基于ICA的语音信号盲分离[摘要]语音信号盲分离处理的含义是指利用BSS技术对一段语音信号进行处理。混合语音信号的分离是盲分离的重要内容，目前的混和语音分离大多是建立在无噪环境中的混叠情形下，主要以盲源分离(BlindSourceSeparation，BSS)，根据信号的统计特性从几个观测信号中恢复出未知的独立源成分。本文重点研究了以语音信号为背景的盲源分离，在语音和听觉信号处理领域中，如何从混有噪声的的混叠语音信号中分离出各个语音源信号，来模仿人类的语音分离能力，成为一个重要的研究问题。具体实现主要结合ICA技术，将语音去噪作为一个预处理过程，对带噪声的混叠语音盲分离进行了研究，本文详细了介绍了FastICA算法，将这种算法应用于实际的语音信号噪声分离中，并将分离出的语音信号与混合前的原信号进行了分析比较验证了通过ICA实现语音信号的盲分离是切实可行的。[关键词]语音信号，盲源分离，独立成分[Abstract]BlindseparationofspeechsignalsprocessingmeansforprocessingreferstoasectionofthespeechsignalofmicrophonedetectedbyBSStechnique.Separatethemixedspeechsignalsisanimportantcontentofblindseparation,themixedspeechseparationismostlybasedonnoisefreeenvironmentintheoverlappingcase,mainlytotheblindsourceseparation(BlindSourceSeparation,BSS),accordingtothestatisticalcharacteristicsofthesignalfromtheobservedsignalsrecoverindependentsourcecomponentisunknown.Thispaperfocusesontheblindsourceseparationusingspeechsignalasthebackground,inspeechandaudiosignalprocessingfield,separatingeachvoicesourcesignalfromnoisymixedspeechsignals,tomimichumanspeechseparationability,hasbecomeanimportantresearchquestion.ConcreterealizationmainlywithICAtechnology,thespeechdenoisingasapretreatmentprocess,theoverlappingspeechblindseparationofmixedwithnoiseisstudied,thispaperpresentstheFastICAalgorithm,thespeechsignalnoiseseparationofthisalgorithmisappliedtothepractice,andtheoriginalvoicesignalmixedwithisolatedanteriorareanalyzedandcompared,verifiedbyICAtorealizetheblindseparationofspeechsignalsisfeasible.[Keywords]speechsignal,blindsourceseparation,independentcomponentanalysis目录1前言....................................................................11.1盲语音信号分离技术的背景及意义.........................................11.2语音的特性.............................................................22语音信号特性及分析........................................................22.1语音的基本特征........................................................22.2语音处理的理论基础.....................................................32.3语音信号的MATLAB应用程序..............................................42.3.1输入语言的MATLBA时域和频谱图程序分析.............................42.3.2混合语音信号的MATLBA时域和频谱图程序分析.........................83盲信号处理...............................................................103.1盲信号处理的基本概念.................................................103.2盲信号处理的方法和分类................................................103.3盲信号处理技术的研究应用..............................................103.4独立成分分析分析......................................................113.4.1独立成分分析的定义................................................113.4.2ICA的基本原理....................................................124FASTICA算法..............................................................144.1数据的预处理..........................................................144.2FASTICA算法..........................................................
