Abstract:

The task of performing speaker recognition over voice recordings is an active research area in the relevant literature in which many applications has been proposed so far.  In this study, speaker recognition is performed over cepstral features extracted from raw voice recordings. Some of the most prominent cepstral feature selection methods, namely, LPC, LPCC, MFCC, PLP and RASTA-PLP are utilized and their contribution to the performance of the applied method is investigated. Obtained features are handled by SVM classification algorithm to finalize the speaker recognition task. As a result, it is observed that cepstral feature selection methods such as LPCC and MFCC combined with SVM classification result in around 97% accuracy.

Keywords: