家庭陪护机器人语音控制系统研究与设计

发布时间：2018-10-08 09:00

【摘要】：随着机器人技术的快速发展以及社会需求的不断增加,清洁、安保、陪护及康复等服务机器人已经逐步走入大众日常生活。为了方便用户使用,本文基于实验室自主研发的一款家庭智能陪护机器人,研究设计了一套机器人语音控制系统,使得用户能够以一种最为亲切自然的方式与机器人进行交互。本文在完成整体方案设计的基础上,对系统涉及的声源定位和语音识别两个技术环节进行了重点研究。声源定位方面,本文在比较分析经典声源定位算法的基础上,选取一种基于到达时延的广义互相关法(GCC:Generalized CrossCorrelation)定位算法。针对该算法在噪声和回响较大环境中效果不够理想的问题,对其进行了改进,增加一个PHAT加权函数,对频谱信号进行处理;同时引入一种抛物线插值算法,在不提高采样频率的前提下,有效提高了空间搜索定位精度。此外,本文还完成了麦克风阵列的拓扑结构设计以及麦克风、声卡等硬件的选型工作。语音识别方面,本文采用Kaldi开源语音识别开发平台,基于准备的训练数据及脚本,进行了单音子、三音子模型及改进算法模型的训练,并对其中涉及的HMM模型、基于加权有限状态机(WFST)的静态空间网络等理论进行了研究,针对其中的核心问题提出了相应的解决方法。通过语音识别精度测试实验,对各个模型的识别精度进行比较,最终确定系统采用基于三音子的LSM语音识别模型。最后,本文制定了上下位机通信协议,搭建了语音控制系统实验平台,在机器人运动控制中,针对控制精度不高的问题,采用了一种PID控制同模糊控制融合的控制策略。在此基础上,进行了声源定位、自主转向以及语音指令识别并实时响应的相关测试。大量的实验结果及初步的应用表明,本文设计的陪护机器人语音控制系统具有较高的定位和指令识别精度,可以满足陪护机器人在家庭环境中的使用需求。
[Abstract]:With the rapid development of robot technology and the increasing demand of society, cleaning, security, escort and rehabilitation service robots have gradually entered the daily life of the public. For the convenience of users, this paper designs a robot voice control system based on a home intelligent escort robot, which is developed independently by the laboratory, which enables the user to interact with the robot in a most friendly and natural way. On the basis of the overall scheme design, this paper focuses on the two technical links of sound source location and speech recognition involved in the system. In the aspect of acoustic source location, based on the comparison and analysis of classical acoustic source localization algorithms, a generalized cross-correlation (GCC:Generalized CrossCorrelation) localization algorithm based on arrival delay is selected. In order to solve the problem that the effect of the algorithm is not ideal in the environment of high noise and reverberation, an PHAT weighting function is added to process the spectrum signal, and a parabola interpolation algorithm is introduced. On the premise of not increasing the sampling frequency, the accuracy of spatial search and location is improved effectively. In addition, the topology design of microphone array and the hardware selection of microphone and sound card are also completed. In the aspect of speech recognition, this paper uses the Kaldi open source speech recognition development platform, based on the training data and script, carries on the training of monosyllabic, tri-phonon and improved algorithm model, and the HMM model involved in it. The theory of static space network based on weighted finite state machine (WFST) is studied, and the corresponding solutions to the core problems are proposed. The recognition accuracy of each model is compared by testing the accuracy of speech recognition. Finally, the LSM speech recognition model based on triphonon is adopted in the system. Finally, the communication protocol of upper and lower computer is established, and the experimental platform of speech control system is built. In order to solve the problem of low control precision in robot motion control, a control strategy combining PID control with fuzzy control is adopted. On this basis, the sound source location, autonomous steering, speech command recognition and real-time response are tested. A large number of experimental results and preliminary applications show that the speech control system designed in this paper has a high accuracy of positioning and instruction recognition, and can meet the needs of the accompanying robot in the home environment.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242

【参考文献】