【正文】 北京 科技大學(xué)本科生畢業(yè)設(shè)計(jì)（論文） 1 摘 要 說話人識別技術(shù)是一種的重要生物認(rèn)證手段，也是 身份鑒別 學(xué)術(shù)會議中的一項(xiàng)重要內(nèi)容。 說話人識別的目的是通過話語找出或核實(shí)說話人的身份 ，可以被用于訪問控制 。它屬于語音信號領(lǐng)域的一個(gè)模式識別問題。 本文 使用交疊分幀的方法短時(shí)化語音信號，使用每幀信號的能頻值區(qū)分 語音信號 和噪聲信號 。特征提取方面，本文 使用線性預(yù)測倒譜系數(shù)和基音頻率來表征生成語音的發(fā)音器官的差異（先天的），用差分線性預(yù)測倒譜系數(shù)和差分基音頻率表征發(fā)音器官發(fā)音時(shí)動(dòng)作的差異（后天的）。四種特征 加權(quán)擴(kuò)維得到的組合 特征矢量 最終表征了一個(gè)特定的 說話人。 分 類決策方面， 本文使用矢量量化的方法完成對說話人語音信號的分類和判決。 本文設(shè)計(jì)的系統(tǒng)是基于 Java 語言和 SQL Server 2020 數(shù)據(jù)庫實(shí)現(xiàn)的。 Java 語言用于實(shí)現(xiàn)語音樣本采集、預(yù)處理、特征提取、分類決策等說話人識別所需的各種算法。 SQL Server 2020 數(shù)據(jù)庫用于存儲已注冊說話人的語音碼本。 本文 在 實(shí)現(xiàn) 系統(tǒng)的基礎(chǔ)上，分析了 組合特征中各分量對說話人識別的貢獻(xiàn)大小。得到的結(jié)論是：用于說話人識別的參數(shù)中，線性預(yù)測倒譜系數(shù)效果最好，差分線性預(yù)測倒譜系數(shù)次之，基音頻率再次之，差分基音頻率效果最差。 根據(jù)這 一結(jié)論，系統(tǒng)通過調(diào)整組合特征中各分量加權(quán)系數(shù)的方式突出貢獻(xiàn)大的分量。實(shí)驗(yàn)表明，調(diào)整后系統(tǒng)識別率顯著提高。 對于 10 名男性語音的碼本庫， 本文實(shí)現(xiàn)系統(tǒng)的識別率可達(dá)到 87%。 關(guān)鍵詞：說話人識別；基音；線性預(yù)測；矢量量化 北京 科技大學(xué)本科生畢業(yè)設(shè)計(jì)（論文） 1 The research of the textindependent speaker recognition system Abstract Speaker recognition technology is one of the important biometric ways, as well as an important part in academic conferences of identification. The purpose of speaker recognition is identifying or verifying the speaker39。s identity through the discourse, which can be used to control access. It is a pattern recognition problem on speech signals. This paper uses the way of overlapping subframe to short the voice signal, and uses the Energy Frequency Value of each frame to separate the voice signal from the noise signal. In the respect of feature extraction, this paper uses LPCC and pitch frequency to characterize pronunciation ans generated voice sounds (congenital), and uses differential LPCC and differential pitch frequency characterize the difference of pronunciation an moves to pronounce(acquired). Ultimately, a ponent feature vector, which is obtained by weighted and united that four features, characterize a particular speaker. In the respect of classification and decision, we use method of vector quantization to plete the classification and sentencing for speakers39。 speech signal. The system that is designed by this paper is achieved based on the Java language and SQL Server 2020 database. Java language is used to implement algorithms needed by speaker recognition, such as voice sampling, preprocessing, feature extraction, classification and decision and so on. SQL Server 2020 database is used to store registered speakers39。 voice codebooks. Based on the system achieved, this paper analyzed contribution to the identification for each position of the ponent features. The conclusion is: for speaker identification parameters, LPCC is the best, differential LPCC is the second best, the pitch frequency is the third best and differential pitch frequency is the worst. Based on this conclusion, the system gives prominence to the position, which has more contribution to the identification, by adjusting weighted coefficient. After adjustment, experiments show that system identification rate improved significantly. The recognition rate of the system pleted by this paper can reach 87% to the codebook database produced by 10 male voices. Key Words： speaker recognition。 pitch。 linear prediction。 vector quantification 北京 科技大學(xué)本科生畢業(yè)設(shè)計(jì)（論文） 1 目 錄 摘 要 ........................................................................................................................ 1 Abstract ........................................................................................................................... 1 引 言 ........................................................................................................................ 4 1 緒論 ............................................................................................................................ 5 本文利用到的聲學(xué)知識 .................................................................................... 5 說話人識別的分類 ........................................................................................... 6 說話人確認(rèn)和說話人辨認(rèn) ...................................................................... 6 文本有關(guān)、文本無關(guān)和文本提示 ........................................................... 7 本文系統(tǒng)實(shí)現(xiàn)概述 ........................................................................................... 7 2 語音信號預(yù)處理 ......................................................................................................... 9 語音信號分幀 ................................................................................................... 9 語音信號端點(diǎn)檢測 ......................................................................................... 10 本章小結(jié) ......................................................................................................... 13 3 說話人識別的特征提取 ............................................................................................ 14 特征參數(shù)的選取 ............................................................................................. 14 基音特征 ......................................................................................................... 15 自相關(guān)函數(shù) ........................................................................................... 15 基音檢測 ............................................................................................... 16 線性預(yù)測倒譜系數(shù) ......................................................................................... 18 線性預(yù)測分析 ....................................................................................... 18 LPCC 求解 ............................................................................................. 20 差分特征 ......................................................................................................... 22 特征的組合 ..................................................................................................... 22 北京 科技大學(xué)本科生畢業(yè)設(shè)計(jì)（論文） 2 本章小結(jié) ......................................................................................................... 23 4 說話人識別的分類決策 .............................................................................