Title: Smoothing factor in discriminative feature adaptation
Other Titles: Vyhlazovací faktor v diskriminativní adaptaci vektoru pozorování
Authors: Zajíc, Zbyněk
Citation: ZAJÍC, Zbyněk; MACHLICA, Lukáš; MÜLLER, Luděk. Initialization of adaptation by sufficient statistics using phonetic tree. In: IEEE 11th international conference on signal processing. Beijing: IEEE Press, 2012, p. 503-506.
Issue Date: 2010
Publisher: [S. n.]
Document type: článek
URI: http://www.kky.zcu.cz/cs/publications/ZbynekZajic_2010_SMOOTHINGFACTORIN
Keywords: diskriminativní trénování;adaptace;vyhlazovací faktor
Keywords in different language: discriminative training;adaptation;smoothing factor
Abstract in different language: While the ML criterion try to maximize the likelihood of the observation states sequence given the correct transcriptions, the DT criteria reflect the recognition error and try to minimized it. Maximum Mutual Information (MMI) criterion in Yu (2006) is one of the DT possibilities. The main problem consists in the optimization process, where mainly the weak-sense auxiliary function is used. Regrettably, it does not guarantee the convergence of the discriminative criterion. In order to adjust the stability of discriminative criteria a smoothing term is involved. Another criteria are e.g. Minimum Phone Error (MPE) or Minimum Classification Error (MCE).
Rights: © Zbyněk Zajíc
Appears in Collections:Články / Articles (KIV)
Články / Articles (NTIS)

Files in This Item:
File Description SizeFormat 
ZbynekZajic_2010_SMOOTHINGFACTORIN.pdfPlný text56,33 kBAdobe PDFView/Open

Please use this identifier to cite or link to this item: http://hdl.handle.net/11025/16952

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.