aas@brolga.cc.uq.oz.au (Alex Sergejew) (05/26/91)
Gidday! Back on Thu, 2 May 1991 11:34:25 GMT I asked the net: > I'm aware that there is an extensive literature on efficiently estimating > state transitions and unknown parameters of hidden and semi markov signal > models from time series data, primarily used in the speech processing > context, but have been making heavy weather of translating the basic > references I've found into useful working code. > > Could anyone please point me in the right direction, whether it be > algorithmic descriptions or (preferably) PD source code. If there is enough > interest I will gladly post a summary. The following were also interested: David Haines <haines%hobart@cs.umass.edu> Robert Wylie <wylie@watnow.uwaterloo.ca> Krishna Nanda <nanda@linc.cis.upenn.edu> Sven.Koenig@a.gp.cs.cmu.edu Wolfgang Nejdl <vexpert!nejdl@relay.eu.net> Rick@vee.lrz-muenchen.de and my thanks to the following whose replies, whilst pointing to helpful references (most of which I was already aware of), unfortunately did not pin down any working code: Mou-Yen Chen <mychen@cs.buffalo.edu> Len Moskowitz <moskowit@paul.rutgers.edu> Robert Goldman <rpg@rex.cs.tulane.edu> I append extracts from their replies. I have informally been reassured that semi Markov and HMM algorithms are not all that difficult in themselves, once one gets past the notation in which they are described. If anyone *does* have more information (or code) I'm sure we'd all love to hear about it! Alex. _--_|\ Alex A Sergejew Internet: aas@cc.uq.OZ.AU / X University of Queensland Voice: +61-7-271-8298 \_.--._/ Brisbane, Qld, Australia Fax: +61-7-271-8567 +-+-+-+-+-+-+-+ >From: Mou-Yen Chen <mychen@cs.buffalo.edu> >Message-Id: <9105030737.AA20871@sirius.cs.buffalo.edu> >In-Reply-To: <1991May2.113425.802@brolga.cc.uq.oz.au> >Organization: CEDAR, SUNY at Buffalo We are using HMM(Hidden Markov Model) in hand-written word recognition. As my knowledge, the direct way to implement MM is the Viterbi Algorithm which is the statistical version of DP(Dynamic Programming). Maybe these references can help you a little: (1) A. Kundu, Yang He and P. Bahl, "Recognition of Handwritten Word: First and Second Order Hidden Markov Model Based Approach", Pattern Recognition, Vol. 22, No. 3, pp.283-297, 1989 (2) L.R. Rabiner and B.H. Juang, "An Introduction to Hidden Markov Model", IEEE ASSP Magazine, Vol. 3, No. 1, Jan.1986, pp.4-16 (3) G. David Forney, JR. "The Viterbi Algorithm", Proc. IEEE, Vol.61, No.3, March 1973. +-+-+-+-+-+-+-+ >From: Len Moskowitz <moskowit@paul.rutgers.edu> >Message-Id: <9105031826.AA17586@paul.rutgers.edu> >In-Reply-To: USENET article <1991May2.113425.802@brolga.cc.uq.oz.au> It's not Markov Models but there was some work on something similar that's embodied in a product called AIM (Abductory Induction Mechanism) available from AbTech Corp (700 Harris St., Charlottesville, VA 22901 USA, phone: 804-977-0686). +-+-+-+-+-+-+-+ >From: Robert Goldman <rpg@rex.cs.tulane.edu> >Message-Id: <9105032022.AA21658@rex.cs.tulane.edu> >In-Reply-To: aas@brolga.cc.uq.oz.au's message of 2 May 91 11:34:25 GMT ... the best presentation of the Viterbi algorithm I've found for presentation to students is the survey article by Forney, as follows: @ARTICLE{Forney, AUTHOR = {G. David Forney}, Title = {The Viterbi Algorithm}, JOURNAL = {Proceedings of the IEEE}, YEAR = {1973}, VOLUME = {61}, NUMBER = {3}, PAGES = {268--278} } I have not found a comparably detailed (i.e., with pseudo-Algol) treatment of the forward-backward algorithm for HMMs. +-+-+-+-+-+-+-+