Voice Identification

History of Voice Identification

Using aural and spectrographic means trained examiners compare one or more suspect (or known source) voices with a questioned or unknown recorded voice to determine identification or elimination of a speaker. In the legal setting, these comparison tasks are referred to as forensic voice identification tasks. Prior to World War II, there was no technology available to support the identification of the voice by ear. Reliability of aural identification of the voice came into question during the famous kidnap-murder of Charles and Anne Morrow Lindbergh's son in 1932. About that same time, Dr. Francis McGehee, a professor of psychology at Johns Hopkins University, reported an experiment to explore how well listeners could identify unfamiliar voices just by listening to them; his initial findings in “aural identification” showed that the error rate for listeners dramatically increased when voices were presented more than one week apart. Long-term memory for voices was thus revealed to be a limiting factor in “voice identification.”

After that initial report, the advent of World War II led to more focused interest in voice identification because of its possible usefulness in the war effort. Technology for that purpose was developed at Bell Laboratories in New York by Dr. Ralph Potter and a team of colleagues. Their work started with the development of an electromechanical acoustic spectrograph in 1941, which led to a device capable of displaying speech visually—a sound spectrograph. Their research showed that “voiceprints” might have practical application, making it possible to identify a speaker with greater certainty than by the aural method. But the war ended before their work was fully developed.

The work of Gray and Kopp was resumed in the early 1960s by Lawrence G. Kersta, an electrical engineer and physicist, also at Bell Laboratories. His interest in voiceprint identification led to a study using 123 male speakers. He reported a correct identification rate of 99 percent for the comparison of isolated words and 98 percent for random words. His research, however, was limited to contemporary samples, and comparison trials were “closed sets” in which examiners knew that a match existed. Nevertheless, Kersta continued his research and eventually opened his own laboratory and forensic service using the voiceprint method until his retirement in 1973.

Kersta's work sparked a great deal of controversy as well as interest from law enforcement, the scientific community, and the courts. In 1966, the Michigan State Police (MSP) began to pursue the practical application of forensic voice identification as developed by Kersta. Under a federal grant the MSP and Dr. Oscar Tosi, a professor and director of the Speech and Hearing Sciences Research Laboratory at Michigan State University, embarked on a three-year research project to determine the validity and reliability of Kersta's voiceprint method. Dr. Tosi's project included the random selection of 250 male speakers from a population of 25,000 English-speaking male students. Twenty-nine examiners with one week of training conducted 34,996 identification attempts under varying conditions. Examiners made visual comparison of spectrographic cues. They were allowed to rate their confidence level regarding the conclusions. The data in the Tosi project are too voluminous to report here, but when only those “almost certain” decisions were statistically considered the error rate for false identification was 2.4 percent and for false elimination, 4.8 percent. Dr. Tosi reported that if voiceprint examiners were allowed to both listen to voices and compare visually speech spectrograms of a minimum of ten words the error rate would be negligible. Several subsequent experiments, using real-life conditions and male and female speakers in a variety of conditions, appeared to support Dr. Tosi's [Page 1693]conclusions. Voiceprint identification, though still somewhat controversial, is now a common police investigative technique.

...