家用陪护机器人语音交互与控制系统研究

发布时间：2018-05-16 23:13

本文选题：老年陪护机器人 + 神经网络　；参考：《昆明理工大学》2017年硕士论文

【摘要】：近年来,随着深度学习技术的高速发展,语音交互技术在很多行业领域得到了广泛的应用。但在现实环境中,由于干扰的存在,语音识别效果差,虽然语音交互软件在市面上已普遍存在,但距离用户广泛接受还有一定的距离。面临中国老龄化加剧现状,目前我国增加了对老年人家用陪护机器人研发的投入,而且将其作为服务机器人领域发展的重要目标。本文利用集合语音分离、语音识别等技术为老年人设计了一款家用语音控制陪护机器人。本文在对基于深度神经网络的单通道语音分离系统仿真的基础上,对语音分离的目标函数,深度神经网络隐层数目以及激励函数进行分析,提出一种适合用于汉语的语音分离系统模型,并实验验证了该系统对分离的有效性。在讨论了基于深度神经网络的汉语孤立词、非特定人语音识别的基础上,使用基于MFCC与RASTA-PLP的组合特征在基于深度神经网络模型上进行语音识别仿真,分析了这种组合特征对语音识别正确率的影响。将语音分离与语音识别结合,构建了陪护机器人语音交互控制系统,采用JAVA编程设计了基于安卓的语音控制系统,并分析一种通过语音控制的硬件系统,通过语音控制机器人,实现人机交流,并验证了陪护机器人语音交互控制系统的性能。本文针对老年人在使用陪护机器人过程中,尤其在有噪音干扰的情况下,容易造成识别不准确的问题,提供一种小词汇量的非特定人、孤立词语音识别系统模型,并实现对陪护机器人的控制。该系统也适合用于其他需要识别孤立词的人机交互系统,比如车载系统。
[Abstract]:In recent years, with the rapid development of deep learning technology, voice interaction technology has been widely used in many fields. But in the real environment, because of the existence of interference, speech recognition effect is poor. Although speech interactive software is widely available in the market, there is still a certain distance from widely accepted by users. Facing the aggravation of aging in China, our country has increased the investment in the research and development of the elderly household escort robot, and regarded it as the important goal of the development of the field of service robot. In this paper, a kind of home speech control escort robot is designed for the elderly by means of aggregate speech separation and speech recognition. Based on the simulation of single channel speech separation system based on deep neural network, this paper analyzes the objective function of speech separation, the number of hidden layers and the excitation function of the deep neural network. A speech separation system model suitable for Chinese is proposed, and the effectiveness of the system is verified by experiments. Based on the discussion of Chinese isolated words based on depth neural network and independent speech recognition, the speech recognition simulation based on the combination of MFCC and RASTA-PLP is carried out on the model of depth neural network. The influence of the combined features on the accuracy of speech recognition is analyzed. By combining speech separation with speech recognition, a speech interactive control system of escort robot is constructed. A speech control system based on Android is designed by JAVA programming. A hardware system based on speech control is analyzed, and the robot is controlled by voice. The man-machine communication is realized and the performance of the accompanying robot voice interactive control system is verified. In order to solve the problem of inaccurate recognition in the process of using accompanying robot, especially in the case of noise interference, this paper provides a small vocabulary model of isolated words speech recognition system. And realize the control of escort robot. The system is also suitable for other isolated word recognition systems, such as vehicular systems.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242

【参考文献】