摘要
为了解决非英语母语学习者在语音识别中出现的语法错误问题,提出了基于神经机器翻译的语法错误检测语音识别中的语言模型。将构建的语言模型与传统的语言模型进行了比较,分析了该语言模型算法在语法错误检测中的性能。结果显示,由于神经机器翻译具有特定的内部结构,可以结合上下文信息进行语音识别,神经机器翻译模型可以更好地进行语法错误检测。通过比较不同语言模型结果,该方法比基于规则的方法有显著优势,神经机器翻译语言模型的精度、召回率、F值分别为0.54、0.52、0.53,证明了该模型具有较好的性能。
In order to solve the problem of grammatical errors in speech recognition of non-English native language learners,a language model for speech recognition and syntax error detection based on neural machine translation is proposed.The performance of the algorithm in grammar error detection is analyzed by comparing the constucted language model with the traditional one.The results show that because of its specific internal structure of the neural machine,neural machine translation can recognize speech with context information,and neural machine translation model can detect grammatical errors better.The results of different language models are compared,and the proposed method has a significant advantage over rule-based methods.The accuracy,recall rate and F value of neural machine translation language model are 0.54,0.52 and 0.53 respectively,which proves that the model has better performance.
作者
吴迪
WU Di(Shaanxi Xueqian Normal University,Xi’an 710000,China)
出处
《信息技术》
2022年第5期82-87,共6页
Information Technology
基金
陕西学前师范学院科研计划项目(2019ZDRS03)。
关键词
自动语音识别
语言建模
神经机器翻译
语法错误检测
automatic speech recognition
language modeling
neural machine translation
syntax error detection
