According to Michael Steinmann, Director Regional Technology for Nuance Communications, typically there are 50 to 100 calls per 100,000 that have fraudulent intent and with increasing access to personal information stored in social media, identity theft is on the rise.
Nuance (you may have known them as Scan Soft or Dragon Dictate) have been leading the field in using voice biometrics (VB) in call centres, mobile apps and consumer electronics. Michael can’t reveal any of Nuance’s Australian clients, but across the globe, their clients include banks, law enforcement, government agencies, health care and more.
So begins my, and your, quick education into VB, stress and distress detection and disambiguation.
VB is used:
- To verify identity
Traditionally ID verification is all about two or more things - something you know (knowledge factor), something you have (possession factor) and something the user is (biometric factor).
VB is one of the strongest and most convenient of these three, with a few seconds of natural speech removing the need for pin numbers, date of birth, mother’s maiden name and so on. All of which are increasingly easy to get hold of from social media sources and targeted malware stealing pins and passwords.
According to Michael VB is virtually fool proof. I ask about recorded voices being used and he smiles “That’s in the movies. VB identifies immediately if it is a recording or a series of words clipped together (a ‘liveness’ test) and it detects if the characteristics are ‘exactly the same’ as a voice print on file – in fact it expects small variations” he says.
Once a voice print is identified as fraudulent it can be black listed and that is the end of that person’s ability to continue. If any further attempts are made from the same fraudster, they will be automatically identified as being disingenuous and asked to leave the call or refused the transaction. For example Michael recounted a case where a particular fraudster from South Africa kept calling to a local Australian Bank, trying to change their voice and fool the system, but in combination with other factors the system could always identify the fraudulent caller.
- To personalise experiences and help disambiguate (clarify) what you want
“Tell me in few words what you want. OK that is about mobile phone activation – is that correct? I will connect you to the right person who can help you”.
Having had considerable need to use call centres lately I can say that the days of being transferred several times to get to the right person are improving. But as part of this process it is also about using natural language understanding or NLU in conjunction with information like the telephone number you are calling from and any open files or unresolved issues you have all in the pursuit of better service.
When NLU is combined with VB, the ability to identify who you are and what you want becomes even easier. The system can automatically combine information from your request with your user-profile to make more informed decisions and transfer you to the most qualified agent. This allows the business to not only reduce customer effort, but create more personalised experiences through the call centre.
- To determine stress and distress
Once you use a NLU/VB system you have a voice print. So far Nuance has more than 32 million registered voice prints and these each have some 120 different points of analysis. “There are VB technologies that can tell if you are tense, stressed or under duress and escalate the call to a ‘high risk status’ for professional handling.”While this technology isn’t currently deployed, there is potential future use cases where people being coerced into fraud (telephony banking or at an ATM) can be identified.
The drivers of this technology are fraud reduction, increased convenience for the end user and automation (more efficiency). Michael says that every company that invests in NLU/VB finds efficiencies and increases customer satisfaction and retention. For example, since Barclays implemented voice biometrics within its call centre, 93 percent of customers have rated the system at least 9/10 for speed, ease of use and security.
Michael is keen to see VB rolled out more widely to things like emergency calls where it may reduce prank calls and decrease delays in assessing action to be taken. He said that its use, especially in emergency disaster relief would virtually eliminate the same person claiming multiple benefits.
He is interested in expanded uses like gender and possibly age identification, as part of the natural language understanding for communicating with smart devices like TV’s and vending machines, in cars to determine if the driver may be impaired by alcohol, drugs and illness (please start the car) or to control secondary systems - even to talk to a coffee machine that recognises who you are and what you like.
Another application is language identification. Instead of presenting customers with a long list of language options to choose from, they can just ask for what they are looking for in their native tongue, and the system can identify and then respond directly to them. This will save the Customer time and relieve the stress of listening to 30 or more language options to choose from.
“What we see today is the tip of the iceberg, entry level stuff. One day we will replace pins and ID verification with VB and have a place where you can centrally register your VB print to do everything” Michael says. The aim would be that customers can simply say “Go ahead and pay my bill” – and it would be as simple as that.