Automatic Pronunciation Evaluation in High-states English Speaking Tests Based on Deep Neural Network Models

This paper introduces our state-of-the-art automatic pronunciation evaluation technologies for high-states English speaking tests. A noise robust implementation of Goodness of Pronunciation based automatic scoring method and novel acoustic adaptation techniques have been proposed. Experimental results show that the proposed techniques outperform conventional methods. Using these posterior based sentimental features combined with supra-segmental prosodic features, the proposed pronunciation evaluation system can provide human expert level performance for high-states English speaking tests.