Regresar

LPC-based Feature Coefficients for Voice Authentication Tasks

Abstract:

Voice authentication is a promising biometric technique based on extracting important information from the speech signal by means of computing a vector of feature coefficients. Based on that, this paper evaluates the effectiveness of linear predictive coefficients when combined with other simple metrics in voice authentication tasks. Linear predictive coefficients were chosen due to their relatively good performance and their not-so-complicated structures when compared to other similar alternatives. All the feature coefficients have been evaluated through an extensive parameter space study in order to apprehend the main limitations and potentials of voice authentication under different scenarios. For such an evaluation, a classifier based on artificial neural networks has been implemented.