摘要
本文提出了一种基于多普勒微波雷达的发音动作检测与命令词识别方法.该方法利用微波雷达的多普勒特性检测发音过程中面部肌肉的微小变化,实现不依赖语音声学信号的命令词识别.本文首先设计实现了一个基于多普勒微波雷达的发音动作检测系统,并基于此系统构建了一个包含2个说话人的命令词识别数据库.然后,本文研究了基于支持向量机和卷积神经网络模型的雷达数据分类方法,并对比了不同模型和特征组合在单话者建模和多话者建模情况下的命令词识别性能.实验结果表明,本文设计的数据采集系统可以有效检测发音动作,所构建的卷积神经网络分类器可以取得90%以上的命令词识别准确率.
This paper proposes a method based on Doppler microwave radar for articulatory movement detection and command word recognition.Utilizing the Doppler characteristics of microwave radars,this method detects small changes in facial muscles during pronunciation,and realizes command word recognition without using acoustic signals.In this paper,an articulatory detection system based on Doppler microwave radars is first designed and implemented.A command word recognition database containing two speakers based on this system is constructed.Then,this paper studies the classification methods of radar data based on support vector machine and convolutional neural network model.The command word recognition performances of different models and features are compared under both single-speaker modeling and multi-speaker modeling conditions.The experimental results show that the data acquisitionsystem designed in this paper can effectively detect the articulatory movement,and the constructed convolutional neural network classifier can obtain an accuracy higher than 90% in the command word recognition task.
作者
吴鹏飞
凌震华
WU Peng-fei;LING Zhen-hua(Department of Electronic Engineering and Information Science,University of Science and Technology of China,Hefei 230027,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2020年第2期426-430,共5页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61871358)资助.
关键词
发音动作检测
多普勒雷达
卷积神经网络
支持向量机
articulatory movement detection
Doppler radar
convolutional neural network
support vector machine
