Call System for Japanese Students of English Using Pronunciation Error Prediction and Formant Structure Estimation

We propose an effective pronunciation learning system using speech processing and recognition technologies. For automatic phonemic error detection, we perform pronunciation error prediction with orthographic text. To improve its reliability, we adopt speaker adaptation and segment input pair-wise classifier for effective feedback to generate personalized instruction. We also adopt formant frequency normalization technique (ML-VTLN and MMSE). The effect of adaptation and normalization is addressed in the paper.