基于DFCNN-CTC端到端的藏族学生普通话发音偏误检测  

DFCNN-CTC end-to-end based detection of Mandarin mispronunciation by Tibetan student

作　　者：甘振业[1,2] 周世华 曾浩 杨鸿武 GAN Zhen-ye;ZHOU Shi-hua;ZEN Hao;YANG Hong-wu(College of Physics and Electronic Engineering,Northwest Normal University,Lanzhou 730070,China;Engineering Research Center of Gansu Province for Intelligent Information Technology and Application,Lanzhou 730070,China;School of Educational Technology,Northwest Normal University,Lanzhou 730070,China)

机构地区：[1]西北师范大学物理与电子工程学院,甘肃兰州730070 [2]甘肃省智能信息技术与应用工程研究中心,甘肃兰州730070 [3]西北师范大学教育技术学院,甘肃兰州730070

出　　处：《西北师范大学学报（自然科学版）》2020年第5期49-53,108,共6页Journal of Northwest Normal University(Natural Science)

基　　金：国家自然科学基金资助项目(11664036,31860285)。

摘　　要：计算机辅助语音训练系统需要检测非母语者的错误发音,并提供详细的指导性反馈,有助于第二语言学习者更有效地提高发音水平.利用深度全序列卷积神经网络(Deep full convolutional neural network,DFCNN)和链接时序分类(Connectionist temporal classification,CTC)技术,建立了一种用于发音偏误检测和诊断任务的端到端语音识别方法.该方法不需要音位信息,也不需要强制对齐,以扩展声韵母为偏误基元,设计了64种偏误类型.实验结果表明,该方法能够有效地检测出错误发音,检测正确率为87.07%,错误拒绝率为7.83%,错误接收率为25.97%.Detecting mispronunciations produced by non-native speakers and providing detailed instructive feedbacks are desired in computer assisted pronunciation training system(CAPT),as it is helpful to L2 learners to improve their pronunciation more effectively.This paper use deep full convolutional neural network(DFCNN)and connectionist temporal classification(CTC)to build an end-to-end speech recognition for mispronunciation detection and diagnosis task.Our approach is end-to-end models,while phonemic information or forced alignment between different linguistic units,are not required.We designed 64 types of mispronunciation with the extended initial and final(XIF)as mispronunciation unit.The experimental results show that this method can effectively detect the wrong pronunciation,achieving a false rejection rate of 7.83%and a false acceptance rate 25.97%.The detection accuracy is 87.07%.

关 键 词：发音偏误检测 卷积神经网络 链接时序分类 端到端 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]