生态翻译理念下智能翻译机器人机交互研究  

Research on the interaction of intelligent translation robots under the concept of ecological translation

作　　者：李指南 LI Zhinan(Xianyang Vocational&Technical College,Xianyang,Shaanxi 712000,China)

机构地区：[1]咸阳职业技术学院,陕西咸阳712000

出　　处：《自动化与仪器仪表》2023年第9期224-228,共5页Automation & Instrumentation

基　　金：2022年咸阳职业技术学院教学改革研究项目课题《互联网+背景下高职学生外语口语焦虑研究》(2022JYB34)。

摘　　要：针对当前翻译理念的改变,研究试图在生态翻译视角下,构建一种新的智能语音识别模型,并将其用于翻译机器人的人机交互系统中。首先构建了DNN-HMM生态语音识别模型,其次使用N-Gram模型优化英文连续文本的翻译,最后分别采用seq2Seq网络和GPT-2神经网络实现两种不同类型的人机交互。结果显示,DNN-HMM模型的平均识别错误率远远低于GMM-HMM模型,仅为3.2%。在多轮人机交互中,DNN-HMM模型的精确率在0.77~0.89之间、召回率在0.78~0.86之间、F1值在0.78~0.85之间,三项检测指标均优于GMM-HMM模型。测试DNN-HMM模型的交互响应时间,93%的单轮交互和94%的多轮交互响应时间均在1 s以内。结合上述指标可以说明此次所构建的语音识别模型能够很好地完成翻译机器人人机交互任务。The progress of AI technology has promoted the research and development of most intelligent products in the field of human-computer interaction design.In view of the changes in current translation concepts,the research attempts to construct a new intelligent speech recognition model from the perspective of ecological translation and apply it to the human-computer interaction system of translation robots.At first,DNN-HMM ecological speech recognition model is constructed.According to the idea of ecological translation,the translation idea of collocation between adjacent words in context in N-Gram model is applied to the continuous translation of English text.Secondly,seq2Seq network and GPT-2 neural network are used to realize two different types of human-computer interaction.The results show that the average recognition error rate of DNN-HMM model is far lower than that of GMM-HMM model,only 3.2%.In multi-round human-computer interaction,DNN-HMM model has better accuracy,recall rate and F1 value.

关 键 词：生态翻译 机器人 人机交互 语音识别系统 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]