This study presents systems submitted by the University of Texas at Dallas,Center for Robust Speech Systems (UTD-CRSS) to the MGB-3 Arabic DialectIdentification (ADI) subtask. This task is defined to discriminate between fivedialects of Arabic, including Egyptian, Gulf, Levantine, North African, andModern Standard Arabic. We develop multiple single systems with differentfront-end representations and back-end classifiers. At the front-end level,feature extraction methods such as Mel-frequency cepstral coefficients (MFCCs)and two types of bottleneck features (BNF) are studied for an i-Vectorframework. As for the back-end level, Gaussian back-end (GB), and GenerativeAdversarial Networks (GANs) classifiers are applied alternately. The bestsubmission (contrastive) is achieved for the ADI subtask with an accuracy of76.94% by augmenting the randomly chosen part of the development dataset.Further, with a post evaluation correction in the submitted system, finalaccuracy is increased to 79.76%, which represents the best performance achievedso far for the challenge on the test dataset.
展开▼
