LVA: The New Fraud Detector?
Herman Cain was innocent of charges that he had affairs, while Jerry Sandusky was not being truthful about his involvement with boys, according to two Layered Voice Analysis (LVA) technology experts.
Lynn Robbins, president of Voice Analysis Technologies of Madison, Wisc., and T.J. Ward, president and CEO of Atlanta-based Investigative Consultants International, made these startling claims in November 2011 after the two statements made by Cain and Sandusky in recent media interviews were analyzed using LVA technology created by Nemesysco Ltd., based in Netanya, Israel.
Nemesysco explains that LVA is not a voice stress analysis technology but instead uses its own patented technology to detect brain activity traces in a subject's voice via a wide spectrum analysis. LVA, Nemesysco says, is comprised of a set of unique signal processing algorithms that identify different types of stress, cognitive processes, and emotional reactions.
Ward says that he used the $15,000 LVA program in assessing Cain's statements made to the press regarding Sharon Bialek, who said Cain sexually assaulted her, and Ginger White, his alleged mistress, who said she had a 13-year affair with the former presidential hopeful.
"When we ran his speech and everything he [said] about not remembering knowing Bialek, he was telling the truth—there was no 'high risk' [or false statement]," Ward says. "When I ran hers, she was telling the truth about their meetings, but as far as his hand up her dress, it wasn't true. It registered as a 'high risk.' With Ginger White, Cain admitted a relationship, but she was going beyond what he was saying about a personal relationship," Ward says.
On the other hand, Sandusky was not telling the truth about allegedly molesting young boys, according to Robbins, who analyzed the former Penn State assistant coach's interview with Bob Costas and found that Sandusky made many high-risk statements.
"LVA is a technology used as an investigative focus tool," Robbins said in an interview. "I don't like saying someone is flat-out lying, but there was deception. There was barely any emotional attachment in his statements. He only showed any kind of emotional attachment to his statements two times, when he said, 'I love to be around kids,' and 'I'm not a pedophile.'
"The majority of his conversation showed embarrassment having to do with people's perception of him, but not guilt. It's pretty abnormal for someone to go through this without guilt, even if they didn't do something.
"He said he worked with many, many young people, but the LVA showed he's holding something back; he's afraid of it. He had no stress, no guilt, but indicators showed high excitement. He really believes in his mind that he was trying to make them feel good about themselves," Robbins says.
What Is Layered Voice Analysis?
Layered Voice Analysis (LVA) technology was invented in Israel in 1997, originally for security purposes. Amir Liberman, founder and CEO of Nemesysco, came up with the idea after a terror event in Israel in 1996.
"I was so upset that it couldn't be stopped," he says. "I wanted to see how I could assist using my knowledge, and led a team of researchers and psychologists to develop LVA."
Nemesysco provides voice analysis solutions for both defense and civilian markets and, based on emotion detection through voice analysis, can optimize truth verification, expose malicious intentions, provide risk assessments, and reveal emotions.
The technology is based on the idea that changes in cortical perception in the brain and interpretation of events manifest themselves in vocal wave form during speech. By using a wide range spectrum analysis to detect minute, involuntary changes in speech wave form, LVA can detect anomalies in brain activity and classify them in terms of stress, excitement, deception, and varying emotional states.
LVA has two basic formulas comprised of unique signal-processing algorithms that extract 129 emotional parameters from each voice segment. These are further classified into nine major categories of basic emotions. Depending on the goal of the analysis, as many as eight "final analysis" formulas can be applied to the emotional parameter data. These include lie stress analysis, arousal level, attention level, emotional level, conflict level, deception patterns match, and additional methods for veracity assessment.
Who Uses LVA?
Nemesysco has worldwide clients and roughly 100 resellers in the United States, Middle East, Europe, Asia, and Russia. In the United States, the technology is sold primarily through Voice Analysis Technologies, founded in 2007.
