Speech Technology Magazine


To Tell the Truth

Did you ever want to ask: Does my car really need to have its engine rebuilt? Is the check actually in the mail? Can you really sell me that bridge?<@SM>
By Judith Markowitz - Posted Sep 10, 2002
Page1 of 1
Bookmark and Share
Did you ever want to ask: Does my car really need to have its engine rebuilt? Is the check actually in the mail? Can you really sell me that bridge? Have you ever wanted to have a device that could analyze what someone is saying and automatically tell you whether that person is lying or telling the truth? You’re lucky because you’re reading this column and I happen to know that what you are looking for is a voice lie-detector (VLD). I also know that there are a number of commercial VLD products that are available for purchase. “But,” you ask, “Do they actually work?” Most of the evidence for and against VLDs comes from anecdotal reports including one I did for Speech Technology Magazine in 1998. Once again, though, you are truly lucky. In March of this year, the Air Force Research Laboratory (AFRL/IFE) released the results of a study of two VLDs. Their research was funded by the US National Institute of Justice and its objective was to evaluate the ability of VLDs to support law-enforcement investigations. Under Duress
Unfortunately, the published results provide no definitive answers. The primary reason for the inconclusiveness of the AFRL’s findings is that the evaluators were not able to find a corpus with enough samples of confirmed deceptive speech to do an assessment of the ability of these systems to detect lies. Instead, they elected to determine whether the systems could detect stress – a condition that the researchers knew how to evaluate. They chose to measure stress because developers of VLDs claim that their systems operate on the same basic principles as polygraphs. Both technologies view lying as a stressful condition and contend that stress is manifested in ways that can be measured. Polygraphs detect and measure stress through galvanic skin response whereas voice lie-detectors use acoustic features of a person’s voice. In particular, most developers of VLDs believe that lying produces microtremors: low-frequency signals between the frequencies 20Hz and 40Hz. The AFRL subjected two VLDs to a series of laboratory tests and a field test. The products were Lantern (The Diogenes Group Inc.) and Veridicator (Trustech Ltd.). Consistent with the claims made by The Diogenes Group, Lantern responded to the presence of microtremors but did not respond to any of the other acoustic indicators of stress. Veridicator did not respond to microtremors, which supports Trustech’s claim that Veridicator does not base its decisions on those low-frequency patterns. It was not possible to verify the company’s claim that its product analyzes multiple features because Trustech will not reveal the nature of those attributes saying that the information is proprietary. The only evidence for Trustech’s claim is that Veridicator generates multiple graphs and waveforms during processing. Another test used what the AFRL calls “known-ground-truth” data that consisted of recorded portions of six polygraph sessions with suspects in two separate murder cases. Both Lantern and Veridicator reported the suspects were lying, which was correct since both suspects later confessed and were convicted. The results are suggestive but, according to AFRL, not definitive because the test sample was extremely small (two suspects) and might simply reflect the ability of VLDs to detect the stress any suspect–innocent or guilty–would exhibit during a polygraph test. Each product was subjected to field testing by an experienced investigator working for a local law-enforcement agency. They used the systems in a variety of felony cases, including homicide, rape and burglary. Both investigators were impressed with the ability of the systems to recognize deception and recommended that VLDs be used as an investigative tools. Neither investigator felt that the findings of voice lie-detectors should be used as evidence to help convict a suspect. Corpus delicti
The AFRL concluded that the VLDs they evaluated could, indeed, detect stress in a person’s voice. This finding encouraged them to expand their assessment to determining whether VLDs can also consistently identify deceptive speech. They plan to work with Walter Reed Army Institute of Research to develop a database of deceptive speech. Recognizing that both innocent and guilty suspects are under stress in an interview situation they intend to use the database to determine whether VLDs are able to distinguish deceptive speech from physiological and biochemical stress. Now, about that bridge I had for sale … Dr. Judioth Markowitz is the associate editor of Speech Technology Magazine and is a leading independent analyst in the speech technology and voice biometric fields. She can be reached at (773) 769-9243 or jmarkowitz@pobox.com.
Page1 of 1