Speaker identification using CVA, SVM, KNN, and TDNN classifiers

Speaker recognition studies are used in many fields today. Especially in security systems, this issue has gained more importance. Speech recognition systems to be created must reach high recognition rates. Speaker recognition is divided into speaker identification and speaker verification. In this study, speaker identification was carried out for the Turkish METUBET and English MNIST databases. MFCC coefficients and pitch frequency values are combined for speaker identification. 40 speakers were used for the METUBET database and 30 speakers were used for the NMNIST database. The CVA, SVM, KNN and TDNN classifiers were used in the study. In speaker identification, the highest speaker identification rate for METUBET was found to be 97.75% with SVM-polynomial kernel and 96.14% with TDNN for MNIST. Speaker recognition result for METUBET was found to be 100% with OVY. Keywords: CVA, KNN, METUBET, MNIST, SVM-polynomial kernel, TDNN