语音去噪技术分类方法与性能评价研究

打开文本图片集

中图分类号：TN912.3 文献标识码：A 文章编号：2096-4706（2025）16-0001-08

Abstract： With the wide application of speech technology in the fields ofcommunication，smart home，medical treatment， railwayoperation and soon，theproblem that speech signal is vulnerable to noise interferencein the processof acquisition andtransmision is becoming more and more prominent.This papersystematicallyreviews thecore methods andresearch progressofspeech denoising technology.Basedonthe statisticalcharacteristicsofnoise，thedenoising methodsaredivided into two categories ∵ stationary noise and non-stationary noise.For stationary noise，the principle，process and limitations of spectral subtraction，Wiener filteringand Minimum Mean SquareError methodareanalyzed.For non-stationary noise，the implementationsateiesandapplicablesenariosofKalmanftermetod，daptiveseaeltioneod，signalspace methodandwavelet hresholddenoisingmethodarediscussed indetail.Inadition，theadvantagesofthejointdenoisingmethod aresummarized，anditispointedoutthatitcaneffctivelydealwithomplexnoiseenvironmentbyfusingmultiplelgorhs. Finaly，thekeyperformanceevaluationindexessuchassignal-tooiseratioandRootMeanSquareErorareintroduced，which provide a theoretical basis for the optimization and selection of denoising algorithms.

Keywords： speech denoising; stationary noise; non-stationary noise; signal-to-noise ratio; Root Mean Square Error

0 引言

在当今数字化社会中，语音技术已广泛融入生活的方方面面。（剩余11299字）