XPost: comp.soft-sys.matlab, comp.dsp, sci.stat.math   
   XPost: comp.speech.research   
   From: dkanejiya@gmail.com   
      
   > > Can we estimate or design a N-state Hidden Markov Model provided we   
   > > have a given Pdf (probability density function) or a CDF (Cummulative   
   > > Density function)?   
   >   
   > Not without making more assumptions or more information.  Do you know   
   > how your pdf evolves with time?  If not, you are rather stuck as you   
   > really don't have much of a clue as to how your observed pdf is built up   
   > from the emmissions of each of your N states.   
   >   
   > Of course I can solve your problem by setting N=1.   
      
   Probably he wanted to know how to estimate or design the transition   
   probabilities of an HMM given observation density functions for each   
   of the states. That should be easier - just follow the standard HMM   
   training but do not re-estimate the pdfs. If the algorithm reestimates   
   them and u don't want to change the algo all over again, just reset   
   them to the original constrained values after each iteration. Now your   
   HMM will have the optimized transition probabilities.   
      
   I just performed the operation the other way around. Constraining the   
   transition probabilities and estimating the best observation   
   distributions. It gave better results :)   
      
   [ comp.ai is moderated ... your article may take a while to appear. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)