training with full covariance matrix.

[asr2010]

I was trying to train the AM with full covariance but there is a problem in
the normalization of the variances.

INFO: cmd_ln.c(691): Parsing command line:
C:\Users\mohamed\sphinx\my_work\trainer\an4\bin\norm.exe \
-accumdir C:/Users/mohamed/sphinx/my_work/trainer/an4/bwaccumdir/an4_buff_1 \
-fullvar yes \
-varfn C:\Users\mohamed\sphinx\my_work\trainer\an4\model_parameters\an4.ci_cont_flatinitial\globalvar

Current configuration:

-accumdir C:/Users/mohamed/sphinx/my_work/trainer/an4/bwaccumdir/an4_buff_1,
-constE 3.0 3.000000e+000
-dcountfn
-example no no
-fullvar no yes
-help no no
-inmeanfn
-inmixwfn
-invarfn
-meanfn
-mixwfn
-mmie no no
-oaccumdir
-regmatfn
-tiedvar no no
-tmatfn
-varfn C:\Users\mohamed\sphinx\my_work\trainer\an4\model_parameters\an4.ci_cont_flatinitial\globalvar

INFO: main.c(146): No -mixwfn specified, will skip if any
INFO: main.c(149): No -tmatfn specified, will skip if any
INFO: main.c(152): No -meanfn specified, will skip if any
INFO: main.c(206): Reading and accumulating counts from
C:/Users/mohamed/sphinx/my_work/trainer/an4/bwaccumdir/an4_buff_1
ERROR: "s3acc_io.c", line 368: Unable to access C:/Users/mohamed/sphinx/my_wor
k/trainer/an4/bwaccumdir/an4_buff_1/gauden_counts
INFO: main.c(428): No means or variances to normalize
WARNING: "main.c", line 514: NO reestimated variances seen, but -varfn
specified
Sun Apr 29 11:03:45 2012

[asr2010]

I forgot to say that the the file /an4/bwaccumdir/an4_buff_1/gauden_counts
already exists. For non-full covariance AM, the training is very smooth.