Automatic accent classification using ensemble methods

Accent classification technologies directly influence the performance of the state-of-the-art speech recognition system. In this paper, we propose a novel scheme for accent classification, which uses decision-templates (DT) ensemble algorithm to combine base classifiers built on acoustic feature subsets. Different feature subsets can provide sufficient diversity among base classifiers, which is known as a necessary condition for improvement in ensemble performance. Compared with those methods of Majority Voting ensemble and Support Vector Machine, our ensemble scheme can achieve the highest performance. On the other hand, we investigate the possible reasons why ensemble systems can provide potential performance, in terms of diversity analysis. In our experiments, a native Mandarin speech corpus and a non-native multi-accent Mandarin speech corpus which contains three typical minorities’ accents in Yunnan, China, are adopted.