Anything you say can and will be used against you: spectrographic evidence in criminal cases.

• Author: Rafferty, Lisa

1. INTRODUCTION

   In 1967, a court had the first opportunity to pass upon the admissibility of spectrographic evidence.(1) Since then, there has been a swirl of controversy surrounding what some feel is an unreliable, highly subjective test,(2) and others feel is an accurate means of identifying unknown speakers.(3) This Note addresses the issue of spectrograph admissibility in the post-Daubert era. Part II relates the history and the mechanics of the spectrograph. Part III looks at how various jurisdictions dealt with the issue of spectrograph evidence under the Frye test in the pre-Daubert era. Part IV examines how the Daubert decision has changed the way federal courts must analyze scientific evidence in order to determine its admissibility. And Part V analyzes how the Daubert test will affect the admissibility of spectrographic evidence. Because the Federal Rules of Evidence favor admissibility, courts should allow spectrographic evidence to be used in criminal cases. However, because of the drawbacks of applying the standard set forth in Daubert to a forensic science such as spectography, courts should cease to use the Daubert test when passing upon the admissibility of spectography. Instead, courts should presumptively admit the evidence, and proceed to conduct a thorough review of the reliability of the particular spectrographic comparison at issue.

2. THE HISTORY OF THE SPECTROGRAPH

   The spectrograph(4) was invented by Dr. Lawrence G. Kersta in 1941.(5) Kersta was working as an engineer conducting communications research for Bell Telephone Laboratories.(6) The spectrograph's original purpose was to study speech and other sound signals as they relate to telephone communication services.(7) The spectrograph takes an aural sample and translates it onto paper for visual analysis. A bar spectrograph, the type used today, measures sound according to three dimensions: frequency, time, and intensity.(8) In general, time is represented along the horizontal axis, frequency is represented along the vertical axis, and intensity will be seen as a gray scale.(9) The spectrogram, then, is a graphic representation of sound.(10)

   The spectrograph is premised on the idea that each person has a unique voice and that this uniqueness can be represented accurately by the spectrograph.(11) Kersta hypothesized that each person has slightly different voice cavities and articulators.(12) These include the nose, the throat, and the two oral cavities formed in the mouth by the tongue.(13) Kersta also hypothesized that the way in which the articulators are used during speech will have a substantial role in determining the uniqueness of a person's voice.(14) Kersta believed that "the chance that two individuals would have the same dynamic use patterns for their articulators would ... be remote,"(15) and thus concluded that each person would have a unique spectrograph for different sounds.(16)

   Spectography analysis is, in essence, a matching of similarities, and a distinguishing of differences, between the spectrograms. In the ideal situation, the two recordings would be made under substantially similar circumstances so as to eliminate as many variables as possible.(17) Along with the visual comparison of the spectrograms, an examiner will also do an aural comparison of the voice samples.(18) The examiner can come to several conclusions: positive identification; probable identification; positive elimination; probable elimination; or no opinion.(19)

   Kersta performed several tests using spectrogram comparison. In 1962, Kersta claimed that he had performed over 50,000 tests which resulted in over 99% accuracy.(20) One experiment used a group of high school females as testers.(21) The students received one week of training and then performed a series of "closed"(22) tests.(23) Kersta claimed that of 2,000 tests done by the high school girls, the error rate was approximately one percent.(24) Kersta also performed a similar test using as subjects fraternal twins of the same sex.(25) In this case, Kersta reported an accuracy of 87%; 84% for female voices and 90% for male voices.(26) Kersta repeated the test using an examiner who had eight months of experience in spectrogram analysis, and Kersta reported that this technician was able to identify sixty identical twins with just one error.(27)

   Many scientists, however, doubted the accuracy of Kersta's experiments.(28) They challenged either the hypothesis that each person's voice was unique, or the results obtained by Kersta in his experiments. One of the main critics was Richard H. Bolt. While Kersta claimed that spectrograph analysis was as accurate as fingerprint analysis, Bolt and his colleagues pointed out that a fingerprint is primary evidence representing the exact physical characteristics of the finger itself.(29) However, a spectrogram is a graphic representation of sound, and thus has a less direct relationship to the physical characteristics of vocal anatomy.(30) Furthermore, a fingerprint is static and will not change its physical characteristics over time. A person's voice, on the other hand, will evolve and can be affected by environmental conditions or even excitement level.(31) Finally, a person has the capability to disguise or alter his voice deliberately, while a fingerprint cannot be altered.(32)

