TECHNOLOGY

RRA® (Remote Risk Assessment)

RRA, the underlying technology in Clearspeed Verbal, is our proprietary risk alerting voice analytics software. Highly scalable and effective, RRA can quickly and economically assess an individual’s level of risk by evaluating specific outputs of human voice and speech. Specifically, RRA can detect and quantify risk in “yes” or “no” responses to customer-defined questions pertinent to a theme, problem, or issue.

Our technology captures these vocal “yes” and “no” responses to automated voice questions, creates an integrated model of each response and then evaluates them from low to high risk. There is no “baseline” required for each voice questionnaire. The presence (or absence) of vocal signatures indicative of distinctive risk-associated reactions is detected, scored and the results are provided to customers via our secure enterprise web application, through an API, or any other methods that customers require.

To do this, RRA analyzes known macro-level stylistic vocal changes elicited in the voice production process. Some of these changes are measurable within spectrographic representations of the voice signal, as well as prosodic and other quantifiable voice characteristics, which have been found to be language agnostic.  RRA technology is not based on micro-feature analysis or signal-level measurements such as jitter, shimmer, tremor, wobble etc. that have traditionally been used for voice stress analysis.

Read More...

In its core function as a reliable, cutting-edge fraud and risk identification tool, RRA has never failed to demonstrate its effectiveness in any of the tens of thousands of instances evaluated. It is however an alerting, not determinative, risk tool, and should never be used as a single risk data source to adversely impact any individual.

The technology has been developed and validated over many years using responses with confirmed risk outcomes. During its development, RRA has undergone, and continues to undergo, complex and iterative evaluations including exhaustive field testing. Over time, these intensive assessment efforts have contributed to the technology’s impressive screening metrics (100% sensitivity, 91% specificity) and accuracy (>95%) for first-order risk vetting applications. Also, to date, for supervised customer screens, RRA has consistently demonstrated positive predictive values of at least 94% (i.e., > 94% probability that questionnaire responses predicted as higher risk are confirmed as such).

Underpinning RRA is the incredibly rich know-how in the content and construct of question design combined with deep expertise in understanding how to spectrogramatically view and interpret the various levels of risk that occur in the voice.

RRA analyzes voice data that is collected through Clearspeed’s automated voice risk questionnaire engine. This engine comprises a host of sub technologies for question delivery, call control quality, and other related technologies that then produce the clearest possible voice signal to be used with RRA.

These automated voice questionnaires are deployed on AWS, and can also be integrated with call center systems, installed on premise, or even completely offline for locations where there is limited cell phone or broadband service. It is scalable to an unlimited number of concurrent calls or connects and can be used either actively or passively. If active, the respondent has been asked to take the questionnaire. If passive, the questionnaire is simply embedded within a call center IVR and is part of the call flow.

What RRA Does

RRA assesses the risk of an individual’s Yes/No voice responses to questions with respect to specific themes, issues, or problems. As an example, some customers want to be confident they do not hire candidates who are also working for unfavorable foreign governments since this would pose an unacceptable risk. How would one find out this information ordinarily? It doesn’t exist, but with RRA, you just ask “Are you currently working for a foreign government?” If someone flags high risk, it is alerting that this is something you should be aware of and a recommendation that you follow up. The alert is an indication of risk but does not make a determination in itself because RRA is not a lie detector. RRA is quite accurate but it should not be used as a sole data source under any circumstances.

RRA Is like a Metal Detector

RRA Objectives

Not unlike TSA screening at airports, the objectives of RRA are to:

  • Help quickly clear and fast-track the majority of people
  • Flag for further due diligence anyone assessed as higher risk

How RRA Works

RRA assesses the risk of an individual’s “yes” or “no” voice responses to customer-defined questions. We obtain these responses remotely through the Clearspeed automated voice questionnaire engine:

We then score the responses and provide the results. Sometimes we alternate each of the pertinent questions with a neutral question. This helps balance the voice signal and increase the assessment accuracy. Sometimes we ask all the questions a second time. This has two purposes – to adjust for voice quality experienced during the first iteration and to allow the individual to hear and understand the questions so they are prepared for the second iteration.  We may also have a short transition between the first and second iterations.

