基于参数修改的中文歌声合成算法的分析-analysis of chinese song synthesis algorithm based on parameter modification.docxVIP

  • 12
  • 0
  • 约3.91万字
  • 约 49页
  • 2018-05-18 发布于上海
  • 举报

基于参数修改的中文歌声合成算法的分析-analysis of chinese song synthesis algorithm based on parameter modification.docx

基于参数修改的中文歌声合成算法的分析-analysis of chinese song synthesis algorithm based on parameter modification

AbstractThespeechsynthesistechnologyisanimportantresearchcontentinthefieldof Human-computer interaction. Singingvoice synthesis is abranch of the speechsynthesis, and ithasalsobecomeahottopicinrecentyears.InordertogenerateaMandarinsong,theTextToSpeech(TTS)technologyandvoicemodificationtechnologyarecombinedtogetherinthis paper.Firstly,thedifferencesbetweenspeechandsingingvoicesignalareanalyzed.Then,themelodycontrolmodelandspectrummodelwhichbasedonGaussianMixtureModelareestablished.Lastly,asingingvoicesynthesissystemisachievedwithvariabletimbre.Theresultsofourresearchcanrevealinherentdifferencesbetweenthespeechandsingingvoice,andhaveanimportanttheoreticalsignificanceforthespeechsynthesis.In addition,oursystemcanbeappliedtoentertainmentareasandevensomemusiccreation fields.Itmakesthesingingvoicesynthesismoreentertainingandinteresting.Themainresearchresultsandinnovationareasfollows:Firstly,thedifferencesbetweenthespeechandthesingingvoicearecomparedby analyzingtheiracousticparameters.Thespeechandsingingvoiceareallproducedbythesameorgans,whiletheperceivedeffectofthemisdifferent.Thedifferencesareanalyzedinthepaper.Speechmaintransfersemanticcontent,andsingingvoicemaintransferemotions. ThesingingvoicehasrichharmonicanditsHHGstillhashighenergy,andthetrendofthefundamentalfrequencyandotherharmonicareverysmooth.Comparedwithsingingvoice,thespeech’strendsarerelatedwithspeech’stones.Secondly,anewsingingmelodycontrolmodelisconstructed.Accordingthedifferencebetweentheminfundamentalfrequency(F0),theF0controlmodelismainlybasedonvibrato. Themelodycontrolmodelisadoptedtoachievethelyricstosongconversionbymodifyingtheacousticparametersofspeechsignal.TheMOStestshowsthattheaverageMOSscoreof syntheticsongbeforetimbreconversionisabove3.29.Thirdly,asinging’sspectrummodificationmodelisproposedinthispaper,whichisbasedonGMM.ConstructingtheSTRAIGHTspectrummodelbyusingGMM,andthesinging’stimbrecanbeconverted.TheABXtestdemonstratesthattheaccuracycanbeupto 100%inthecaseofk=0or1,anditcanbehigherthan62.5%

您可能关注的文档

文档评论(0)

1亿VIP精品文档

相关文档