A century ago, a German speech scientist named Ludimar Hermann got his hands on an Edison phonograph and began to study acoustic phenomena he called formants. Half a century later, researchers from Bell Labs used these formants – which (in case you were wondering) are the spectral maximums that result from vocal resonance (ahem) – as the basis for an early speech platform called Audrey that could recognize uttered digits.
In the years since, speech recognition and analytics have evolved into an interdisciplinary field that combines computing, linguistics, mathematics, and even social sciences. During my own decades at Bell Labs, I remember with great fondness the amazing research that emerged from the speech groups. Today, recognition, interpretation, and use of spoken language by computer systems have become an essential aspect of our day-to-day lives.
It was therefore a delight to spend time recently with Clearspeed. Born in 2016 with roots in the Stanford Business School, the company's solution offering, Clearspeed Verbal, combines voice analysis with an AI-enabled Remote Risk Assessment (RRA) technology. As you’d expect, using voice technology for security analytics sits right in the wheelhouse of yours-truly, so I was excited to connect with Clearspeed. Here’s a summary of what I learned:
“We formed Clearspeed to disrupt vetting with fast, accurate, and scalable voice technology,” explained Alex Martin, CEO and Co-Founder. “Unlike lie detectors, which make a binary determination of lie or no-lie, we follow an interdisciplinary process that identifies risk along a continuum and - rather than delivering a determinative outcome - we provide alerts that teams can use in their risk processes. We do this using automated voice questionnaires which analyze the characteristics of human voice that we have determined, through years of testing, to be associated with risk.”
How the Clearspeed Verbal platform accomplishes this goal is fascinating: It starts by using speech technology to collect vocal responses from some person to a series of yes and no questions. It then measures various differences in the structures and features of those audible responses from an algorithmic risk quantification they have validated over many years. Screening processes would presumably use this method to quickly identify persons of potentially high risk, or to rapidly clear the low risk majority.
Now – before you jump to conclusions that this is some turbo-charged lie detection system, let me share what Clearspeed explained when I asked about this: The system is designed to simply collect answers to questions, and analyze these human reaction-based responses on a low-high spectrum. When a response does not match a pre-determined voice signature based on a proprietary risk analysis, then it flags that response as potentially indicating risk.
Here's where the security comes in: Clearspeed’s customers direct employees, candidates, or other individuals to be vetted through the automated questionnaire, which is simply a few minute phone call that can be delivered in any language. The voice output is then used to determine risk for questions that might be deemed pertinent to that business – perhaps related to some sensitive issue or problem. The platform identifies risk often missed by legacy fraud or security platforms, and sorts the group to provide clarity to the human expert charged with making a hiring, claim, or security-related decision.
“We’ve had success with financial services companies using our technology to detect and prevent fraud,” explained Martin. “We’ve also had success supporting military organizations such as United States Special Operations in their vetting of individuals. Organizations such as energy companies and insurance have also put our voice technology to good use in reducing risk. Think of us like the metal detector in the airport. Our technology clears people quickly, and provides alerts missed for further follow-up.”
The Clearspeed team cited a DoD field evaluation where their technology apparently performed at extremely high levels of accuracy. This contrasted with polygraphs and voice stress analyses, which showed accuracies in the 50 to 75% range - levels deemed to low for deployment as a dependable risk identification or screening tool. Avoidance of human bias and the consistent use of AI were cited as major factors in the high Clearspeed results.
From the perspective of TAG Cyber industry analyst, I can confidently conclude that this methodology and platform have high potential for growth. In fact, the Clearspeed commercial solution seems like a winner, because the processes being targeted for disruption are some of the most tedious and error-prone business and government activities that might be imagined (#government, #clearances, #bureaucracy – ugh).
If you are currently manage any business process that requires vetting of humans for purposes ranging from highly consequential applications such as military, nuclear, or industrial control – to more mundane applications such as personnel application processing, then you’d be wise to contact Clearspeed to solicit their story. I suspect you will like what you hear – and please be certain to share your experiences with us here.
Stay safe and healthy.