Parametric influence of speaker verification system — An experimental study

With the advent of servicing industries that needs to cater to a large number of their clients it has become necessary to offer self-help options to their clients. Very often the self-help channel is the interactive voice response (IVR) channel and the mode of interaction is spoken speech. The affect of this is the need for the self-help platform to be able to verify the identity of the user before making the self-help services. In this context speech based biometric has gained significance in recent times. There are several advantages of using speech as a biometric; but there has been a caution in using speech as a biometric in actual practice. This paper is part tutorial and part experimental. We first present a speaker verification system based on statistical modeling and then present experiment results to verify the influence of the train data, speech parameters and other information on the performance of the speaker verification system.