AUTOMATIC DIALECT IDENTIFICATION SYSTEM BASED CORE-SETS

摘要

Automatic language identification has increasing importance among speech processing applications. It can be used to route calls to human operators (commerce, emergency), preselect suitable speech recognition system (information systems) and has many uses in security applications. In this paper, we focus on automatic dialect identification system based Gaussian Mixture Model (GMM), which measure acoustic characteristics over shifted delta cepstral.The concept of this system is based Multi-Class L2-Support Vector Machines (L2-SVMs) and Minimal Enclosing Ball (MEB) equivalence reduced in Core-Sets approach usingone of the two clustering algorithmsNearest Neighboror Fuzzy C-Mean.The results reveal a significant impact on the identification of five Arabic Maghrebian dialects based on our own corpus. The system achieved an identification rate about 76.09 % on 10-second utterances.