Thepreviouslyproposedsyllable-synchronousnetworksearch(SSNS)algorithmplaysaveryimprotantroleintheworddecodingofthecontinuousChinesespeechrecognitionandachievessatisfyingperformance.Severalrelatedkeyfactorsthatmayaffecttheoverallworddecodingeffectarecarefullystudiedinthispaper,includingtheperfectingofthevocabulary,thebig-discountTuringre-estimatingoftheN-Gramprobabilities,andthemanagingofthesearchingpathbuffers.Basedonthesediscussions,correspondingapproachestoimpovingtheSSNSalgorithmareproposed.COmparedwiththepreviousversionofSSNSalgorithm,thenewversiondecreasestheChinesecharactererrorrate(CCER)intheworddecodingby42.1%acrossadatabaseconsistingofalargenumberoftestingsentences(syllablestrings).
