Hi, all, I am training a isolated word model by using some vedio features. But the idea is same as audio features. But when I use HInit to initialize my HMM, it shows a error [+2126]. All the information are listed as below:

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

HInit -A -D -T 1 -S trainlist_0(oh).txt -M model/hmm0 -H model/proto/0(oh)hmm_vi_W.txt 0(oh)_vi_W  

 No HTK Configuration Parameters Set 

 Initialising  HMM 0(oh)_vi_W . . .  

 States   :   2  3  4  5  6  7 (width) 

 Mixes  s1:   1  1  1  1  1  1 ( 16  ) 

 Num Using:   0  0  0  0  0  0 

 Parm Kind:  USER 

 Number of owners = 1 

 SegLab   :  NONE 

 maxIter  :  20 

 epsilon  :  0.000100 

 minSeg   :  3 

 Updating :  Means Variances MixWeights/DProbs TransProbs 

  - system is PLAIN 

62 Observation Sequences Loaded 

Starting Estimation Process 

  ERROR [+2126]  ViterbiAlign: No path found in 1'th segment 

 FATAL ERROR - Terminating program HInit 

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

And when I use HCompv to obtain the vFloors file, I find all the variances are 0.

Could anyone tell me what happen and how it comes, please?

By the way, I use a "htkread" matlab program to read my feature file, I find all of them are OK. The data looks perfect! 

Thank you in advance!

--- (Edited on 2/17/2013 9:21 pm [GMT-0600] by tfpeach) ---