Most of Robbins's clients are in law enforcement, the military, insurance, prisons, customs and borders, and theft prevention. She has approximately 200 customers, which include the Department of Defense, the U.S. Immigration and Customs Enforcement Department, U.S. Special Forces, the HIDTA (High Intensity Drug Trafficking Areas) program, and several prisons and sheriff's offices. In addition, a number of commercial and corporate entities use LVA for consumer research, uncovering fraud in insurance, call centers, and government benefits such as Medicare/ Medicaid, pre-employment screening in human resources, and credit risk assessment by financial institutions.
Business Use of LVA
But LVA has much more wide-ranging applications than fraud detection. Nemesysco offers several products, including LVA-I, RA7, QA5, and LVA 6.50.
The company's QA5 software development kit provides voice analysis for automated quality assurance and customer relationship management. The QA5 technology engine integrates into existing call center solutions to analyze prerecorded and/or ongoing calls and alerts managers when problematic call events occur.
Additionally, QA5 can be used on archived calls to identify patterns of interaction within calls and evaluate specific aspects of customer service quality, as well as provide insights about customer preferences, likes, and dislikes. QA5 outputs can also be used for automated report generation, staff and shifts optimization, and automated scoring systems.
According to Liberman, statistics show that 3 percent to 5 percent of all calls to contact centers are "bad," meaning that they show increasing customer anger and dissatisfaction. "Using QA5, managers can identify problematic calls as they develop in real time," he says. "Very few bad calls will go under the radar—we have a very low false-negative rate."
Liberman says that costs for LVA depend on volume of tests per year, and are also based on the number of seats in the call center.
Nemesysco's RA7 product is designed for insurance and financial fraud detection. The open architect platform integrates into the contact center through a set of conversation scripts, while analyzing callers' voices for emotional indications of risk. RA7 automatically generates a report per case to indicate its final risk level and suggest topics for further investigation. Managers and investigators can review reports, listen to recorded interviews, and make decisions faster, the company says.
For example, a car insurance call center operator may ask seemingly innocuous questions, such as, "Did you leave your keys in the car?" when a client reports a stolen car.
The company says that RA7 can identify a person's state of mind by analyzing key vocal properties in his speech; identify various types of stress, cognitive processes, and emotional reactions; create an "emotional signature" of an individual's speech at a given moment; and detect deceptive motivation, criminal intention, and general credibility by identifying key emotional signatures reflected in the voice.
Identifying insurance fraud in obtaining government benefits is particularly problematic, according to Robbins, who is working with a large company in Washington, D.C., that is investigating Medicaid/Medicare fraud.
"We provide their investigators with the investigation focus tool to home in on suspicious claims," she says. "It also helps to clear these cases quickly if the solution is found. That is what the LVA is best at, finding answers through effective questioning. There are billions spent on insurance fraud a year, especially in Medicaid and Medicare fraud. We always use a conservative twenty percent savings. The fact is—in almost every case we have been involved with—if there is fraud, we find it. The figure is far higher than twenty percent."
Liberman says Nemesysco's fraud detection rate depends on its clients, as well.
"Since fraud level is around thirty to thirty-five percent in the insurance world and we can identify at least most of it,we have seen cost reduction of around these numbers with some of our customers," he says. "These savings are not created by the system alone, as the system will never make a final decision about any case, but it makes it possible, using a proper and focused investigation protocol, to reveal the false or inflated claims and bring them to their rightful size."
Nemesysco also puts its money where its mouth is.
"We have a return of investment guarantee of at least 200 percent profit on every dollar spent with us," Liberman says. "If you do not meet the projection savings, we pay back the proportional part."
Criticism of LVA
Some critics confuse LVA with lie detectors and voice stress analysis, Liberman says. LVA is not the same as voice stress analysis technology, he says, and does not use any previously known method for detecting voice stress, such as micro tremors.
When he first began to develop LVA, Liberman says, the idea was to build the ultimate lie detector. "What we found out later is that there is no such thing."
Liberman says that since lying is not associated with a single unified set of feelings that are measurable, a real lie detector does not exist.
While Robbins says that his technology is 95 percent accurate, there are plenty of naysayers.
In a 2007 article in the International Journal of Speech Language and the Law, two Swedish linguists, Anders Eriksson and Francisco Lacerda, were highly critical of LVA.
