语音识别算法在ARM-linux平台上的研究与实现
发布时间：2019-04-19 19:36
【摘要】：随着电子技术和移动互联网的飞速发展，移动终端变得越来越贴近人们的日常生活，，更自然的人机交互方式显得更加重要。语音识别作为一个更自由和方便的人机交互的方式，走入人们的生活。而且随着移动终端的普及，ARM平台成为现在硬件平台的热点，所以研究语音识别在ARM-Linux平台上的实现，成为一个交叉的热点。论文模拟了日常家居家电控制的一套语音控制系统，目的在于在ARM-Linux平台上实现一个小词汇量的连续的非特定人的语音识别系统。论文首先深入的研究了语音识别的基本思想和基本流程，然后分阶段研究了语音识别的前处理算法、两种特征提取算法和三种比较重要的识别算法，其中重点研究和应用了隐形马尔可夫模型(HMM)，并深入研究了HMM的三个基本算法：前向后向算法、Viterbi算法、Baum-Welch算法。论文在选定了HMM作为实现算法之后，具体设计了系统的软件模块流程。结合剑桥大学开发的HTK工具包进行了对语音样本的HMM模型训练，并将训练出的模板运用语音识别引擎进行识别。然后将语音识别模块进行交叉编译，并植入ARM-Linux平台，建立一个基于ARM-linux平台的语音识别系统。论文为HMM在嵌入式中的应用做了基础性的探索，使语音识别进入日常生活的应用又迈进了一步。
[Abstract]:With the rapid development of electronic technology and mobile Internet, mobile terminals become more and more close to people's daily life, and more natural man-machine interaction becomes more and more important. Speech recognition, as a freer and more convenient way of human-computer interaction, enters people's lives. And with the popularity of mobile terminals, ARM platform has become the hot spot of hardware platform now, so the research on the realization of speech recognition on ARM-Linux platform has become a cross-focus. In this paper, a set of speech control system for household appliances control is simulated in order to realize a small vocabulary continuous speaker-independent speech recognition system on the ARM-Linux platform. In this paper, the basic idea and flow of speech recognition are deeply studied, and then the pre-processing algorithm, two feature extraction algorithms and three more important recognition algorithms are studied in stages. Three basic algorithms of HMM: forward-backward algorithm, Viterbi algorithm and Baum-Welch algorithm are studied in detail. The hidden Markov model (HMM),) is studied and applied in detail. After choosing HMM as the implementation algorithm, the software module flow of the system is designed. Combined with the HTK toolkit developed by Cambridge University, the HMM model of speech samples is trained, and the trained templates are recognized by speech recognition engine. Then the speech recognition module is cross-compiled and implanted into the ARM-Linux platform to establish a speech recognition system based on ARM-linux platform. This paper makes a fundamental exploration for the application of HMM in embedded system, and makes speech recognition step forward into the application of daily life.
【学位授予单位】：河北科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TN912.34;TP368.1
本文编号：2461218
[Abstract]:With the rapid development of electronic technology and mobile Internet, mobile terminals become more and more close to people's daily life, and more natural man-machine interaction becomes more and more important. Speech recognition, as a freer and more convenient way of human-computer interaction, enters people's lives. And with the popularity of mobile terminals, ARM platform has become the hot spot of hardware platform now, so the research on the realization of speech recognition on ARM-Linux platform has become a cross-focus. In this paper, a set of speech control system for household appliances control is simulated in order to realize a small vocabulary continuous speaker-independent speech recognition system on the ARM-Linux platform. In this paper, the basic idea and flow of speech recognition are deeply studied, and then the pre-processing algorithm, two feature extraction algorithms and three more important recognition algorithms are studied in stages. Three basic algorithms of HMM: forward-backward algorithm, Viterbi algorithm and Baum-Welch algorithm are studied in detail. The hidden Markov model (HMM),) is studied and applied in detail. After choosing HMM as the implementation algorithm, the software module flow of the system is designed. Combined with the HTK toolkit developed by Cambridge University, the HMM model of speech samples is trained, and the trained templates are recognized by speech recognition engine. Then the speech recognition module is cross-compiled and implanted into the ARM-Linux platform to establish a speech recognition system based on ARM-linux platform. This paper makes a fundamental exploration for the application of HMM in embedded system, and makes speech recognition step forward into the application of daily life.
【学位授予单位】：河北科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TN912.34;TP368.1
【参考文献】
相关期刊论文 前5条
1 刘维亭,朱志宇;基于小波网络和HMM的语音识别方法[J];电声技术;2004年11期
2 丁爱明;;作为说话人识别特征参量的M FCC的提取过程[J];电子工程师;2006年01期
3 荆嘉敏,刘加,刘润生;基于HMM的语音识别技术在嵌入式系统中的应用[J];电子技术应用;2003年10期
4 胡珊珊;刘加;王国梁;;嵌入式实时英语语音识别系统的设计和实现[J];电子技术应用;2006年09期
5 谢家春;葛永军;;基于ARM的车载语音识别系统设计与实现[J];电子世界;2013年01期
相关博士学位论文 前1条
1 吴斌;语音识别中的后处理技术研究[D];北京邮电大学;2008年
相关硕士学位论文 前7条
1 付维;基于HMM的机器人语音识别系统的研究[D];武汉科技大学;2011年
2 周冀辉;基于ARM9的语音控制智能终端的设计[D];华中科技大学;2010年
3 郭秋平;基于ARM系统的Linux平台移植研究[D];浙江大学;2006年
4 孙长斌;基于特定人的小词汇量孤立词语音识别[D];吉林大学;2007年
5 顾亚强;非特定人语音识别关键技术研究[D];国防科学技术大学;2009年
6 胡优;语音识别算法及其在嵌入式中的应用[D];电子科技大学;2010年
7 黄文涛;基于神经网络的嵌入式语音识别系统研究[D];广东工业大学;2012年
本文编号：2461218
本文链接：https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/2461218.html
