【作者】 王娜；

【导师】 邵克勇；

【作者基本信息】 东北石油大学 ， 模式识别与智能系统， 2011， 硕士

【摘要】 语音,作为人类最自然的交流工具,是人类获取资源与信息的重要来源。在信息技术高速发展的今天,让计算机能“听懂”人类的语音,是人—机进行沟通的最便捷的形式之一,语音识别就是这样的一门技术。近些年来,语音识别技术一直是计算机应用的热点。同样,机器人技术也逐渐成为现代自动化技术发展的标志之一。因此,将语音识别技术和机器人控制技术相结合,更是体现了当今最高技术上的自动化。本课题利用博创科技的旅行家—Ⅱ号机器人,以实现对机器人的语音控制为目标,针对语音信号的特征参数提取问题、语音识别算法的性能优化问题及机器人运动测试进行探讨研究,具体工作如下:提出了一种新的语音特征参数提取的方法。在传统的基于人耳听觉特性的MFCC特征参数基础上,将其与基于人的发声机理的共振峰参数结合,构成新的语音特征参数(MFCC+共振峰)。该方法从语音的发声机理和人耳听觉两方面出发,提取语音参数,此算法较传统方法包含信息量多,准确性高,且抗噪声能力强。针对传统的隐马尔科夫模型训练过程中对初值依赖性强、容易陷入局部最优的缺陷和DTW算法实时性的要求,分别提出了三步混合算法(TSMS)和DTW高效算法。经过仿真实验表明,TSMS算法的适应性更强,收敛速度更快,识别的准确率更高。DTW高效算法能够在满足实时性的前提下,减小计算量。给出两个改进算法的实验测试结果,验证了两者的优势。最后,设计机器人语音控制系统。创建Microsoft Speech SDK程序文件,编写DTW高效算法的端口程序和机器人运动测试程序,并对运动测试结果作了分析和说明。更多还原

【Abstract】 Voice, as the most natural human communication tool, is important sources of human getting resources information. With rapid development in information technology, so making the computer "understand" the human voice, is one of the most convenient forms of human - machine communication, speech recognition is such a technology. In recent years, speech recognition technology has been the focus of computer applications. Similarly, robot technology has gradually become one of the signs the development of automation technology. Therefore, the combination of speech recognition technology and robot control technology reflects the high technology of today’s automation.In this paper, we use Borch technology raveler -Ⅱrobot to achieve the goal of robot voice control, for speech signal feature extraction problem, the performance of speech recognition algorithms and motion tests of robot ,what we do are as follows:We propose a new way of speech feature parameters extraction.At the traditional MFCC feature parameters based on human auditory characteristics, combined with the formant parameters based on the person’s voice mechanism, form the new voice feature parameters (MFCC + resonance peak). The method is based on he voice mechanism of sound and the human auditory, it includes more speech parameters information than the traditional algorithms,also it has higher accuracy and strong resistance to noise.For the fefect of the high dependence on the initial value、easy to fall into local optimal and the requirements of real-time DTW algorithm during the training process of traditional hidden Markov model,we proposed three steps hybrid algorithm (TSMS) and DTW efficient algorithms. The simulation results show that, TSMS algorithms have stronger adaptability, faster convergence and higher recognition accuracy. DTW algorithm can efficiently satisfy the real-time requirements and reduce the amount of calculation. The given experimental results of two improved algorithms verifyied the advantages of them.Finally, we designed the robot voice control system. Create Microsoft Speech SDK file, write the port procedures of DTW efficient algorithms and robot testing procedures, and analyzed the test results.更多还原