Using Voice for Unique ID / Aadhaar authentification
By Nikhil Pahwa on July 12th, 2010
A few days ago, while writing about the authentication API from India’s unique identification authority Aadhaar, we suggested that biometric scans should not be a prerequisite for Aadhaar authentication for services, since that would end up limiting the usage of Aadhaar only to physical verification: mobile devices cannot do fingerprint and Iris scans yet. At the same time, Aadhaar authentication would be essential for reducing KYC costs for mobile transactions as well, which Nokia has suggested, is a substantial cost (Read: Nokia’s Mobile Payment Suggestions)
At VAS Asia, Speech solutions company Nuance Communications showcased an implementation that they’ve done, for voice based authentication of rural daily wage laborers for India’s National Rural Employment Guarantee Scheme. Now the problem with the NREGS has been, as was reported by NDTV a few weeks ago, cases of faking identities to collect wages on behalf of people who weren’t enrolled, or even in existence. In conversation with MediaNama, Sunny Rao Managing Director (India & South East Asia) for Nuance explained how they’re authenticating workers.
When a worker registers for a program, they go through a voice enrollment process for authentication: the coordinator asks them to answer a few questions using their voice, creating a voice password. When someone turns up for work, their attendance is marked using a similar voice authentication process, and money is also collected after verification using the voice.
But what about a long term verification, like is required in the UID? The UID’s biometric authentication is using fingerprints or an iris scan; how does one authenticate voice over a long period of time, or when, say, someone has a sore throat – the voice changes. Rao said that they use something called password ageing. There’s a certain time limit, after which the password expires, and users have to authenticate themselves again; they have an operational procedure where reaches out to people once in six months, to keep the voice profile updated. “There are 28 characteristics to your voice, around five of which change when you have a sore throat. 23 remain the same, and we can make a confident determination on that basis. In case of an extreme situation like Laryngitis, we fall back on DTMF authentication.”
But is this foolproof? Rao said that they use methods like liveness detection to check for imposters, check for multi-factor authentication, where they may ask users questions to which answers only the registrant knows. He says a commercial deployment results in a false success of less than 1%: “if we have 1 million people enrolled, it is the percentage of imposter population that is trying to hack into the account. Out of those 1000, it is less than 1% who could on one particular access which we may incorrectly recognize.”
Rao said that the usage of voice authentication is more service dependent, and is a layer that can be added over and alongwith a UID number – perhaps used for rural banking, for remote access or verification.