Abstract
The problem of automatic accent identification is important for severalapplications like speaker profiling and recognition as well as for improvingspeech recognition systems. The accented nature of speech can be primarilyattributed to the influence of the speaker's native language on the givenspeech recording. In this paper, we propose a novel accent identificationsystem whose training exploits speech in native languages along with theaccented speech. Specifically, we develop a deep Siamese network-based modelwhich learns the association between accented speech recordings and the nativelanguage speech recordings. The Siamese networks are trained with i-vectorfeatures extracted from the speech recordings using either an unsupervisedGaussian mixture model (GMM) or a supervised deep neural network (DNN) model.We perform several accent identification experiments using the CSLU ForeignAccented English (FAE) corpus. In these experiments, our proposed approachusing deep Siamese networks yield significant relative performance improvementsof 15.4 percent on a 10-class accent identification task, over a baselineDNN-based classification system that uses GMM i-vectors. Furthermore, wepresent a detailed error analysis of the proposed accent identification system.