78 29.49158 Type of Exam 1 23.75264 0.80540 0.372 Within Cells 78 1.85307 Type of Quest. 1 0.92264 0.49790 0.483 Interaction Type Exam and Type Question 1 0.24178 0.13048 0.719 67 68 Table 34. Multivariate Analysis of Variance for Duration on Verified Subjects, Combined Tests One and Two. SOURCE OF F SIG. OF VARIATION df MS RATIO F Main Effect Within Cells 78 166.08155 Type of Exam 1 729.13786 4.39024 0.039 Within Cells 78 21.37721 Type of Quest. 1 0.18906 0.00884 0.925 Interaction Type Exam and Type Question 1 0.01379 0.00065 0.980 APPENDIX F MULTIVARIATE ANALYSIS OF VARIANCE TABLES NONVERIFIED SUBJECTS ON COMBINED TESTS ONE AND TWO APPENDIX F Table 35. Multivariate Analysis of Variance for Fundamental Frequency on Nonverified Subjects, Combined Tests One and Two. SOURCE OF F SIG. OF VARIATION df MS RATIO F Main Effect Within Cells 43 2360.39701 Type of Exam 1 352.0285? 0.14914 0.701 Within Cells 43 38.12355 Type of Quest. 1 0.90000 0.02361 0.879 Interaction Type Exam and Type Question 1 245.78750 6.44713 0.015 Table 36. Multivariate Analysis of Variance for Amplitude on Nonverified Subjects, Combined Tests One and Two. SOURCE OF F SIG. OF VARIATION df MS RATIO F Main Effect Within Cells 43 32.13978 Type of Exam 1 7.07251 0.22005 0.641 Within Cells 43 1.05798 Type of Quest. 1 1.14469 1.08196 0.304 Interaction Type Exam and Type Question 1 1.87843 1.77549 0.190 69 70 Table 37. Multivariate Analysis of Variance for Duration on Nonverified Subjects, Combined Tests One and Two. SOURCE OF F SIG. OF VARIATION df MS RATIO F Main Effect Within Cells 43 101.32980 Type of Exam 1 136.35734 1.34568 0.252 Within Cells 43 11.26052 Type of Quest. 1 31.80278 2.82427 0.100 Interaction Type Exam and Type Question 1 5.86984 0.52128 0.474 APPENDIX G MEANS AND STANDARD DEVIATIONS TABLES VERIFIED AND NONVERIFIED SUBJECTS ON TESTS ONE AND TWO KNOWLEDGE-RELEVANT-CONTROL TYPE QUESTION APPENDIX G Table 38. Fundamental Frequency Means and Standard Deviations for Verified Subjects on Test One/ Knowledge-Relevant Type Question. Type of Type onuestion Examination Knowledge-Relevant Control Truthful Mean --- --- s.d. --- --- Deceptive Mean 115.1 119.7 s.d. 33.3 27.6 Table 39. Fundamental Frequency Means and Standard Deviations for Verified Subjects on Test Two/ Knowledge-Relevant Type Question. Type of Type of Question Examination Knowledge-Relevant Control Truthful Mean --- --- s.d. --- --- Deceptive Mean 123.8 124.3 s.d. 38.5 37.3 Table 40. Fundamental Frequency Means and Standard Deviations for Nonverified Subjects on Test Two/ Knowledge-Relevant Type Question. Type of Type onuestion Examination Knowledge-Relevant Control Truthful Mean 175.0 191.0 s.d. 0.0 0.0 Deceptive Mean --- --- s.d. --- —-- 71 72 Table 41. Amplitude Means and Standard Deviations for Verified Subjects on Test One/Knowledge-Relevant Type Question. Type of Type of Question Examination Knowledge-Relevant Control Truthful Mean --- --- s.d. --- --- Deceptive Mean 25.2 25.7 s.d. 2.5 3.0 Table 42. Amplitude Means and Standard Deviations for Verified Subjects on Test Two/Knowledge-Relevant Type Question. Type of Type of Question Examination Knowledge-Relevant Control Truthful Mean --- --- s.d. --- --- DeceptiVe Mean 25.6 25.4 s.d. 3.1 3.1 Table 43. Amplitude Means and Standard Deviations for Nonverified Subjects on Test Two/Knowledge- Relevant Type Question. Type of Type of Question Examination Knowledge-Relevant Control Truthful Mean 15.3 20.6 s.d. 0.0 0.0 Deceptive Mean --- --- Sod. _-- --- 73 Table 44. Syllabic Duration Means and Standard Deviations for Verified Subjects, on Test One/Knowledge- Relevant Type Question. Type of Type onuestion Examination Knowledge-Relevant Control Truthful Mean --- --- s.d. --- --- Deceptive Mean 44.1 42.0 s.d. 7.2 9.2 Table 45. Syllabic Duration Means and Standard Deviations for Verified Subjects, on Test Two/Knowledge- Relevant Type Question. Type of Type of Question Examination Knowledge-Relevant Control Truthful Mean --- --- s.d. --- --- Deceptive Mean 41.8 40.0 s.d. 5.2 3.9 Table 46. Syllabic Duration Means and Standard Deviations for Nonverified Subjects, on Test Two/Knowledge- Relevant Type Question. Type of Type of Question Examination Knowledge-Relevant Control Truthful Mean 32.0 38.0 s.d. 0.0 0.0 Deceptive Mean --- --- Sod. --- --_ REFERENCES REFERENCES Alpert, M., Kurtzberg, R.L. and Friedhoff, A.J., "Transient Voice Changes Associated with Emotional Stimuli”, Archives of General Psychiatry, Vol.8, No. 4, p.365, April, 1963. Arthur, R.O.,'The G.S.R. Unit”, Journal of Polygraph Studies,Vol. V, No. 6, May-June 1971. Barland, G.H., “The Reliability of Polygraph Chart Evaluations”, Paper presented at the American Polygraph Association Seminar, Chicago, IL, August 15, 1972. Bersh, P., ”A Validation Study of Polygraph Examiner Judgements”, Journal of Applied Psychology, Vol. 53, pp. 399-403, 1969. Bigland, B., and Lippold, O.C.J., "Motor Unit Activity in the Voluntary Contractions of Human Muscle“, Journal Physiology, Vol. 125, pp. 322-335, 1954. Brenner, Malcolm, 'Stagefright and Steven's Law", Paper presented at Eastern Psychological Association Convention, April 1974. Coker, C.H., Umeda, N. and Bowman, C.P., ”Automatic Synthesis from Ordinary English Text,” Speech Synthesis, edited by James L. Flanagan and Lawrence R. Rabiner, Stroudsburg, PA., Dowden, Hutchinson and Ross, Inc., pp. 440-441, 1973. Dektor Counterintelligence and Security, Inc., 5508 Port Royal Road, Springfield, VA., 1977. 74 75 Edel, E. and Jacoby, J., "Examiner Reliability in Polygraph Chart Analysis: Identification of Physiological Responses“, Qpprnal of Applied Psychology, Vol. 60, pp.632-634, 1975. Ekman, Paul, Friesen, Wallace V. and Scherer, Klaus R., “Body Movement and Voice Pitch in Deceptive Interaction", Semiotica, Vol. 16, No. 1, pp.23-27, 1976. Ellson, D.G., Davis, R.C., Saltzman, I.J. and Burke, C.J., ”A Report of Research on Detection of Deception“, (Contract N6 our-18011 with the Office of Naval Research) Available from the University of Indiana, Bloomington, IN, 1952. Friedhoff, Arnold J., Alpert, Murray and Kurtzberg, Richard L., 'An Effect of Emotion on Voice", Nature, Vol. 193, No. 4813, pp.357-358, January 27, 1962. Fry, 0.8., “Experiments in the Perception of Stress”, Language and Speech, Vol. 1, 1958. Hall, M.E. and Hutchinson, J., I'Spectrographic Analysis of Interspeaker and Intraspeaker Variabilities of Speakers Under Stress", Human Communication, Vol. 1, Spring, 1976. Horvath, F. and Reid, J., "The Reliability of Polygraph Examiner Diagnosis of Truth and Deception", The Journal of Criminal Law, Criminology and Police Science, Vol. 62, No. 2, 1971. Horvath, Frank, “Detection of Deception: A Review of Field and Laboratory Procedures and Research“, Polygraph, Vol.5, No. 2, pp. 107-145, June 1976. Horvath, Frank, ”Detecting Deception: The Promise and the Reality of Voice Stress Analysis”, Journal of Forensic Science, JFSCA, Vol.27, No.2, April 1982. pp. 340-351. 76 Horsley, V. and Schafer, E.A., “Experiments on the Character of the Muscular Contractions which are Evoked by Excitation of the Various Parts of the Motor Tract”, Journal Physiology, Vol. 7, pp.96-110, 1886. Jaspar, H.H. and Andrews, H.L., ”Brain Potentials and Voluntary Muscle Activity in Man”, Journal Neurophysiology, Vol. 1, pp. 87-100, 1938. Keeler, L., ”A Method for Detecting Deception", American Journal of Police Science, Vol. 1 (1): 38, 1930. Kreifeldt, J.G. and Yao, S., “A Signal to Noise Investigation of Nonlinear Electromyographic Processors“, IEEE Trans. Bio. Eng., Vol. 21, 1974. Kubis, J., Studies in Lie Detection: Computer Feasibility Considerations. Technical Report 62-305, Arlington, VA, Armed Services Technical Information Agency, June 1962, prepared for Air Force Systems Command, Contract No. AF20 (602) - 2270, project no. 5534, Fordham University, 1962. Larson, J., ”Modification of the Marston Deception Test”, Journal of the American Institute of Criminal Law, Criminology and Police Science, Vol. 12, pp.390-399, 1921. Lee, C.D., "The Instrumental Detection of Deception", Springfield, IL, Thomas, 1953. Lehiste, I. and Peterson, G., "Vowel Amplitude and Phonemic Stress in American English“, Journal of Acoustical Society of America, Vol. 31, pp.428-435, 1959. Lieberman, Philip, ”Some Acoustic Correlates of Word Stress in American English", Journal of the Acoustical Society of America, Vol. 32, No. 4, April, 1959. 