Kolmogorov-Smirnov Test in Text-Dependent Automatic Speaker Identification

In this paper Kolmogorov-Smirnov test, a non-parametric statistical test, is introduced for text-dependent automatic speaker identification (ASI) with Mel-frequency cepstral coeffients (MFCCs) based speech features. In the case of closed-set ASI, the identity (Id) of the unknown speaker is assigned to the Id of that reference speaker to whom the number of MFCCs pair having same distributions is maxima with 88.26% accuracy at 8% level of significance. In open-set ASI, after determining the identity of the unknown speaker it is verified whether the unknown speaker is truly the reference speaker or not by comparing the number of matched MFCCs pair to the threshold value previously set for that reference speaker with 87.24% identification efficiency at 2% level of significance and false speaker detection rate is 99.5% at 10% level of significance.