Speaker Classification using DTW and A Proposed Fuzzy Classifier

Abstract

The speaker classification is considered with how the identity of the speaker is represented as a unique class label. This identity is characterized by the voice features belong to the speaker. The speaker classification has many application related to the security and forensic systems. There are many classification methods that could be used in speaker classification but the such classifier must has the ability to discriminate the between voice feature vectors which overthought there are a small differences. In this work, a proposed fuzzy classifier has been used for speaker classification using fuzzy inner product (FIP) and Mel frequency Cepstral coefficients (MFCC) features. This proposed classifier is evaluated by a comparison with Dynamic Time Warping (DTW) as traditional method. The proposed classifier was more accrued than DTW, since it classify speakers in ELSDSR data set with 90.91% while the accuracy of DTW classifier was 77.27%.