資源描述:
《畢設(shè)論文 話說(shuō)人識(shí)別的系統(tǒng)設(shè)計(jì).doc》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫(kù)。
1、北京科技大學(xué)本科生畢業(yè)設(shè)計(jì)(論文)摘 要說(shuō)話人識(shí)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(huì)議中的一項(xiàng)重要內(nèi)容。說(shuō)話人識(shí)別的目的是通過(guò)話語(yǔ)找出或核實(shí)說(shuō)話人的身份,可以被用于訪問(wèn)控制。它屬于語(yǔ)音信號(hào)領(lǐng)域的一個(gè)模式識(shí)別問(wèn)題。本文使用交疊分幀的方法短時(shí)化語(yǔ)音信號(hào),使用每幀信號(hào)的能頻值區(qū)分語(yǔ)音信號(hào)和噪聲信號(hào)。特征提取方面,本文使用線性預(yù)測(cè)倒譜系數(shù)和基音頻率來(lái)表征生成語(yǔ)音的發(fā)音器官的差異(先天的),用差分線性預(yù)測(cè)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(shuō)話人。分類(lèi)決策方面,本文使用矢量量
2、化的方法完成對(duì)說(shuō)話人語(yǔ)音信號(hào)的分類(lèi)和判決。本文設(shè)計(jì)的系統(tǒng)是基于Java語(yǔ)言和SQLServer2000數(shù)據(jù)庫(kù)實(shí)現(xiàn)的。Java語(yǔ)言用于實(shí)現(xiàn)語(yǔ)音樣本采集、預(yù)處理、特征提取、分類(lèi)決策等說(shuō)話人識(shí)別所需的各種算法。SQLServer2000數(shù)據(jù)庫(kù)用于存儲(chǔ)已注冊(cè)說(shuō)話人的語(yǔ)音碼本。關(guān)鍵詞:說(shuō)話人識(shí)別;基音;線性預(yù)測(cè);矢量量化本文在實(shí)現(xiàn)系統(tǒng)的基礎(chǔ)上,分析了組合特征中各分量對(duì)說(shuō)話人識(shí)別的貢獻(xiàn)大小。得到的結(jié)論是:用于說(shuō)話人識(shí)別的參數(shù)中,線性預(yù)測(cè)倒譜系數(shù)效果最好,差分線性預(yù)測(cè)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(guò)調(diào)整組合特征中各分量加權(quán)系數(shù)的方式
3、突出貢獻(xiàn)大的分量。實(shí)驗(yàn)表明,調(diào)整后系統(tǒng)識(shí)別率顯著提高。對(duì)于10名男性語(yǔ)音的碼本庫(kù),本文實(shí)現(xiàn)系統(tǒng)的識(shí)別率可達(dá)到87%。3--北京科技大學(xué)本科生畢業(yè)設(shè)計(jì)(論文)Theresearchofthetext-independentspeakerrecognitionsystemAbstractSpeakerrecognitiontechnologyisoneoftheimportantbiometricways,aswellasanimportantpartinacademicconferencesofidentification.Thepurposeofspeaker
4、recognitionisidentifyingorverifyingthespeaker'sidentitythroughthediscourse,whichcanbeusedtocontrolaccess.Itisapatternrecognitionproblemonspeechsignals.Thispaperusesthewayofoverlappingsub-frametoshortthevoicesignal,andusestheEnergyFrequencyValueofeachframetoseparatethevoicesignalfrom
5、thenoisesignal.Intherespectoffeatureextraction,thispaperusesLPCCandpitchfrequencytocharacterizepronunciationorgansgeneratedvoicesounds(congenital),andusesdifferentialLPCCanddifferentialpitchfrequencycharacterizethedifferenceofpronunciationorganmovestopronounce(acquired).Ultimately,a
6、componentfeaturevector,whichisobtainedbyweightedandunitedthatfourfeatures,characterizeaparticularspeaker.Intherespectofclassificationanddecision,weusemethodofvectorquantizationtocompletetheclassificationandsentencingforspeakers'speechsignal.Thesystemthatisdesignedbythispaperisachiev
7、edbasedontheJavalanguageandSQLServer2000database.Javalanguageisusedtoimplementalgorithmsneededbyspeakerrecognition,suchasvoicesampling,preprocessing,featureextraction,classificationanddecisionandsoon.SQLServer2000databaseisusedtostoreregisteredspeakers'voicecodebooks.KeyWords:speake
8、rrecognition;pitch;