[hts-users:02405] Re: Any easy way of getting an adapted model set?

[Hui LIANG <tshlmail-hts@xxxxxxxxx>]

Thank you for the quick reply, Dr. Zen!

However, I didn't succeed in obtaining an adapted model set by the AX command of HHEd. Here is the output message of HHEd. HADAPT:SAVEFULLC=TRUE was ignored by HHEd.

My HHEd command was:
./HTKTools/HHEd -A -D -T 1 -C ./trn.cnf -H ./clustered.cmp.mmf -J . -w ./clustered.cmp.adapted2.mmf ./AX.hed ./context.cmp.tiedlist

Thank you again!

Hui LIANG

----------------------------------------

HHEd
 496517/435372 Models Loaded [7 states max, 2 mixes max]

AX ./csmaplr/SA.cmp.cmllr6
 Setting HMMSet adapt XForm
  Using baseclass macro "dectree_cmp.base" from file ./dectree_cmp.base
  Using xform macro "SA.cmp.cmllr6" from file ./csmaplr/SA.cmp.cmllr6
Attached 232489 XFormInfo structures
Attached 232489 AInfo structures
Created 25 ObsCaches (of 26 possible)
Attached 232489 MInfo structures

Saving new HMM files ...
Edit Complete

HTK Configuration Parameters[26]
  Module/Tool     Parameter                  Value
# HADAPT          SAVEFULLC                   TRUE
  HADAPT          SAVESPKRMODELS              TRUE
# HADAPT          BANDWIDTH       IntVec 3 39 39 39 IntVec 1 0 IntVec 1 0 IntVec 1 0 IntVec 3 4 4 4
# HADAPT          BLOCKSIZE       IntVec 3 40 40 40 IntVec 1 1 IntVec 1 1 IntVec 1 1 IntVec 3 5 5 5
# HADAPT          DURSPLITTHRESH  Vector 5 200.0 200.0 200.0 200.0 200.0
# HADAPT          SPLITTHRESH     Vector 5 2460.0 500.0 500.0 500.0 1000.0
# HADAPT          SMAPSIGMA               1.000000
# HADAPT          USESMAP                     TRUE
# HADAPT          DURTRANSKIND               CMLLR
# HADAPT          TRANSKIND                  CMLLR
# HADAPT          DURUSEBIAS                  TRUE
# HADAPT          USEBIAS                     TRUE
# HADAPT          DURREGTREE      dectree_dur.tree
# HADAPT          REGTREE         dectree_cmp.tree
# HADAPT          DURADAPTKIND                TREE
# HADAPT          ADAPTKIND                   TREE
  HMODEL          TRACE                        512
  HADAPT          TRACE                         61
                  SHRINKOCCTHRESH  Vector 5  100.0 2000.0 2000.0 2000.0 100.0
#                 MAXSTDDEVCOEF                 10
#                 DURVARFLOORPERCENTILE          1.000000
#                 APPLYDURVARFLOOR              TRUE
#                 VFLOORSCALESTR  Vector 5 0.01 0.01 0.01 0.01 0.01
                  APPLYVFLOOR                 TRUE
                  NATURALWRITEORDER              TRUE
                  NATURALREADORDER              TRUE


--- On Thu, 18/2/10, Heiga ZEN (Byung Ha CHUN) <heiga.zen@xxxxxxxxxxxxxxxxx> wrote:

> From: Heiga ZEN (Byung Ha CHUN) <heiga.zen@xxxxxxxxxxxxxxxxx>
> Subject: [hts-users:02404] Re: Any easy way of getting an adapted model set?
> To: tshlmail-hts@xxxxxxxxx
> Cc: hts-users@xxxxxxxxxxxxxxx
> Date: Thursday, 18 February, 2010, 13:30
> Hi,
> 
> Hui LIANG wrote:
> 
> > Just a quick question! I have an average voice model
> set and a set of speaker-specific CMLLR transforms. Is there
> any simple way such that I can obtain an adapted model set
> (in other words, to merge the average voice model set and
> the set of speaker-specific CMLLR transforms)?
> > 
> > I just know that in order to obtain such an adapted
> model set, I can perform one more iteration of CMLLR
> transform estimation with HERest by setting both
> HADAPT:SAVESPKRMODELS and HADAPT:SAVEFULLC to TRUE. But I
> prefer not to do CMLLR transform estimation any more.
> 
> You can use HHEd AX command with HADAPT:SAVEFULLC=TRUE.
> 
> Best regards,
> 
> Heiga ZEN (Byung Ha CHUN)