基于深度神经网络的普通话智能评测关键技术研究  

Research on Key Technologies of Mandarin Intelligent Evaluation Based on Deep Neural Network

作　　者：洪世勇[1] Hong Shiyong(School of Information Engineering,Jiujiang Vocational University,Jiujiang 332000,China)

机构地区：[1]九江职业大学信息工程学院,江西九江332000

出　　处：《安徽电子信息职业技术学院学报》2021年第2期1-5,共5页Journal of Anhui Vocational College of Electronics & Information Technology

基　　金：江西省教育厅科学技术研究项目(GJJ203909)。

摘　　要：为改进传统普通话水平测试第四题测评中的不足,实现测评的智能化,减轻工作人员工作量,提升考试的公平性,采用基于深度神经网络的普通话智能评测关键技术,设计使用深度神经网络-隐马尔科夫模型的语音识别模型(DNN-HMM),识别考生表述的语音,更好地反映考生真实发音的标准程度;使用条件随机场训练分句模型,更加精准地判定考生发音的流畅程度;采用矢量空间模型,较好地评估考生词汇语法使用的规范程度。In order to improve the shortcomings in the assessment of the fourth question of the traditional Putonghua proficiency test,realize the intelligence of the assessment,reduce the workload of the staff,and improve the fairness of the test,using the key technology of Putonghua intelligent evaluation based on deep neural network,a speech recognition model based on deep neural network hidden Markov model(DNN-HMM)is designed to recognize the speech of candidates and better reflect the standard degree of their real pronunciation.Using conditional random field training clause model,we can judge the fluency of examinees'pronunciation more accurately;using vector space model,we can better evaluate the standard degree of examinees'vocabulary and grammar use.

关 键 词：深度神经网络 普通话智能评测 语音识别模型 矢量空间模型 条件随机场训练分句模型 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]