  • Automated phone call (no humans, no bias)
  • No personally identifying information about the caller required
  • Call is 2–10 minutes long, depending on the number of questions
  • Without voice stress analysis
  • Questionnaire can be conducted in any language
  • At any time of the day
  • Anywhere in the world
  • Without biometrics

How Accurate is RRA?

Analysis of tens of thousands of questionnaires and results from real-world use cases have demonstrated RRA’s veracity, efficacy, and reliability, in the following ways:

  • Systematic quantification of vocal signals along a continuum of risk from low to high
  • Accuracy in US DoD field evaluation was >95%
  • Impressive screening metrics based on independent field-validation: 100% sensitivity, 91% specificity, 96.9% PPV, 100% NPV (note: Type II error rate may be underestimated)
  • Consistent flagging precision of > 94% in customer deployments based on known outcomes provided by these customers
  • RRA has regularly reduced customers’ existing false negative blind spots by identifying high risk individuals missed by client-based practices and procedures.

RRA Biometrics, Personally Identifiable Information (PII), and Baselines

RRA does not track PII for people who take our voice questionnaires. It makes no difference who is calling, their race, sex, gender, etc, or which specific questions they will be asked, what language they are in, or where they are located. RRA works the same way irrespective.

RRA does not use biometrics, does not baseline anyone, and does not store or use any personal history about anyone to perform its analysis.

RRA is Not Lie Detection or Voice Stress Analysis

Lie Detection is the process of making a final, binary determination of truth, “lie” or “no lie.” Technology used to assist human experts in the one on one application to help make that final determination are true lie detectors.

Voice Stress Analysis (VSA), developed in the 1960s, is a manually administered process that evaluates frequencies associated with the voice below 20Hz (typically 8-14 Hz). VSA is used almost exclusively during face-to-face, specialized interviews involving specific interrogation techniques. The goal is to identify truth and deception and gain admissions and confessions, similar to polygraph testing.

How Do You Know RRA Works?

Our customers have validated our technology by comparing our results to their confirmed results/outcomes from other indisputable means. In each case, we were not privy to the known outcomes ahead of the use of the tech. This has been done in many languages, cultures, and for both commercial and military use cases. We can share these references with you when necessary.

Crafting RRA Questions

Type: The questions you ask someone are based on the risk you’re trying to prevent or mitigate.  Generally they are slightly intimidating because the subject matter is usually serious – whether it’s health safety, security, or fraud, these are serious issues faced by all of us. They should be as direct as possible and not vague in any way.

Number: Our technology has been shown to be most effective at between 1 and 7 pertinent questions.

We believe the most important issue for Legal teams is to ensure that results from using Clearspeed Verbal are not used as the sole source of data to adversely impact anyone. Clearspeed Verbal provides a risk alert and should be used as such – to alert someone that additional follow up may be required and not as a final determination of risk. It is up to Risk Managers, HR, and other human experts to come to a determination of risk for any individual, using additional sources of evidence and interviews.

In our implementations, the Legal team has primarily wanted  to understand the legality of the RRA technology and any liability the organization may face. The questions usually center on whether RRA is lie detection technology and subject to the 1988 Employee Polygraph Protection Act (“EPPA”).

RRA does not fall under the EPPA. The most relevant clause in this Act is the one referring to lie detection and specifically whether the technology is being used to render a diagnostic opinion regarding the honesty or dishonesty of an individual. RRA provides a risk alert and not any diagnosis about honesty – it cannot tell whether someone is lying and is not measuring deception of any kind. That would be the purview of other technologies such as voice stress analysis, an actual polygraph, or through one on one interviews.