Automatic dialect recognition using feature fusion

In this paper, an automatic dialect recognition system is proposed using feature-level fusion of mel frequency cepstral coefficient (MFCC) and teager energy operator (TEO) based features. Support vector machine (SVM) based classifier is used in the classification phase. The systematic evaluation of the proposed system is performed on Malayalam dialect database, created in studio environment. The database consists of four dialects with 300 speech samples each. While MFCC based system reports an accuracy of 65%, TEO system gives an accuracy of 73.33% The combined system shows an improvement with overall accuracy of 78%. The experiment shows the promise of feature fusion in automatic dialect recognition system.