语音识别在核电厂控制室中的应用研究

Study on Application of Speech Recognition in Control Room of Nuclear Power Plant

当前,核电厂数字化控制室以显示器作为主要媒介,为操纵员提供监控手段。相比于常规控制室,数字化控制室可以提供更加充分和丰富的监控信息。同时,其显示的信息量有限,导致操纵员不可避免地执行一些导航和查询类次要任务,影响人机交互效率。通过对语音模型训练算法、模型训练脚本、基于统计语言模型的算法进行试验及研究,并对核电厂主控制室相关词汇进行针对性训练、制定标准化语音指令,使操纵员可以通过视觉及听觉两方面与核电厂仪控系统终端进行交互,从而打破单模态输入、输出限制。试验结果表明,采用语音交互的方式,在保证指令准确的前提下,使操纵员完成指定任务所需时间缩短了约50%。这证明语音交互可以有效提升操纵员的人机交互效率。

Currently,digital control rooms in nuclear power plants use monitors as the primary medium to provide manipulators with the means to monitor and control.Compared with the conventional control room,the digital control room can provide more adequate and rich monitoring information.At the same time,the limited amount of information displayed leads to the manipulator inevitably performing some secondary tasks such as navigation and query,which affects the efficiency of human⁃computer interaction.By testing and researching the speech model training algorithm,model training script,and algorithm based on statistical language model,as well as conducting targeted training on the vocabulary related to the main control room of a nuclear power plant,and formulating standardized speech instructions,the manipulator can interact with the terminal of the instrumentation and control system of a nuclear power plant through both visual and auditory aspects,which breaks the unimodal input and output limitations.The test results show that by using speech interaction,the time required for the manipulator to complete the assigned task is reduced by about 50%under the premise of ensuring the accuracy of the instructions.This proves that speech interaction can effectively improve the manipulator’s human⁃computer interaction efficiency.