VOICEPRINTS

They’re the ‘fingerprints’ of the voice--a system of identification--but they’ve taken their lumps when offered as evidence in criminal trials. At issue: How scientifically reliable are they? Now, law enforcement officials and scientists in Whittier have teamed up in a high-tech effort to validate the method, in hopes of stirring a change in the legal winds.

Twenty years ago, a young man whose face could not be seen but whose voice rang with clarity described during a CBS television documentary how he had torched buildings during the Watts riot three months earlier. Acting on independent information, police arrested an 18-year-old as an arson suspect. When investigators questioned him at County Jail, they taped the interview.

That tape with the suspect’s “voice exemplar” on it and a subpoenaed copy of the CBS recording was sent to the New Jersey laboratory of Lawrence G. Kersta, a physicist who had developed a system of identification that had become popularly know as the voiceprint. Kersta examined both tapes and said the voices matched. He later testified to that effect at the trial of the 18-year-old, who was convicted.

The conviction was overturned, however, when an appellate court ruled that the verdict was based primarily on the voiceprint. The judges said the process had “not yet reached a sufficient level of scientific certainty to be accepted as identification evidence in cases where the life or liberty of a defendant may be at stake.”

Another Reversal

A decade later, the California Supreme Court overturned another conviction when it rejected voiceprint evidence “until the time when there is demonstrated solid scientific approval in support” of it. Since then prosecutors have been leery about the admissibility of voiceprints in California courtrooms, and since 1976 none has been successfully introduced in a criminal case in the state.

But a change may be in the wind because of a meticulously researched high-tech project run by a small team of deputy sheriffs and speech scientists working in a suite of offices in the Sheriff’s Academy near Whittier. It is aimed at scientifically validating the voice identification technique not only for use in the courtroom but as an investigative tool as well.

The success of the project, which is supported by the Justice Department and the Secret Service, depends on the progress made during the next two years by the three-man group of audio and computer experts working under the auspices of the Special Investigations Bureau, which is headquartered at the academy. The undertaking is the most ambitious attempt yet to elevate the voiceprint process into a law enforcement method as scientifically acceptable as fingerprinting has become.

The voiceprint was originated during World War II by audio specialists working for the U.S. Army. The military wanted to identify voices of enemy radio dispatchers. Strategists figured if they could establish that the same voice was broadcasting from different places they could track enemy troop movements.

It was years before the procedure was introduced into the courtroom.

Although voiceprints now have been accepted as evidence in courtrooms of various levels, including federal jurisdiction, in more than 25 states, the process “still is in its infancy compared to fingerprints,” said Hirotaka Nakasone, a member of the sheriff’s team. He likened its state of the art to handwriting analysis.

What sets the Sheriff’s Department’s work apart from most previous voiceprint research principally is the incorporation of highly sophisticated computers. In the past, voice patterns after being recorded were plotted on paper graphs, which were subject to variable interpretations, depending on what the examiner did or did not see.

Laws of Probability

The team involved in the current project has developed computer hardware and software designed to strip away the variables so that examiners, who themselves have become far more sophisticated during the last few years, will be able to satisfy scientific laws of probability in presenting their findings.

The two principals involved in the sheriff’s project appear, at first blush, an improbable pair.

Craig Melvin, 38, is a tall athletic-looking deputy sheriff who joined the department 13 years ago and was a street cop before he joined the special investigations audio unit eight years ago. He was educated as a meteorologist but developed the skills he brings to the current project by switching sciences when he went to work in the acoustical laboratory of Lockheed Aircraft.

Nakasone, also 38, is a slight, bespectacled academician and civilian consultant on the project. A native of Japan and still legally a resident of that country, he has been in the United States for 12 years, during which time he acquired a bachelor’s, a master’s and a Ph.D. in speech science at Michigan State University, then was an assistant professor there in the same discipline.

The two met six years ago in a class at Michigan State and began to talk about developing a voice identification system, according to Nakasone, “that would be workable in a forensic environment” and about how computers could be employed to make it so.

Both were aware that a committee of the National Academy of Sciences had warned as recently as 1979 that voice tape evidence “should be used with great caution” because it “is plagued with technical uncertainties.” But at the same time, the academy committee suggested that the concept “could develop into a mature endeavor built on scientific understanding.”

That is exactly what Melvin, Nakasone and the deputy sheriff who does their computer programming, Mike Clynes, are aiming for.

The National Institute of Justice, which oversees grants offered by the Department of Justice, awarded the Sheriff’s Department $220,000 to undertake the project. The Secret Service contributed $60,000. Two private foundations donated about $25,000 in high-tech equipment being used in the project. And the Sheriff’s Department earmarked $250,000 of its own for the work.

Telephoned Threats

The Secret Service’s interest, Melvin said, is attributable to the fact that “so many phone threats” are made to people highly placed in government. “The Secret Service wants a way of classifying voices,” he explained.

In the next few months, Melvin and Nakasone plan to build a 400-voice data base for their computer analysis system. “Two hundred male and 200 female voices. We’ll probably use a lot of cadets at the academy. And we’ll want them to sound similar to make the computer’s task more difficult,” Melvin said. “If the voices sounded grossly different, it would make it easy for the computer to distinguish the variations.”

Melvin said the speakers will not only talk into a live microphone but their voices will be recorded when they are using a telephone and when they are equipped with a body transmitter. Each volunteer will be tested on two separate occasions to allow for any changes that might occur in his or her voice over a period of time.

During testing, the volunteers will read 10 times from prepared texts and speak spontaneously 10 times, giving the audio experts 40 different samples per voice for their data base.

Advisory Monitors

The National Institute has appointed a four-member advisory panel to monitor the program, including a Lockheed Aircraft acoustical expert, a Georgetown University research psychologist who also has a background in mathematics and judicial philosophy, a lieutenant in the Michigan State Police Department who is a certified voice examiner (as are Melvin and Nakasone), and a Los Alamos, N.M., National Laboratory computer and speech scientist.

Lt. Ben Nottingham, head of the Special Investigations Bureau and the sheriff’s liaison with the project, cited an encouraging example of strides that have been made in preliminary computer voiceprint identification.

Comedian Rich Little, who is gifted at imitating the voices of others, volunteered to participate in an experiment in a lab at the Michigan State Police Headquarters. He did his impressions of Humphrey Bogart and John Wayne, among others.

Despite the entertainer’s skills as an impressionist, Melvin said, when compared against the originals: “Rich Little is Rich Little. And Bogart is Bogart. And John Wayne is John Wayne.”