[hts-users:02074] FlatCluster error 7120

[Girish Malkarnenkar <girish1m@xxxxxxxxx>]

Dear Sir/Madam,

 

I am running the hts demo after substituting the mgc files with facial parameters and the lf0 files with dummy empty files as my goal is to generate facial parameters after training the system on them. However during the initialization & reestimation of phones I got an error as shown below:

 

HTK Configuration Parameters[9]
  Module/Tool     Parameter                  Value
#                 MINDUR                         5
#                 MAXSTDDEVCOEF                 10
#                 APPLYDURVARFLOOR              TRUE
#                 DURVARFLOORPERCENTILE          1.000000
#                 VFLOORSCALESTR  Vector 4 0.01 0.01 0.01 0.01
#                 MINLEAFOCC                     0
#                 NATURALWRITEORDER              TRUE
#                 NATURALREADORDER              TRUE
#                 APPLYVFLOOR                 TRUE

Initialising  HMM /home/hp/facialexps/proto/ver1/state-5_stream-4_mgc-585_lf0-0.prt . . .
 States   :   2  3  4  5  6 (width)
 Mixes  s1:   1  1  1  1  1 ( 585  )
 ( 585  )
 Num Using:   0  0  0  0  0
 Mixes  s2:   2  2  2  2  2 (  0  )
 ( MSD  <=  0 )
 Num Using:   0  0  0  0  0
 Mixes  s3:   2  2  2  2  2 (  0  )
 ( MSD  <=  0 )
 Num Using:   0  0  0  0  0
 Mixes  s4:   2  2  2  2  2 (  0  )
 ( MSD  <=  0 )
 Num Using:   0  0  0  0  0
 Parm Kind:  USER
 Number of owners = 1
 SegLab   :  @
 maxIter  :  20
 epsilon  :  0.000100
 minSeg   :  1
 Updating :  Means Variances MixWeights/DProbs TransProbs

 - system is PLAIN
96 Observation Sequences Loaded
Starting Estimation Process
  ERROR [+7120]  FlatCluster: Failed to make 2 clusters at iter 0

 FATAL ERROR - Terminating program /cygdrive/c/Users/hp/Desktop/PROJECTPHASE1/FINAL/HTK-3.4/htk/HTKTools/HInit
Error in /cygdrive/c/Users/hp/Desktop/PROJECTPHASE1/FINAL/HTK-3.4/htk/HTKTools/HInit  -A    -C /home/hp/facialexps/configs/trn.cnf -D -T 1 -S /home/hp/facialexps/data/scp/train.scp                -m 1 -u tmvw    -w 3 -H /home/hp/facialexps/models/qst001/ver1/cmp/init.mmf -M /home/hp/facialexps/models/qst001/ver1/cmp/HInit -I /home/hp/facialexps/data/labels/mono.mlf -l @ -o @ /home/hp/facialexps/proto/ver1/state-5_stream-4_mgc-585_lf0-0.prt

 

The HTKBook identifies the solution as reducing the no of clusters requested. Can anyone tell me where I should reduce this? Also what is the cause of this problem apart from less data? Could it be because I used empty lf0 files while creating the cmp files or that I switched off usegv in the config.pm?

 

Yours sincerely

 

Girish Malkarnenkar