"Our review of scientific studies will show that these machines perform at chance level when tested for reliability," the authors wrote. "Given such results and the absence of scientific support for the underlying principles, it is justified to view the use of these machines as charlatanry, and we argue that there are serious ethical and security reasons to demand that responsible authorities and institutions should not get involved in such practices."
When asked about the Cain LVA analysis, Eriksson told USA Today, "As we show in our paper, the principles upon which the LVA is based have no more to do with emotion or deception detection than throwing a pair of dice. So the people investigating Mr. Cain would have obtained equally reliable results by doing precisely that."
However, a 2006 scientific paper from the University of Florida for the Counterintelligence Field Activity, a United States Department of Defense agency, that was also critical of LVA has since been discredited by journal referees at J. Mack Robinson College, Georgia State University. "The work it reports contains a number of serious scientific flaws; as a result, the claims made in the manuscript are not adequately supported by the evidence presented. This work does not merit publication in its current form," said the referees.
According to Liberman, the Eriksson and Lacerda criticism was unfair and the authors never contacted the company. "I would prefer hearing from someone who actually used it," he says. "Eriksson and Lacerda never laid hands on the actual professional LVA products. Instead, they just reconstructed it based on two of our patents published ten years ago."
Robbins also points out that LVA technology should not be confused with a lie detector. "I understand constructive criticism if a person is educated on any technology. I don't understand criticism if they criticize out of ignorance or generalize their criticism because 'all lie detectors are ineffective.' Some criticism comes when the objective is to better [lie detectors'] chance of existing. We have long been criticized by the polygraph organizations and their embedded supporters. I understand why they are critical. Studies have long shown polygraph to be an ineffective tool. There are far better tools out there and far less expensive [ones]. Those who understand the LVA 6.50 understand its value. There are many who just think it is another lie detector. That is far from the truth."
Duke University, for one, believes in LVA technology. Last year, professors at Duke University's Fuqua School of Business published an article in the American Finance Association's Journal of Finance, "Analyzing Speech to Detect Financial Misreporting." Professors William Mayew and Mohan Venkatachalam analyzed the voices of chief executive officers and chief financial officers during 1,647 conference calls from 691 companies between January and December 2007.
"When managers exhibit high negative affect during conference calls, it results in both lower average return-on-assets over the subsequent two quarters and a lower likelihood of meeting or beating the upcoming quarter consensus analyst earnings forecast," Mayew and Venkatachalam wrote.
The authors said that in a laboratory setting, they generated a speech sample of "misreporters" and truth-tellers to provide construct validity for the vocal dissonance marker using automated vocal emotion analysis software based on LVA technology. They also said that the finding lends support for the LVA-based cognitive dissonance measure.
"We examined whether vocal markers of cognitive dissonance are useful for detecting financial misreporting," the authors wrote. "We used speech samples of CEOs during earnings conference calls, and generated vocal dissonance markers using automated vocal emotion analysis software. We [assessed] construct validity for the software-generated dissonance markers by correlating them with four dissonance-from-misreporting proxies obtained in a laboratory setting. We find a positive association between these proxies and vocal dissonance markers generated by the software, suggesting the software's dissonance markers have construct validity.
"Applying the software to a CEO's speech," they continued, "we find that vocal dissonance markers are positively associated with the likelihood of irregularity restatements. Our results provide new evidence on the role of vocal cues in detecting financial misreporting."
The authors did issue a caveat, though.
The Future of LVA
Liberman acknowledges that the adoption of LVA technology has been slow going, but remains positive about the future.
"We have been pushing water up the hill for fifteen years now, and we proved our case little by little despite very noisy opposition at times…. Changing one's expectations of how to use our technology due to already established misconceptions adds a new set of challenges to our task list," he says. "It was an important time for us to learn the strengths as well as the weaknesses of LVA, how to turn its raw analysis into actionable items to bring it to market readiness and how to overcome its limitations. I have no doubt one day it will reach the critical mass and explode, and I think this day is very close indeed. [This] will be a great year for us."
Staff Writer Michele Masterson can be reached at firstname.lastname@example.org.
A window into the future of speech technology.