   After Kersta's initial tests, K.N. Stevens conducted a similar test using untrained examiners.(33) Steven conducted both open and closed tests. His results did not approach those obtained in the Kersta tests. Stevens had his examiners perform both visual and aural comparisons, but these comparisons were done separately. In the closed tests, Stevens reported that visual examination of the spectrograms in the closed tests yielded a false identification rate of 6% to 28%.(34) In the open test, visual examination resulted in a false identification rate of 31% to 47%.(35) For aural comparisons, however, Stevens reported 6% to 8% error of false identification.(36) This study has been criticized extensively, however. Oscar Tosi, Professor of Speech and Heating Sciences at Michigan State University, pointed out that none of the examiners had any training in spectrogram comparison.(37) Furthermore, each student examiner took only an average of one minute before reaching a conclusion.(38)

   Tosi had the opportunity to conduct his own study in 1970.(39) His study is presently the most comprehensive analysis of the accuracy of spectrogram comparison. Tosi's examiners were college students who all had one month of training in spectrogram analysis.(40) The students made only a visual comparison of the spectrograms and were forced to make either a positive identification or a positive elimination within fifteen minutes.(41) Some trials involved as many as forty known voice spectrograms in both open and closed tests.(42) An important aspect of Tosi's study was that speech samples were obtained by various means and under various circumstances. Among the different types of recordings were: direct speech into a tape recorder in a quiet environment; speech recorded through a telephone line in a quiet environment; and speech recorded through a telephone line in a noisy environment.(43) Tosi recorded 11,644 trials and structured the comparisons to include open tests, telephone transmission, non-contemporary samples, and fixed or random context in an attempt to "closely simulat[e] practical legal cases."(44)

   The results of Tosi's study, while not approaching the accuracy claimed by Kersta, were much more optimistic than those of the Stevens study. Tosi reported a false identification rate of 6% and a false elimination rate of 13%.(45) Furthermore, based on the examiner's rating of confidence in their conclusions, Tosi hypothesized that the error rate would have been even smaller had the examiners been able to return a conclusion of "no opinion." Had that been the case, Tosi estimated that the error rate would have been 2% for false identifications and 5% for false eliminations.(46)

   In subsequent studies, Tosi reported similar results. Tosi had examiners compare long samples (thirty-two seconds) of speech of both American and Polish subjects.(47) The subjects recorded samples under three different conditions: normal voice; under "stress"; and attempting to disguise their voices.(48) Examiners were able to make a positive identification for 96% of the American subjects and for 94% of the Polish subjects when speaking in their normal voice.(49) When speaking under stress or with a disguised voice, the correct identification decreased to 92%.(50) Although there continued to be concern among scientists about the effect of background noise or disguised voices, Tosi asserted that these variables would only increase the number of false eliminations or no opinions. They would not increase the chance that the wrong person would be improperly identified.(51)

   Because of the tremendous law enforcement potential of the spectrograph, the Federal Bureau of Investigation invited the National Academy of Sciences to evaluate the procedure in 1976.(52) The Governing Board of the National Research Council(53) approved the project and appointed the Committee on Evaluation of Sound Spectrograms in July 1976.(54) Richard H. Bolt was appointed chairman, and the committee included Professor Tosi.(55) The Committee performed no independent experiments and its activities were limited to informal discussions. The Committee did not come to any conclusion as to the admissibility of spectrograph evidence in court. Instead, the Committee made broad generalizations about the current sophistication of the spectrograph:

   At the present time, the technique of voice identification is a practical methodology that is rather widely used, but that lacks a solid theoretical basis of answers to scientific questions concerning the foundations of voice identification. This disparity between practice and theory appears to be recognized by...