Australian company Auraya, already a world leader in voice biometric authentication, hopes to break into the lucrative US market with its recently-launched ArmorVox Speaker Identity System.
ArmorVox is the direct descendent of a system developed by Dr Clive Summerfield in 2005 for Centrelink, the Australian government's social security payments agency, whose call centre receives 33 million enquires a year.
"For most [voice authentication] applications you need an account number and some other item of speech, such as a phrase such as 'my voice is my password'," Summerfield told CSO Online.
"In the Centrelink application they tested people saying their Centrelink number, their name and a passphrase which was one of five secret passphrases that they'd recorded on the system... it's less than 30 seconds of speech."
That was sufficiently accurate to authenticate more than 95 per cent of users, which in turn means that less than 5 per cent needed to be authenticated by a human operator.
The Centrelink experience demonstrated a commercial demand for voice biometrics, but there were technical shortcomings. Auraya was established in 2006 to develop improved technology for this market, with Summerfield as chief executive.
Since then Auraya's technology has been sold in the US, Europe and the Middle East, and is used by some Australian banks and a government office in New Zealand.
Auraya's advances have now been incorporated into the ArmorVox voice biometrics software for Windows and Linux launched at the end of January.
ArmorVox can process 50,000 verifications per hour on an 8-core server. It can be delivered from customer premises equipment or the cloud, or as a hosted authentication service.
A web services API allows ArmorVox to be incorporated into Interactive Voice Response (IVR) and other applications.
"An internet banking application could telephone your mobile phone number and authenticate that it is actually you doing this international transaction," Summerfield said.
At the heart of ArmorVox is Auraya's patented ImpostorMap technology, which significantly reduces the risk of an imposter impersonating a legitimate user's voiceprint — ironically by filling the system's voiceprint database with the voiceprints of imposters.
In most voice biometric systems, the voice sample provided during an authentication attempt is measured in various ways, and those parameters are compared with those of the sample provided by the user when they were enrolled in the system. If they match closely — that is, if the overall difference between the two samples is less than a predetermined threshold value — authentication is successful.
"What you find in any speech database is that around 80 per cent of the voiceprints are quite secure, that is, very hard for imposters to break in. 10 to 20 per cent are more susceptible" Summerfield said.
ImpostorMap measures the relative security of users' voiceprints by trying to break them using a selection of the other speakers.
"For every person enrolled in the system there's at least ten others that are actually claiming to be the same person, i.e. they're saying the same account numbers, they're claiming the same name, and they're claiming the same date of birth," Summerfield said.
"We then set the security thresholds for each of the voiceprints loaded in the system according to the security performance of each of those voiceprints individually." The result is a significant improvement in the system's overall accuracy.
Contact Stilgherrian at Stil@stilgherrian.com or follow him on Twitter at @stilgherrian
Register Today. Hear from Rob Livingstone, Michael Barnes, Steve Quane and Dave Asprey amongst others on the Evolution, Trends, Solutions and the Future of Cloud Security, limited seats register today through CSO.