77 Lieberman, Philip, ”Perturbations in Vocal Pitch“, Journal of the Acoustical Society of America, Vol. 33, No.5, May, 1961. Lieberman, Philip and Michaels, Sheldon B., "Some Aspects of Fundamental Frequency and Envelope Amplitude as Related to the Emotional Content of Speech", Journal of the Acoustical Society of America, Vol.34, No. 7, July, 1962. Lindsley, D.S., ”The Psychology of Lie Detection”, In G.J. Dudycha (ed.) ngchologyyfor Law Enforcement Officers, Springfield, IL, Thomas, pp. 89-125, 1955. Lippold, O.C.J., Redfearn, J.W.T. and Vuco, J., “The Relation Between the Stretch Reflex and the Electrical and Mechanical Rhythmicity in Human Voluntary Muscle", Journal Physiology, Vol. 138, pp. 14-15, 1957. Lippold, O.C.J., Redfearn, J.W.T. and Vuco, J., "The Rhythmical Activity of Groups of Motor Units in the Voluntary Contraction of Muscle”, Journal Physiology, Vol. 137, PP.473-487, 1957. Lippold, O.C.J., ”Tremor and Oscillation in the Stretch Reflex Arc“, Journal Physiology, Vol. 202, pp. 55-57, 1969. Lippold, O.C.J., “Physiological Tremor", Scientific American, March, 1971. Marshall, J. and Walsh, E.G., "Physiological Tremor“, Journal Neurolggy, Neurosurgery and Psychiatry, Vol. 19, pp. 260-267, 1956. Marston, W.M., "The Lie Detector Test”, New York, Smith, 1938. McGlone, Robert E. and Hollien, Harry, "Partial Analysis of Acoustic Signal of Stressed and Unstressed Speech", Carnahan Crime Countermeasure Conference Proceedingg, pp. 19-21, 1976. 78 Reid, J.E. and Inbau, F.E., Truth and Deception: The Polygraph (”Lie Detector”) Technique, Baltimore, MD., William and Wilkins, 1966. Rubenstein, Leonard, 'Electro-Acoustical Measurement of Vocal Responses to Limited Stress“, Behavioral Research and Therapy, Vol. 4, pp. 135-138, 1966. Schafer, E.A., "On the Rhythm of Muscular Response to Volitional Impulses in Man“, Journal Physiology, Vol. 7, pp. 111-117, 1886. Shipp, T. and McGlone, R., "Physiologic Correlates of Acoustic Correlates of Psychological Stress“, Acoustical Society of America Convention, Los Angeles, 1973. (Also reported in 1975 Carnahan Crime Countermeasure Conference by McGlone, R., "Tests of the Psychological Stress Evaluator (PSE) as a Lie and Stress Detector”). Simonov, P.V. and Frolov, M.V., “Utilization of Human Voice for Estimation of Man's Emotional Stress and State of Attention", Aerospace Medicine, Vol. 44, March, 1973. Streeter, Lynn A., Krause, Robert M., Geller, Vallerie, Olson, Christopher and Apple, William ”Pitch Changes During Attempted Deception", Journal of Personality_and Social Psychology, Vol. 35, No. 5, pp. 345-350, 1977. Thackray, R.T. and Orne, M.T., "A Comparison of Physiological Indices in Detection of Deception", Psychophysiology, Vol. 4, pp. 329-339, 1968. VanBuskirk, C. and Fink, R.A., "Physiologic Tremor, An Experimental Study", Neurology, Vol. 12, pp. 361-370, Minneapolis, 1962. Voice Identification, Inc., P.O. Box 714, Somerville, N.J. 08876, publication on PM Analyzer, “Speech Processing with the PM Pitch Analyzer". 79 Welch, B.L., ”On the Comparison of Several Mean Values: An Alternative approach," Biometrika, Vol. 38, pp.330-336, 1951. Wicklander, D. and Hunter, F., "The Influence of Auxilliary Sources of Information in Polygraph Diagnosis", Journal of Police Sgience and Administration, Vol. 3, pp. 405-409, 1975. Williams, Carl E. and Stevens, Kenneth N., ”Emotion and Speech: Some Acoustical Correlates", Journal of the Acoustical Society of America, Vol.52, No.4, (Part 2), pp.1238-1250, 1972. "IIIIIIIIIIIIT