VoiceVault Achieving Impressive Levels of Accuracy in Real World Implementations
The latest release of VoiceVault’s voice biometric-based speaker verification engine is achieving impressive levels of performance, according to this press release. Based on the performance of a real-world application involving authentication of individuals initiating high-value transactions through a smartphone, the company has found that False Acceptance Rate (FAR) was 0.01%, while the associated (False Reject Rate (FRR) was 5%.
In this “real-world implementation,” voice biometrics were treated as part of a “4-factor security cocktail.” The voice biometric is the “something your are.” Given that a smartphone is in play, the other three factors can be “something you know” (an account number or PIN), the phone is “something you have” and the fourth factor is often an “anti-spoofing” technology, referring techniques or technologies that detect such things as tape recordings or “fake caller ID” designed to fool a contact center’s automated number identification (ANI) detection.
VoiceVault’s achievement is noteworthy. Opus Research is often asked to compare the accuracy of voice biometric-based caller authentication solutions based on the accuracy of the underlying voice biometric engine. “Your mileage may vary” has been our universal answer. It sounds glib, but FARs, FRRs and EERs are both highly variable and (thankfully) tunable. Impressive levels of accuracy are reached “in the lab.” In the real world, implementers find that accuracy is vulnerability to the quality of communications lines, background noise and even the quality of the initial voice sample from which the voiceprint was electronically distilled.
That said, the “sub 1%” false acceptance rate is much sought after, especially in banks where a single imposter can create havoc and give rise to significant losses. False reject rates in the 5% range can be managed by customer care applications without causing callers too much pain. In point of fact, they are not ever fully rejected, instead they are given the chance to “try again” at least once before another factor (usually the answer to challenge questions) are invoked.
In VoiceVault’s chosen industries – healthcare and financial services – customers will seldom, if ever, rely solely on a single factor, like a voiceprint. Multifactor authentication is both a requirement and a best practice. In this instance, greater accuracy of voice-based authentication translates directly to greater efficiency and better customer experience, which is especially crucial for high-value customers in financial services.