Design of smart synthetic speech answer-sheet system based on deep neural network and CR-DNN
Qingzhu Wu, Shaowei Xiong, Zhengyu Zhu · 2021 IEEE Conference on Telecommunications, Optics and Computer Science (TOCS) · 2021
Inspired by the success of utterance-based neural networks in deep feature extraction, in this study we propose the idea of classification- and regression-based deep neural network (CR-DNN) for detection of synthetic speech answer-sheet on intelligent oral English language learning app. In which, CR-DNN is composed of several classification-based and regression-based DNNs and every DNN can be seen as a block. Furthermore, the deep feature is extracted by CR-DNN firstly and then used for the input of detection system. The experimental results show that the deep feature extracted from CR-DNN can give good performance.