基于语音识别的智能英语翻译机器人人机交互系统  

Human-machine interaction System of Intelligent English translation robot based on Speech Recognition

作　　者：梁亚敏[1] 李亚峰 LIANG Yamin;LI Yafeng(Xian Yang Normal University,Xianyang Shanxi 712000,China)

机构地区：[1]咸阳师范学院,陕西咸阳712000

出　　处：《自动化与仪器仪表》2022年第9期196-200,205,共6页Automation & Instrumentation

基　　金：省级课题《需求分析理论下大学英语课程设置的教学改革与研究》(SGH17H511);校级课题《生态翻译学理论在奥斯汀作品孙致礼译本中的应用与研究》(XSYK18030);校级课题《咸阳师范学院2017“青蓝人才”培养项目》(XSYQL201703)。

摘　　要：针对翻译机器人人机交互的智能化需求,提出一种语音识别的翻译机器人交互系统。其中,为提高语音识别的质量,在语音信号预处理基础上,分别采用DNN-HMM和WaveNet卷积网络对语音进行识别和合成,然后采用Seq2Seq模型和GPT-2神经网络实现人机的单轮交互和多轮交互。结果表明,通过以上的综合方案,语音识别的准确率平均为97%;单轮和多轮交互训练取得良好效果,其中多轮交互的精确率、召回率提升至85%以上;最后将以上方案部署到系统中进行测试,能满足翻译的人机交互需求。Aiming at the intelligent requirements of human-computer interaction of translation robot, a speech recognition translation robot interaction system is proposed. In order to improve the quality of speech recognition, dnn-hmm and WaveNet convolution network are used to recognize and synthesize speech based on speech signal preprocessing, and then seq2 seq model and gpt-2 neural network are used to realize single round and multi round human-computer interaction. The results show that the average accuracy of speech recognition is 97% through the above comprehensive scheme;Single round and multi round interactive training are achieved good results, and the accuracy and recall rate of multi round interactive training are increased to more than 85%;Finally, the above scheme is deployed to the system for testing, which can meet the human-computer interaction needs of translation.

关 键 词：语音识别 交互模式 HMM Seq2Seq 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]