Joint-Factor Analysis (JFA) and I-vectors have been shown to be effective for speaker verification and language identification. Channel factor adaptation has also been used for language and accent identification. In this paper, we show how these techniques can be used successfully in the task of accent classification, and we achieve good accuracy on a 14 accent problem using a novel iterative classification framework based on an iterative linear/quadratic classifier. These results compare favourably with recent results obtained using other non-fused acoustic techniques.

Index Terms: Ivector, accent classification, discriminant analysis, confidence measure