Speaker Recognition Based on Semantic Indexing

Abstract

In this paper, a new pitch extraction method is established to be employed inimproving the performance of the eigenvoices problem. This required indexing thepitch of the voice in a document matrix and then mapping the voice documents intopreserved semantic features. The proposed voice recognition system was built to beoperated in two phases; enrollment and recognition. Closed dataset of differentvoices belong to different sexes and ages of speakers were enrolled in the firstphase. The results of the recognition phase were promising of about 81% for bothsexes. This ensures the successful recognition task and the efficiency of theproposed system.