近年來隨著資訊科技化，人工智慧逐漸深入了我們的生活。深度學習的發展更讓語音辨識技術向前邁進了一大步，不僅能提高人機交互性，還可以應用於醫療等方面。我們用基於深度學習的語音識別技術進行錯誤發音的檢測，以此幫助有構音異常的人找出發音錯誤的地方以增加口說熟練度，並且輔助醫師進行診斷與治療。 本論文「基於卷積遞迴神經網路之構音異常評估技術」，延續過去學者的研究，提出基於CRNN-CTC 改善的系統，來提升錯誤發音檢測 (Mispronunciation Detection, MD) 的效果，達到構音異常的評估。本研究利用卷積遞迴神經網路 (Convolutional Recurrent Neural Network, CRNN) 與連結時序分類 (Connectionist Temporal Classification, CTC) 來訓練網路模型。並加入注意力機制，對構音異常評估的性能進行改善，以提升評估效果。實驗結果表明該方法用於構音異常的檢測有著良好效果。 ;In recent years, with the advancement of Information Technology, artificial intelligence has gradually penetrated into our lives. The development of deep learning has made speech recognition technology a big step forward, not only can improve human-computer interaction, but also can be applied to medical treatment and other aspects. In this paper, continuing the research of past scholars, we propose a system which is based on improved CRNN-CTC algorithm that can improve the effect of mispronunciation detection and achieve the evaluation of Articulation Dis-orders. We use Convolutional Recurrent Neural Network (CRNN) and Con-nectionist Temporal Classification (CTC) with attention model to train the model. The experimental results show that this method has a good effect in the auto-matic evaluation of abnormal articulation.
