A Robust Online Korean Teaching Support Technology Based on TCNN

Abstract

The emergence and development of multimedia forms provide technical support for online Korean language teaching. However, in many aspects, there are still many problems in online Korean teaching, such as noise interference, inaccurate translation, and unstable translation models. In this paper, we propose a Korean speech enhancement model based on temporal convolutional neural network and GRU neural network. We explore a Korean speech enhancement technology based on deep neural network, to make Korean speech teaching clearer and smoother, and to provide a robust support technology for online Korean teaching. First, we construct a temporal convolutional neural network to process and extract temporal feature in language data. Second, we introduce the sliding window mechanism and the maximum pooling structure to extract the feature in the speech time series data effectively and reduced the data scale. Third, we employ the Bi-GRU neural network and encoder-decoder for temporal data enhancement, which effectively avoids the problem that the hidden layer information cannot be effectively used in the traditional model, thereby improving the prediction accuracy and speed of speech data. The experimental outcomes demonstrate the effective evaluation performance of the method proposed in this paper.