【正文】 ............................................... 5 ...................................................................................... 5 ........................................................................... 6 ........................................................................... 7 ............................................................................................. 7 ...................................................................................... 7 ...................................................................................... 8 ......................................................................................... 8 .................................................................................. 9 .................................................................................. 9 .............................................................................. 11 ...........................................12 .................................................................................13 .................................................................................14 .....................................................................................14 ..............................................................................14 ............................................................................................15 LPCC倒譜系數(shù) ................................................................................15 Mel頻率倒譜系數(shù) ..........................................................................16 三、語音識(shí)別主要算法 ...........................................................................................17 .....................................................................................17 ..........................................................................18 .....................................................................................19 ........................................................................................20 HMM和 ANN的混合模型 ..............................................................................21 四、隱含馬爾可夫模型算法 ....................................................................................23 HMM的基本理論和數(shù)學(xué)描述 .......................................................................23 HMM的三個(gè)基本問題及解決算法 ................................................................24 HMM算法的改進(jìn) ........................................................................................31 河南理工大學(xué)畢業(yè)設(shè)計(jì)（論文）說明書 IV HMM的結(jié)構(gòu)和類型 .....................................................................................33 HMM算法實(shí)現(xiàn)的問題 .................................................................................34 五、基于 Matlab環(huán)境下的語音識(shí)別算法實(shí)現(xiàn) ..........................................................35 .....................................................................................35 Matlab中 HMM算法的實(shí)現(xiàn) .....................................................................36 ........................................................................................36 .................................................................................36 .....................................................................................37 ............................................................................................38 六、結(jié)束語 .............................................................................................................39 回顧 ...........................................................................................................39 展望 ...........................................................................................................39 七、致謝 ................................................................................................................40 參考文獻(xiàn) ................................................................................................................40 河南理工大學(xué)畢業(yè)設(shè)計(jì)（論文）說明書 1 一、 前言 語音識(shí)別的發(fā)展歷史 作為智能計(jì)算機(jī)研究的主導(dǎo)方向和人機(jī)語音通信的關(guān)鍵技術(shù)，語音識(shí)別技術(shù)一直受到各國(guó)科學(xué)界的廣泛關(guān)注。經(jīng)過統(tǒng)計(jì)，識(shí)別效果明顯達(dá)到了預(yù)期目標(biāo)。 Matlab 是一款功能強(qiáng)大的數(shù)學(xué)軟件，它附帶大量的信號(hào)處理工具箱為信號(hào)分析研究，特別是文中主要探討的聲波分析研究帶來極大便利。重點(diǎn)是從理論上研究隱馬爾可夫 (HMM)模型算法，對(duì)經(jīng)典的 HMM 模型算法進(jìn)行改進(jìn)。本文基于語音信號(hào)產(chǎn)生的數(shù)學(xué)模型，從時(shí)域、頻域出發(fā)對(duì)語音信號(hào)進(jìn)行分析，論述了語音識(shí)別的基本理論。河南理工大學(xué)畢業(yè)設(shè)計(jì)（論文）說明書 I 摘 要 語音識(shí)別主要是讓機(jī)器聽懂人說的話，即在各種情況下，準(zhǔn)確地識(shí)別出語音的內(nèi)容，從而根據(jù)其信息執(zhí)行人的各種意圖。語音識(shí)別技術(shù)既是國(guó)際競(jìng)爭(zhēng)的一項(xiàng)重要技術(shù)，也是每一個(gè)國(guó)家經(jīng)濟(jì)發(fā)展不可缺少的重要技術(shù)支撐。在此基礎(chǔ)上討論了語音識(shí)別的五種算法：動(dòng)態(tài)時(shí)間伸縮算法 (Dynamic Time Warping， DTW)、基于規(guī)則的人工智能方法、人工神經(jīng)網(wǎng)絡(luò) (Artificial Neural Network， ANN)方法、隱馬爾 可夫 (Hidden Markov Model，HMM)方法、 HMM 和 ANN 的混合模型。 語音識(shí)別算法有多種實(shí)現(xiàn)方案，本文采取的方法是利用 Matlab 強(qiáng)大的數(shù)學(xué)運(yùn)算能力，實(shí)現(xiàn)孤立語音信號(hào)的識(shí)別。本文應(yīng)用隱馬爾科夫模型 (HMM) 為識(shí)別算法，采用 MFCC(MEL 頻率倒譜系數(shù) )為主要語音特征參數(shù)，建立了一個(gè)漢 語數(shù)字語音識(shí)別系統(tǒng)，其中包括語音信號(hào)的預(yù)處理、特征參數(shù)的提取、識(shí)別模板的訓(xùn)練、識(shí)別匹配算法；同時(shí)，提出利用 Matlab 圖形用戶界面開發(fā)環(huán)境設(shè)計(jì)語音識(shí)別系統(tǒng)界面，設(shè)計(jì)簡(jiǎn)單，使用方便，系統(tǒng)界面友好。 關(guān)鍵詞 ：語音識(shí)別算法； 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 recognition technology is not only an important internationally peted technology,but also an indispensable foundational technology for the national economic on the mathematical model from the speech signal,this paper analyze audio signal from the time domain,frequency domain proceeding,and discussed the basic theory of speech recognition algorithm are discussed:Dynamic Time Warping(DTW)、 Rulebased Artificial Intelligence,Artificial Neural Network(ANN),Hidden Markov Model(HMM),HMM bined with 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