文件名称：A new look at discriminative training for hidden Markov models
文件大小：243KB
文件格式：PDF
更新时间：2012-05-31 07:31:32
HiddenMarkov model; Discriminative learning; Minimum
DiscriminativetrainingforhiddenMarkovmodels(HMMs)hasbeenacentralthemeinspeechrecognitionresearchformanyyears.Onemostpopulartechniqueisminimumclassi?cationerror(MCE)training,withtheobjectivefunctioncloselyrelatedtotheempiricalerrorrateandwiththeoptimizationmethodbasedtraditionallyongradientdescent.Inthispaper,weprovideanewlookattheMCEtechnique in two ways. First, we develop a non-trivial framework in which the MCE objective function is re-formulated as a rationalfunction for multiple sentence-level training tokens. Second, using this novel re-formulation, we develop a new optimization methodfor discriminatively estimating HMM parameters based on growth transformation or extended Baum–Welch algorithm. Technicaldetails are given for the use of lattices as a rich representation of competing candidates for the MCE training.ó 2007 Elsevier B.V. All rights reserved