Pitch-Aware RNN-T for Mandarin Chinese Mispronunciation Detection and Diagnosis

Subsequently, Zhang et al. [1] adopted Mispronunciation Detection and Diagnosis (MDD) systems, an autoregressive model, the Recurrent Neural Network Transducer leveraging Automatic Speech Recognition (ASR), face two (RNN-T) [9], for MDD. This approach aims to capture main challenges in Mandarin Chinese: 1) The two-stage models the temporal dependence of mispronunciation patterns, showing create an information gap between the phoneme or tone classification better performance than Connectionist Temporal Classification stage and the MDD stage.