A computerized attendant would ask the shopper to repeat numbers and then compare the person's speech to the voiceprint saved. If it matches, the credit card can be billed. But if it doesn't match, no purchase is made.
A person could also potentially buy items over the phone, without the Internet. Retail items in a store or in a catalog would have a Voice Pay product number as well as a phone number listed that the customer could call. The shopper enters the product number and goes through the process of repeating numbers randomly generated by the computerized attendant.
"I would not rely on it exclusively," said Andreas Stolcke, a senior research engineer at SRI in Menlo Park, CA, and a visiting researcher at the International Computer Science Institute in Berkeley.
Stolcke, whose expertise lies in speech and speaker recognition, said that the way sound is produced in a person's vocal chords and then manipulated by the mouth and tongue is difficult to model. Other variables including the recording channel and background noise add to the difficulty of analyzing the sound and then matching it later.
"It's a little like predicting the weather," said Stolcke. "You can measure the state of the weather at a particular instant and create a model of how it develops from one day to the next, but it's an imprecise model, and you can't always predict what will happen the next day."
But not everyone is as skeptical. Voice Pay's website launched on April 30, 2007 and to date has more than 200,000 people signed up to the use the system once it's adopted by banks. That could be sooner rather than later. Ogden said he's already in discussion with 52 banks around the world, one of which could be yours.