【作者】 高海英；

【导师】 刘京诚；

【作者基本信息】 重庆大学 ， 光学工程， 2006， 硕士

【摘要】 一个完整的迎宾机器人系统需要综合运用多种人工智能技术,其目标是建立起一个“人”的模型,让机器人听懂人的语言就是其中之一。机器人语音识别系统是迎宾机器人头部的重要组成部分,机器人语音识别系统主要是对人的声音进行语音识别并做出判断,然后输出相应的动作指令控制机器人的动作。机器人要能够对非特定人的语音进行识别,当机器人接到语音信号后,对语音进行预处理,端点检测,特征向量提取,再通过对语音库里的特征向量进行模式匹配,输出识别结果。机器人识别的范围由其训练过的语音库决定,特定的语音对应特定的动作。传统的机器人听觉系统一般是以PC机为系统运行平台对机器人进行控制,其特点是用一台计算机作为机器人的信息处理核心通过接口电路对机器人进行控制,处理能力比较强大,语音库比较完备,系统更新以及功能拓展比较容易。缺点是比较笨重,不利于机器人的小型化和复杂条件下进行工作,此外功耗大,成本高。近几年来微型处理器性能的提高促进了机器人的迅速发展,特别是大存储量及高速运算DSP处理芯片的出现使机器人在脱机状态下,独立完成复杂的语音信号处理和动作指令成为可能。论文在硬件上使DSP芯片TMS320VC5509与音频芯片TLV320AIC23相结合作为机器人的语音采集系统,采用DSP作为语音识别处理器,具有较快的处理速度,并采用DSP集成开发环境CCS 3.1完成对DSP芯片的外部接口配置;在软件开发上,论文分析了语音信号的特点,对语音信号进行预处理和端点检测,语音特征向量采用美尔频率倒谱系数(MFCC),模式匹配和训练采用DTW动态规整算法,实现了机器人语音指令的识别。在动作控制上,采用FPGA芯片EP1CT100C8作为机器人头部动作逻辑控制器,使机器人能够根据非特定人的语音命令做出规定的头部动作。DSP的使用使机器人在脱机状态下能够独立快速完成复杂的语音信号处理和动作指令控制,FPGA系统的开发降低了时序控制电路和逻辑电路在PCB板所占的面积,使机器人的“大脑”的语音处理部分微型化,低功耗。一个体积小,低功耗,高速度能完成特定范围语音识别和动作指令的机器人系统的研制具有很大的实际意义。更多还原

【Abstract】 An integrated guiding robot system need use many kinds of artificial intelligence technologies together, whose goal is to build up a man model.It is one of them that we should make the robot understand the languages. The speech recognition system of the robot, which is the important part of theguiding robot head, mainly recognizes the voice and makes the decision, then gives out the action instruction to control the robot. The robot should recognize the voice of nonspecial persons. When receiving speech signals, it preprocesses the speech, detects the end point and extracts characteristic vector. Then it gives out the recognition results through the pattern matching of the characteristic vector in the speech bank. The recognition ranges of the robot are decided by the trained speech bank, in which the special speech matches the special action.The traditional robot hearing system usually controls the robot using PC as the running platform of the system. Its traits are as following: it uses a computer as the information processing core of the robot to control it through the interface circuit.Its processing ability is quite powerful, and speech bank is complete.At last it is quite easy to renew the system and develop the function.But it also has some disadvantages, firstly it is a little heavy, which makes the robot unavailable in the conditions of miniaturization and complexity. Then its power consumption is high and costs much.In the recent years, the improvement of the microprocessor performance promotes the robot to develop fast, especially the appearance of DSP which has mass memory and high-speed operation makes it possible that a robot can complete the complex speech signal processing and action instruction off line.In this paper TMS320VC5509 and TLV320AIC23 are made together to be the speech acquisition system of a robot in the hardware. DSP is used to be the speech recognition processor, whose processing speed is quite fast. CCS 3.1 is used to finish configuring the peripheral interface of DSP. In development of the software the characteristics of speech signals are analyzed, speech signals are preprocessed and detected by end points. Speech eigenvector adopts MFCC, and pattern matching and更多还原