摘要
Thepreviouslyproposedsyllable-synchronousnetworksearch(SSNS)algorithmplaysaveryimprotantroleintheworddecodingofthecontinuousChinesespeechrecognitionandachievessatisfyingperformance.Severalrelatedkeyfactorsthatmayaffecttheoverallworddecodingeffectarecarefullystudiedinthispaper,includingtheperfectingofthevocabulary,thebig-discountTuringre-estimatingoftheN-Gramprobabilities,andthemanagingofthesearchingpathbuffers.Basedonthesediscussions,correspondingapproachestoimpovingtheSSNSalgorithmareproposed.COmparedwiththepreviousversionofSSNSalgorithm,thenewversiondecreasestheChinesecharactererrorrate(CCER)intheworddecodingby42.1%acrossadatabaseconsistingofalargenumberoftestingsentences(syllablestrings).
出版日期
2000年05月15日（中国期刊网平台首次上网日期，不代表论文的发表时间）
