语音自动识别技术及其在电信彩铃业务中的应用研究

发布时间：2018-06-11 11:04

本文选题：语音识别 + 隐马尔可夫模型　；参考：《西安电子科技大学》2014年硕士论文

【摘要】：语音识别技术(Automatic Speech Recognitio n)又称ASR技术,它是一种将人声语音信号转化为可以被计算机程序识别的文字信息,从而识别、了解讲话人所发出指令和意图的技术。语音识别技术的终极目标是实现机器与人之间可通过自然语言进行互动。二十世纪五十年代,美国电话电报公司(ATT)贝尔实验室里建成第一台语音识别系统——Audry系统,它仅能识别十个英文数字。上世纪八十年代末期,数字信息技术、计算机软硬件技术的发展掀起前所未有的高潮,为语音识别的研发带来坚实的技术支撑。语音识别技术在商业领域首先得以应用,并迅速扩展到社会生产和生活的每一领域,自身也受到不断更新的技术的要求。本文首先在第一章简要介绍国内外语音识别技术的发展现状。其次,从ASR技术的基本模型方面,介绍了端点检测、预处理、模块生成、模块匹配和模块管理等环节中的模式识别和数字信号处理的原理。本文通过分析和研究语音识别技术的发展历程,分别通过HMM(隐马尔科夫模型)和DTW(动态时间规正法)设计汉语语言的语音识别系统。HMM具有较强的时间序列建模能力,对时间进行特征参数的训练,每个语音对应自己的隐马尔科夫模型,语音与隐马尔科夫模型进行对应匹配,完成识别过程。HMM通常适用在需要识别大词汇量的情况。DTW能够很好的处理语音信号特征参数的时间长短不同性问题,具有识别速度快、系统费用少和有效的小词汇量处理作用。每节结尾都通过Matlab进行仿真,并提供脚本。最后对整个彩铃语音识别平台的系统进行了系统的介绍。彩铃语音识别平台通过运用声动炫铃系统中的IVR架构设计技术,并结合语音识别和语音合成等语音领域的关键技术,采用voice xml标准语言工具研发的一个应用系统。用户通过输入一个特定的号码进入服务器,用语音或者按键选出歌手名或者歌曲的形式传输给系统命令,就能获得自己定制彩铃。通过采用专用的语言界面,方便用户的操作过程,主要使用VXML技术实现电话流程。彩铃语音识别系统还设有彩铃语音识别平台场景及脚本代码。
[Abstract]:Automatic speech recognition (ASR) is a technology that converts speech signals into text information that can be recognized by computer programs, so as to recognize and understand the instructions and intentions issued by speakers. The ultimate goal of speech recognition technology is to realize the interaction between machines and people through natural language. In the 1950s, AT & T Bell Labs built the first speech recognition system, the Audry system, which recognized only ten English numbers. In the late eighties of last century, the development of digital information technology, computer software and hardware technology set off an unprecedented climax, which brought solid technical support for the research and development of speech recognition. Speech recognition technology has been first applied in the commercial field, and has been rapidly extended to every field of social production and life. It is also subject to the requirements of constantly updated technology. In the first chapter, this paper briefly introduces the development of speech recognition technology at home and abroad. Secondly, from the basic model of ASR technology, the principles of pattern recognition and digital signal processing are introduced, such as endpoint detection, preprocessing, module generation, module matching and module management. By analyzing and studying the development of speech recognition technology, this paper designs the speech recognition system of Chinese language by hmm (Hidden Markov Model) and DTW (dynamic time normalization method), respectively. Each speech has its own Hidden Markov Model, and the speech and Hidden Markov Model are matched. The completion of the recognition process. Hmm is usually applicable to the need to identify large vocabulary. DTW can deal with the different time of speech signal feature parameters well. It has the advantages of fast recognition, low system cost and effective small vocabulary processing. The end of each section is simulated by Matlab, and the script is provided. Finally, the system of CRBT speech recognition platform is introduced systematically. The platform of color ring tone recognition adopts voice xml standard language tool to develop an application system, which is based on the design technology of voice xml architecture, and combined with the key technologies in speech field such as speech recognition and speech synthesis. By entering a specific number into the server and selecting the singer's name or song by voice or keystroke to transmit to the system command, the user can obtain his own custom color bell. By using a special language interface to facilitate the user's operation process, the main use of VXML technology to achieve the telephone flow. The CRBT speech recognition system also has the scene and script code of CRBT speech recognition platform.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN912.34

【相似文献】