一种基于MELP模型600bps声码器的设计-豆柴文库

您所在位置：网站首页 / 一种基于MELP模型600bps声码器的设计.docx / 文档详情

在线预览结束，喜欢就下载吧，查找使用更方便

5 金币

下载文档

/ 2

下载提示

如果您无法下载资料，请参考说明：

1、部分资料下载需要金币，请确保您的账户上有足够的金币

2、已购买过的文档，再次下载不重复扣费

3、资料包下载后请先用软件解压，在使用对应软件打开

文本预览

一种基于MELP模型600bps声码器的设计
Abstract：
Thedesignofa600bpsspeechcoderbasedonMELPmodelisintroducedinthispaper.Thisspeechcodercanmeettherequirementsofreal-timevoicecommunicationinlow-bandwidthandlow-powerenvironments.Firstly,themainprinciplesandtheoriesofspeechcodingbasedonMELPmodelarestudied.Secondly,thehardwareandsoftwarearchitectureofthecoderaredesigned.Finally,theexperimentalresultsofthespeechcoderareanalyzedandevaluated,showingthattheperformanceofthecodecmeetsthedesignrequirements.
1.Introduction
Speechcodingisanimportanttechnologyinthefieldofdigitalsignalprocessing.Withthedevelopmentofcommunicationtechnology,thedemandforvoicecodingwithhighquality,lowrateandlowpowerconsumptionisincreasing.Inthispaper,a600bpsspeechcoderbasedonMELPmodelisdesigned.Comparedwithotherspeechcodingalgorithms,MELP(MixedExcitationLinearPrediction)modelcanachievebetterperformanceunderlowbitratecoding.
2.PrinciplesandTheoriesofMELPModel
MELPmodelisawidelyusedspeechcodingalgorithm.Itisbasedontheanalysis-by-synthesisprinciple,whichmeansthatthespeechsignalissynthesizedfromasetofmodelparameters.ThemaincomponentsoftheMELPmodelareexcitation,linearpredictionfilterandLPCcoefficients,aswellaspitchandgainparameters.
TheexcitationcomponentinMELPmodelincludestwoparts:whitenoiseandperiodicpulsetrain.Theperiodicpulsetrainisgeneratedusingacodebook,andthewhitenoiseisgeneratedusingafilter.Thelinearpredictionfiltermodelsthespectralenvelopeofvoicedspeech,andtheLPCcoefficientsareusedtoestimatethefiltercoefficients.Thepitchparameterisusedtoestimatethefundamentalfrequencyofthevoicedspeech,andthegainparameterisusedtoadjusttheamplitudeoftheexcitationsignal.
3.HardwareandSoftwareArchitectureofMELPCoder
ThehardwareandsoftwarearchitectureoftheMELPcoderisdesignedbasedontheprinciplesandtheoriesofMELPmodel.ThecoreoftheMELPspeechcoderisadigitalsignalprocessor(DSP).TheDSPisusedtoimplementtheMELPmodelandperformsignalprocessing.
Theinputspeechsignalissampledatarateof8kHz,andthenpreprocessedtoremoveDCoffset,normalizetheamplitudeands