Breaking down voice biometrics
24 February, 2014
category: Biometrics, Corporate, Financial
It is a common misconception that voice biometrics and voice recognition are one in the same. Recognition, however, lacks the biometric component. This confusion and the fragmented terminology used to characterize the technology are responsible for at least part of the apprehension that surrounds voice biometrics.
“Voice recognition, to me, means talk and type – I say a sentence and the computer types that sentence,” says Walter Hamilton, executive director and vice chair of the International Biometrics & Identification Association. “The computer doesn’t know who I am, it just knows what I’m saying.”
Hamilton is among those who see promise in voice biometrics noting that if the pros and cons of voice can be navigated effectively, it could have a very big future.
Voice recognition is a common component in everything from PCs to smart phones. It is the addition of a voiceprint – an enrolled voice sample to be used as a verification factor – that gives rise to voice biometrics.
“Voice biometrics, more formally knows as speaker recognition, answers the question ‘who is speaking?’ rather than ‘what they’re saying?’” says Hamilton. “Voice biometrics is not voice or speech recognition, it is figuring out whether the person’s utterance matches their known, registered voiceprint.”
Voiceprints are the biometric reference that an individual can use to verbally claim their identity by matching against a known sample for that identity.
“It is a one-to-one verification,” explains Hamilton, “typically used for access to a computer, physical access to a building or to conduct a telephonic transaction with a financial institution.”
But voiceprints can also be used for surveillance.
“The other implementation of voice biometrics involves no presumption or claim of identity,” says Hamilton. “Instead, this scenario uses an utterance and attempts to find a match from a previously unregistered voiceprint.”
Currently used in law enforcement and military intelligence, Hamilton explains that this implementation of voice can be used to monitor electronic communications and determine if the captured voices belong to wanted criminals or those on a watch list.
The more commercially understood use of voice biometrics uses telephony. A familiar scenario is calling a financial institution and speaking a passphrase or account number to authenticate.
“This requires an enrollment process that establishes a voiceprint for each user and can then be used to verify identity at the time of transaction,” explains Hamilton. “The matching of an utterance and passphrase takes place on the bank side.”
Nuance’s subtle touch
Exemplifying voice biometrics’ rise in popularity is Nuance Communications. The number of individuals using the company’s voice solutions has jumped from 10 million to 32 million in the past two years and more than 300 organizations use Nuance voice biometric solutions today. This roster includes enterprises like Barclays, TD, T-Mobile and Vodafone as well as a number of government organizations.
The increase in number is due to voice biometrics being deployed in smart phone apps and away from traditional telephony, says Bretislav Beranek, senior principal solutions marketing manager at Nuance.
Nuance maintains three distinct biometric algorithms, each offering a different user experience. The company’s text-dependent voice biometric is used to authenticate an individual with a predefined passphrase.
The passphrase can be common to all users in a system or unique to each individual, e.g. a phone number or account number, Beranek explains. “To create a voiceprint, the user is asked to speak their passphrase three times and for verification, the user is asked to speak the passphrase just once.”
The second of Nuance’s algorithms is text-prompted and is used to authenticate an individual with a random passphrase. “For example, the user could be asked to speak a random set of digits, a set of letters, a set of words or a combination of any of these,” explains Beranek.
Hamilton believes that the third of Nuance’s algorithms is the one that is likely to be most attractive from a user’s perspective. The company’s text-independent algorithm is used to authenticate an individual during a conversation, eliminating the need for a canned passphrase.
“Once you establish voice communication with your financial advisor, guess what, they have a voiceprint from your previous call in which a high level of confidence was already established,” explains Hamilton. “Now you can engage in casual conversation with your financial advisor who is flagged on the computer screen as to whether there is high or low confidence in the identity claimed on the call.”
Dubbed FreeSpeech, it’s a solution that Beranek feels is as fast as it is effective.
“FreeSpeech delivers authentication capabilities direct to the contact center during a live conversation between a caller and an agent,” explains Beranek. “It analyzes the caller’s voice to determine their identity, requiring just 10 to 20 seconds of speech to accurately identity an individual.”
FreeSpeech improves the experience for both the user and enterprise. On the enterprise side, Nuance’s voice biometric solutions promise to deliver operational savings, increased automation rates and fewer cases of fraud. It’s the user experience, however, that is likely to impress.
“The user can authenticate to systems without the need to remember a password, PIN or security question and they aren’t required to carry a token or card,” explains Beranek.
As Beranek explains, Nuance is using voice biometrics to analyze more than 100 unique characteristics of a person’s voice, including physical characteristics such as the shape and size of their vocal tract as well as behavioral characteristics such as rhythm of speech and accent. “The combination of these characteristics make every human’s voice unique, much like the unique ridges of a fingerprint,” he explains.