Pindrop launches voice recognition engine and solution to combine customer experience with security
Pindrop has launched a deep neural network-powered biometric engine, Deep Voice, which provides passive customer voice identification, and powers the new Passport multi-factor solution for call centers. The new capabilities increase the company’s addressable market from $2.5 billion to $7.8 billion, according to an announcement.
Deep Voice provides the passive authentication capabilities of Pindrop Passport, which also integrates “phone-printing” technology which identifies the device the call is calling from, and “tone-printing” behavior analysis technology, which also works passively as users interact with the call center.
Deep voice is the first deep neural network speaker recognition system to work “end-to-end,” being applied directly to audio data, rather than speech characteristics defined by human operators, Pindrop Vice President of Research Matt Garland told Biometric Update in an exclusive interview.
This new approach to AI use enables Deep Voice to perform voice matching more effectively using short utterances than traditional voice biometric systems, according to Garland, which allows Pindrop to extend its anti-fraud technology to a customer experience application. Pindrop’s existing fraud protection customers, who all had either serious fraud problems or exposure, are a ready market.
“What we discovered is that a lot of those businesses also had customer experience initiatives, where they were trying to figure out how to get to know their customers more intimately, how to recognize them earlier, how to improve their experience and at the same time reduce their costs,” Garland says. “This opens up that avenue for us.”
Garland notes that by running in the background from the beginning of the call, Passport removes the need for knowledge-based authentication and other methods that make customer identification a significant pain point in call center customer experience, and one that businesses have previously been unable to avoid.
“They want to know who they’re dealing with as soon as they get on the line, and they spend a fair bit of time establishing that in the call even if they don’t have a significant fraud problem. Typically, they’ll spend up to a minute or more doing that.”
Passport can identify customers before they are connected with a human representative, without requiring them to use a particular phrase or speak for a certain length of time.
“We actually start analyzing that call, in some cases even before it lands at the IVR,” Garland explains. “We have what we call “pre-ring” analysis that occurs to determine from the carrier networks whether the call looks authentic. As the call enters the IVR we begin collecting the background noise, the keypress information, or the speech utterances if it’s a speech-enabled IVR.”
The company says this difference can reduce call times by up to 55 seconds, and call center operations costs by up to $1 per call.
“Our newly patented Deep Voice engine is creating new opportunities for Pindrop as we see voice eating all other interfaces,” said Vijay Balasubramaniyan, CEO and co-founder of Pindrop. “Establishing trust, security and identity is key to unlocking brand loyalty with consumers. With today’s product expansion we are delivering an end-to-end voice identity platform for enterprises looking to reimagine the customer experience as voice becomes the dominant interface of choice.”
Deep Voice’s use of raw voice data departs from the typical voice recognition, according to Garland, in which algorithms are based on “hand-crafted features that are based on a theory of speech production or speech perception.”
Pindrop draws on data from over half a million calls, he says, to fuel the neural network’s text-independent technology.
“When you have that scale of data you can begin to recognize patterns that allow you to improve the accuracy and the effectiveness of a voice biometric system. So we turned the machines loose on the data, so to speak.”
The combination of the resulting algorithms with device identification and behavioral analysis provides protection from enrollment attacks and emerging attack technologies like voice synthesis, which has become sophisticated enough to fool most voice biometric systems, Garland says.
Pindrop believes this combination of security and customer experience from multi-factor passive authentication will greatly expand its potential customer base, and Garland reports that early adopters of Deep Voice and Passport are excited.
As previously reported, Pindrop’s fraud detection and customer authentication capabilities were certified for use with Amazon Connect in November.