基于倒谱的语音特征提取算法的╲t 数字语音识别系统.docVIP

PAGE 分类号： TP311.1 2011届本科生毕业设计 题目：基于倒谱的语音特征提取算法的 数字语音识别系统 作 者 姓 名： 学 号： 2007110119 系(院)、专业： 信息工程学院 计算机科学与技术 指导教师姓名： 辛政华 指导教师职称： 助教 2011 年3 摘 要 语音识别是研究让机器能够听懂人类口述的自然语言的一门学科，其最终目标是实现入与机器进行自然语言通信。因此我和另两位同学合作开发一个关于基于Mtalab只有0到9十个数字的数字信号处理系统。介绍了语音识别的关键技术，主要有语音识别单元选取的端点检测、特征参数提取技术、模式匹配准则及模型训练技术三个方面，分别由我和其他两位同学完成。我主要负责特征参数的提取技术。在语音信号处理中，常用的语音特性是基于Mel频率的倒谱系数（MFCC）以及一些语音信号的固有特征，如基音频率等。倒谱可以较好地将语音信号中的激励信号和声道响应分离，并只需要用十几个倒谱系数就能较好地描述语言信号的声道响应，在语音信号处理中占有很重要的位置。本论文设计了基于倒谱的语音特性参数提取算法，并在Matlab中予以实现。 关键词：语音识别；特征参数；模式匹配；倒谱；MFCC ABSTRACT Speech recognition belong to the study of the machine which is a hot spot of application can understand human verbal natural language of a discipline, with the ultimate goal to achieve the natural language communication with machines. So I and two other students develop a Mtalab based only on the ten digits 0 to 9, digital signal processing systems. So my other two students and I develop a digital signal .processing system base On ten digits 0 to 9 using Matlab,the key technology of speech recognitim ,include endpoint detection able to select speech recognition units,feature extraction,pattern criteria.we take charge for three aspects respectively. I’m responsible for the characteristic parameters extraction. In speech signal processing, voice characteristics of commonly used is based on Mel frequency cepstrum coefficients (MFCC), and some of the inherent characteristics of speech signals, such as the pitch frequency. Cepstrum can better motivate the speech signal separation of signal and channel response, and only need a dozen cepstrum can be used to describe the language of the channel response signals in speech signal processing plays an important position. In the paper, cepstrum-based speech feature extraction algorithm and Matlab, to be achieved. Key words: Speech recognition;