A need exists for a general stress measure which is nonobtrusive and can be applied in actual operational circumstances without interfering with operator performance. Ideally, the technique should be objective and quantifiable, simple to collect and analyze, reliable across individuals, and valid for specific test situations. Previous research (Lieberman, 1961; Borgen and Goodman, 1976; Brenner, Branscomb, Schwartz, 1978) in the field of voice analysis has indicated that the psychophysiological stress-state of the speaker may manifest itself in the acoustic domain by changes in the frequency modulation (8 to 12 Hz) of the fundamental frequency (50 to 250 Hz) of the voice. This study attempted to determine if such frequency modulation changes can be used to detect the amount of situational stress in the voice while subjects performed a four-choice information processing task at different presentation rates. The subjects consisted of 31 male Naval officers and 11 male civilians, all college graduates between the ages of 24 and 39. At the time of testing, the participants were all enrolled in a graduate degree program. The subjects were divided into Group I—Jet, Group II—Prop, and Group III—Staff. A Response Analysis Tester (RATER) presented a four-choice discrimination task in which the subject was required to match a response key to each of four stimuli (numbers -one, two, three, and four) appearing in a display window. The sequence of stimuli was randomly presented in an automatic-paced mode for nine 1-minute tests. The stimuli presentation rates were set at one symbol per 1.5 seconds, .75 second, and .50 second. Before the first three tests, the subjects were instructed to press the correct key and not verbalize the number. The next three tests, the subjects verbalized the number and did not press the key. The last three tests, the subjects verbalized the number and simultaneously pressed the correct key. At the end of each block of three tests, the subjects estimated self-performance as percent correct and rated stress on a scale of one (no stress) to seven (high stress). The subjects were allowed as many practice sessions as needed to reach a criterion of 90 percent correct responses at the 1.5 second rate. All subjects reached the criterion within three 1-minute sessions. After practice at all rates, a 1-minute pretest was given using a self-paced mode that allowed the subject to control the stimuli by responding as quickly and accurately as possible. The pretest was used as a measure of each individual's baseline responding rate. Also, a self-pacing posttest was given at the end of the nine 1-minute tests. The voice was analyzed by a device manufactured by Dektor, Incorporated, called a Psychological Stress Evaluator (PSE), developed specifically as a deception-detection instrument. Voice signals were initially recorded on magnetic tape, then processed through filtering circuits and displayed on a strip chart for subsequent visual analysis and interpretation. The coded charts were subjectively scored without any knowledge of conditions by two interpreters trained to recognize stress patterns in speech. The subjective scoring criteria was translated into electronic equivalents and automated on a Varian 73 computer for voice pattern recognition analysis. A comparison of scores from the subjective versus the automated outputs was analyzed for extent of concordance. Significant main effects for percent correct responses were obtained for groups, presentation rate, and groups X presentation rate interaction. No significant differences were found in the correct responses of the subjects when the number was verbalized or not verbalized. The Staff/Civilian Group produced significantly fewer correct responses than either the Jet or Prop Groups at the .75 second rate. Preliminary voice stress analyses for 15 subjects show correlations between performance scores and stress ratings. The results are discussed as to the potential application for a nonobtrusive measure of stress in vocal communication systems that require operator workload assessments.
