摘要
针对机器人缺乏对语音指令的深层识别能力,提出一种基于DNN/GMM的深层信息识别方法.首先,在全息地图环境数据基础上,运用双隐层深度神经网络构建家庭环境神经网络模型;其次,基于该模型对语音指令进行目标对象深层信息识别,并据此提取相关预备指令;再次,基于语音指令高斯混合模型识别语音指令类型;最后,依据不同指令类型选定最优服务对象深层信息识别方法,之后提取服务指令.在一般家庭中构建实验环境,结果验证了该方法的正确性和有效性,且使得机器人依据指令的深层信息能够更加准确地理解并执行指令.
According to robot lacks of ability of recognizing deep information of speech instructions, this paper proposes a deep infor- mation recognition method based on DNN/GMM. First of all, on the basis of the environmental data of holographic map, to build neu- ral network model of family environment by improved double hidden layer deep neural network; Secondly, mining deep information of target object of speech instruction based on this model, and then extracting related prepare instruction;Thirdly, to recognize the type of speech instruction based on Gaussian Mixture Model;Finally,to choose the best method of service object recognition by different in- struction type, and then extract service instruction. To build experimental environment in the general family, the experiment results show that the accuracy and validity of this method,and also enables the robot more accurately understand and executive instructions.
出处
《小型微型计算机系统》
CSCD
北大核心
2015年第6期1347-1352,共6页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61305113)资助
