PERFORMING ACCURATE SPEAKER RECOGNITION BY USE OF SVM AND CEPSTRAL FEATURES
Zülfikar Aslan, Mehmet Akın · DergiPark (Istanbul University) · 2019
The task of performing speakerrecognition over voice recordings is an active research area in the relevantliterature in which many applications has been proposed so far. In this study, speaker recognition isperformed over cepstral features extracted from raw voice recordings. Some ofthe most prominent cepstral feature selection methods, namely, LPC, LPCC, MFCC, PLP andRASTA-PLP are utilized and their contribution to the performance of theapplied method is investigated. Obtainedfeatures are handled by SVM classification algorithm to finalize the speakerrecognition task. As a result, it is observedthat cepstral feature selection methods suchas LPCC and MFCC combined with SVM classification resultin around 97% accuracy.