摘 要

语音识别主要是让机器听懂人说的话，即在各种情况下，准确地识别出语音的内容，从而根据其信息执行人的各种意图。语音识别技术既是国际竞争的一项重要技术，也是每一个国家经济发展不可缺少的重要技术支撑。本文基于语音信号产生的数学模型，从时域、频域出发对语音信号进行分析，论述了语音识别的基本理论。在此基础上讨论了语音识别的五种算法：动态时间伸缩算法(Dynamic Time Warping，DTW)、基于规则的人工智能方法、人工神经网络(Artificial Neural Network，ANN)方法、隐马尔可夫(Hidden Markov Model，HMM)方法、HMM和ANN的混合模型。重点是从理论上研究隐马尔可夫(HMM)模型算法，对经典的HMM模型算法进行改进。

语音识别算法有多种实现方案，本文采取的方法是利用Matlab强大的数学运算能力，实现孤立语音信号的识别。Matlab 是一款功能强大的数学软件，它附带大量的信号处理工具箱为信号分析研究，特别是文中主要探讨的声波分析研究带来极大便利。本文应用隐马尔科夫模型(HMM) 为识别算法，采用MFCC(MEL频率倒谱系数)为主要语音特征参数，建立了一个汉语数字语音识别系统，其中包括语音信号的预处理、特征参数的提取、识别模板的训练、识别匹配算法；同时，提出利用Matlab图形用户界面开发环境设计语音识别系统界面，设计简单，使用方便，系统界面友好。经过统计，识别效果明显达到了预期目标。

关键词：语音识别算法；HMM模型；Matlab；GUI

ABSTRACT

Speech Recognition is designed to allow machines to understand what people say,and accurately identify the contents of voice to execute the intent of people.Speech recognition technology is not only an important internationally competed technology,but also an indispensable foundational technology for the national economic development.Based on the mathematical model from the speech signal,this paper analyze audio signal from the time

I

domain,frequency domain proceeding,and discussed the basic theory of speech recognition technology.Five algorithm are discussed:Dynamic Time Warping(DTW)、Rule-based Artificial

Intelligence,Artificial

Neural

Network(ANN),Hidden

Markov

Model(HMM),HMM combined with ANN.The focus is put in the theoretical studies of Hidden Markov(HMM) model algorithm,and the classical HMM algorithm is improved.

Speech recognition algorithm is realized in various programs,this article taking the method is to use Matlab powerful mathematical operation ability to realize the recognition of speech signal isolation. Matlab is a powerful mathematic software with a mass of toolboxes dealing with signal processing. It gives a terrific shortcut to the research of signal processing,especially the wave analysis. We can characterize the sound with key parameters such as intensity, frequency etc. In this paper, Markov model (HMM) recognition algorithm using MFCC (MEL frequency cepstral coefficients) as the main voice characteristic parameters, the establishment of a Chinese digital speech recognition system, including the preprocessing of the speech signal,the extraction of characteristic parameters the training of the recognition template,identifying matching algorithm;the same time,the use of Matlab graphical user interface development environment designed speech recognition system interface,is designed to be simple,easy to use,friendly interface. Besides,to of the voice recognition is another target.After statistics,recognition result obviously is made out as the expected goal.

Key words：Speech recognition algorithm；HMM model；Matlab；GUI

II

目录

一、前言 ................................................................................................................................. 1

1.1语音识别的发展历史 .............................................................................................. 1 1.2语音识别研究现状 .................................................................................................. 1 1.3语音识别系统的分类 .............................................................................................. 3 1.4语音识别系统的基本构成 ...................................................................................... 3 1.5语音识别技术难点 .................................................................................................. 4 1.6语音识别发展前景 .................................................................................................. 5 二、语音信号分析 ................................................................................................................. 6

2.1语音学知识 .............................................................................................................. 6

2.1.1音素和音节 .................................................................................................. 6 2.1.2汉语的声调 .................................................................................................. 7 2.1.3语音信号产生模型 ...................................................................................... 8 2.2语音信号数字化和预处理 ...................................................................................... 9

2.2.1数字化 .......................................................................................................... 9 2.2.2预加重处理 .................................................................................................. 9 2.2.3防混叠滤波 ................................................................................................ 10 2.2.4加窗处理 .................................................................................................... 11 2.3语音信号的时域分析 ............................................................................................ 12

2.3.1短时能量分析 ............................................................................................ 12 2.3.2短时平均过零率 ........................................................................................ 14 2.3.3短时自相关函数和短时平均幅度差函数 ................................................ 15 2.3.4语音端点检测 ............................................................................................ 17 2.4语音信号的频域分析 ............................................................................................ 17

2.4.1滤波器组法 ................................................................................................ 17 2.4.2傅立叶频谱分析 ........................................................................................ 18

III