Speech enhancement on smartphone voice recording

摘要

Speechenhancementischallengingtaskinaudiosignalprocessingtoenhancethequalityoftargetedspeechsignalwhilesuppressothernoises.Inthebeginning,thespeechenhancementalgorithmgrowthrapidlyfromspectralsubtraction,Wienerfiltering,spectralamplitudeMMSEestimatortoNon-negativeMatrixFactorization(NMF).Smartphoneasrevolutionarydevicenowisbeingusedinallaspectoflifeincludingjournalism;personallyandprofessionally.Althoughmanysmartphoneshavetwomicrophones(mainandrear)theonlymainmicrophoneiswidelyusedforvoicerecording.ThisiswhytheNMFalgorithmwidelyusedforthispurposeofspeechenhancement.Thispaperevaluatespeechenhancementonsmartphonevoicerecordingbyusingsomealgorithmsmentionedpreviously.WealsoextendtheNMFalgorithmtoKulback-LeiblerNMFwithsupervisedseparation.ThelastalgorithmshowsimprovedresultcomparedtoothersbyspectrogramandPESQscoreevaluation...