Biometric Classification of Frequency Following Responses to English Vowels

Auditory evoked potentials are a candidate for the development of future biometric methods with a high security level. As a step towards evaluating the suitability of auditory evoked potentials in biometrics, this work utilizes the frequency following response (FFR) to four short English vowels obtained from 22 normal-hearing adult subjects. Classification algorithms based on Support Vector Machines are used to exploit features extracted from the FFRs, in both time and frequency domains. This approach gave a highest subject identification accuracy of 86.36%. An analysis of the effect of signal quality and stability on the subject identification classification performance was also conducted. The results of the study demonstrate the potential for establishing a biometric identification system using speech-evoked FFRs.