CASIA OpenIR  > 毕业生  > 硕士学位论文
基于嵌入式ARM模块的语音识别算法优化
其他题名Optimization of Algorithms in Embedded Speech Recognition Module Based on ARM
陆盟
学位类型工学硕士
导师李成荣
2010-05-31
学位授予单位中国科学院研究生院
学位授予地点中国科学院自动化研究所
学位专业计算机应用技术
关键词语音识别 Arm 端点检测 拒识 语音增强 Speech Recognition Arm End-point Detection Oov Rejection Speech Enhancement
摘要语音识别是通过识别和理解过程把人类的语音信号转变为文本或命令的技术,是信息处理技术研究的一个热点。而基于PC的语音识别技术理论上已经趋于成熟,而随着嵌入式技术的发展,嵌入式语音识别系统以其体积小巧、可携带性以及软硬件的可裁剪性为语音识别技术提供了一个新的研究方向,在嵌入式设备上实现语音识别技术已经成为语音识别发展的重点和难点。迄今为止,嵌入式语音识别在理论上的研究尚不成熟,还有很多问题等待解决,这也严重困扰了它的市场化应用。 在此背景下,本课题在已有ARM嵌入式语音识别模块上进行有关算法的优化和改进工作。论文分别从理论分析、系统硬件平台的总体设计、系统软件的优化等方面,对语音识别在ARM模块上的应用做了研究。 1、本文详细介绍了语音识别的发展历史与研究现状;具体陈述语音识别技术的基本原理和主要研究方法,并引出语音识别技术中最常用到的两种算法DTW和HMM的数学模型,为进一步的语音识别研究打下基础。 2、本文分析设计了嵌入式语音识别系统的总体方案,主要包括以下两个部分:语音识别系统的控制系统、语音的输入输出系统;文中详细介绍了各个系统的作用以及它们之间的连接方式。 3、在系统软件的改进和优化方面,首先提出了现有系统遇到的一些实际的问题,并由此提出改进方案,具体包括端点检测算法、拒识算法和语音增强算法,给出它们的理论依据和实验结果。 经过实验,本文的改进算法可以提高嵌入式语音识别模块的系统性能,证明了改进算法的有效性。 论文的结尾对以后的工作进行了合理的展望。
其他摘要Speech Recognition is a type of technology, which helps to transfer the speech signals into homogeneous text or command during the process of recognition and comprehension. Speech recognition technology is on the hot spots for its importance and difficulty in research recently. However, speech recognition based on PC drives to maturity stage theoretically and yields substantial results in practice. As the development of embedded technology, with the portability and changeability both in software and hardware, ARM provides speech recognition technology with a new research direction. Speech recognition based on ARM has been the emphasis and difficulty of speech recognition technology. However, the research level of embedded speech recognition is still immature now, which is the biggest obstacle of its marketization. Under this background, the dissertation does the improvement and optimization on currently available embedded speech recognition module based on ARM, which has been classified in theoretical analysis, speech recognition system’s hardware platform design, system’s software analysis and optimization respectively. 1、The dissertation introduces the development history and research status of speech recognition in detail; Gives the specific introduction about the basic principles and the main research methods of the speech recognition technology; two common mathematical models are derived, dynamic time warping and hidden markov model, as the basement for the further research of speech recognition. 2、The dissertation analyses and designs the general architecture for the system, mainly including the following two parts, speech recognition’s control system and speech signal input & output system. The dissertation introduces the function of each system as well as their relationship in detail. 3、In software optimization, the paper firstly presents problems of currently available embedded system. Then it proposes solution accordingly and gives their theoretical basis and experimental results. According to the experiments, the improving algorithms of system software can enhance system’s performance correspondingly, which exemplify the effectiveness of improving algorithms. In the end of the dissertation, rationalization proposals are given to future work.
馆藏号XWLW1463
其他标识符200628014629081
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/7529
专题毕业生_硕士学位论文
推荐引用方式
GB/T 7714
陆盟. 基于嵌入式ARM模块的语音识别算法优化[D]. 中国科学院自动化研究所. 中国科学院研究生院,2010.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
CASIA_20062801462908(839KB) 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[陆盟]的文章
百度学术
百度学术中相似的文章
[陆盟]的文章
必应学术
必应学术中相似的文章
[陆盟]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。