Nearest Neighbor Classification Approach for Bilingual Speaker and Gender Recognition

Abstract

Language plays a very important role in our daily life. Throughout the world various languages are being spoken. Language factor is also important in the speaker recognition technology. In the current work speaker recognition is performed for bilingual speaker and gender recognition. Nearest neighbor classification is a well-known method which analyzed for speaker recognition and gender recognition technology in the current work and also tested the performance of ensemble k-NN approach for speaker recognition problems. The paper discusses about bilingual data-based speaker and gender recognition with multi-handset trained dataset. Well-known mel frequency cepstral coefficient (MFCC) is used to get the glottal characteristics of a speaker. Classification accuracy with individual and mixed utterances for both speaker and gender recognition is achieved maximum with weighted nearest neighbor approach. Overall accuracy of speaker recognition is 83.2% and gender recognition 93.2% is obtained.