With the advance of voice authentication systems, administrators have discovered that even they are prone to identity theft. However, researchers at Carnegie Mellon University’s Language Technologies Institute (LTI) have developed a way to protect voiceprints in an effort to reduce this theft.
Bhiksha Raj, associate professor of language technologies at Carnegie Mellon, teamed with Manas Pathak, a recent Ph.D. graduate of the LTI, to develop a method of converting a voiceprint into an alphanumeric string that can function as a password.
The team’s efforts have produced a system that draws on multiple ways a person speaks and converts them into hundreds of alphanumeric strings. To authenticate a person, the system completes an initial registration. When the system then compares the strings with the ones it has on file from the registration, it looks for enough matches among the strings. At a certain level of matching, it authenticates the user. The system can also add additional security by assigning a number to a smart phone.
In test situations with standardized speech, the system achieved a 95% accuracy rate.
Raj and Pathak will present their work during a keynote address on September 21 at the Information Security Conference in Passau, Germany.