Voice passphrase variability evaluation for speaker recognition

Abstract

We propose a method of voice passphrase variability calculation. To calculate the variability, we transform an acoustic passphrase into a sequence of formants, then into a multi-dimensional histogram. We then compute a value which characterizes the entropy of the sequence. We provide a computer simulation and conclude that using this variability for passphrase creation (e.g. during the enrollment process) helps to significantly increase the performance of speaker verification and speaker identification systems.