Experiment on Voice Identification

Abstract

A two-year experiment on voice identification through visual inspection of spectrograms was performed with the twofold goal of checking Kersta's claims in this matter [Nature 196, 1253[Nature 196, -1257[Nature 196, (1962'] and testing models including variables related to forensic tasks. The 250 speakers used in this experiment were randomly selected from a homogeneous population of 25 000 males speaking general American English, all students at Michigan State University. A total of 34 996 experimental trials of identification were performed by 29 trained examiners. Each trial involved up to 40 known voices, in various conditions: with closed and open trials, contemporary and noncontemporary spectrograms, nine or six clue words spoken in isolation, in a fixed context and in a random context, etc. The examiners were forced to reach a positive decision (identification or elimination) in each instance, taking an average time of 15 minutes. Their decisions were based solely on inspection of spectrograms; listening to the identification by voices was excluded from this experi- ment. The examiners graded their self-confidence in their judgments on a 4-point scale (1 and 2, uncertain; 3 and 4, certain). Results of this experiment confirmed Kersta's experimental data, which involved only closed trials of contemporary spectrograms and clue words spoken in isolation. Experimental trials of this study, correlated with forensic models (open trials, fixed and random contexts, noncontemporary spectrograms), yielded an error of approximately 6% false identifications and approximately 13% false eliminations. The examiners judged approximately 60% of their wrong answers and 20% of their right answers as "uncertain."This suggests that if the examiners had been able to express no opinion when in doubt, only 74% of the total number of tasks would have had a positive answer, with approximately 2% errors of false identification and 5% errors of false elimination. The different conditions existing between experimental trials of identification or elimination performed by the examiners of this study and the tasks performed by a professional examiner in real cases are discussed.

Contributions of Fundamental Frequency and Formant Frequencies to Speaker Identification

“…If, as Tosi et al [1972] suggest, spectrograms are to be used forensically to objectively display similarities among voices, it might be useful to have data bearing on the contributions of these two cues to speaker identifi cation.…”

Effects of selected vocal disguises upon spectrographic speaker identification

“…It has been demonstrated repeatedly that certain speakers are more difficult to identify by their spectrograms than others (Kersta, 1962b;Young and Cambell, 1967;Stevens et al, 1968;Tosi et al, 1972). Hence, it is not surprising that certain speakers in this study were more difficult to identify.…”

“…Subsequently, Tosi et al (1972) attempted to simulate a number of conditions encountered in an actual criminal investigation utilizing speech spectrograms. A wide range of results was obtained in this study, depending on the particular experimental variables being combined.…”

A Tutorial on Text-Independent Speaker Verification

“…In 1972, Tosi et al [70] tried to demonstrate the reliability of voiceprint technique by means of a large-scale study in which they claimed that the scientific community had accepted the method by concluding that "if trained voiceprint examiners use listening and spectrogram they would achieve lower error rates in real forensic conditions than the experimental subjects did on laboratory conditions. "…”