[hts-users:01167] Re: bigger full_all.list was used.

["Heiga ZEN (Byung Ha CHUN)" <zen@xxxxxxxxxxxxxxx>]

Hi,

Umud wrote (2008/02/10 17:33):

> Hi,heiga, i used the big full_all.list,the information is as below

OK, I see.
It indicates that HMM matching in tiedlist generation rather than making unseen models 
dominates the computational cost.
Can you please try the following patch code?
I expect it is not so effective but it can reduce it slightly.
BTW, can you please add -pg option for linking?

Regards,

Heiga ZEN (Byung Ha CHUN)


diff -u -r1.53 HHEd.c
--- HHEd.c      31 Jan 2008 17:56:46 -0000      1.53
+++ HHEd.c      10 Feb 2008 09:17:51 -0000
@@ -840,8 +840,8 @@
    int i;
    StateInfo *ai,*bi;

-   if (a->numStates!=b->numStates ||
-       a->transP!=b->transP) return FALSE;
+   if (a->numStates!=b->numStates) return FALSE;
+
    for (i=2;i<a->numStates;i++) {
       ai=a->svec[i].info; bi=b->svec[i].info;
       if (ai!=bi && (!equivState ||
@@ -5520,7 +5520,7 @@
          if (q->type=='h') {
             hmm=(HLink) q->structure;
             for (i=seen;i!=NULL;i=i->next)
-               if (EquivHMM(hmm,i->owner)) break;
+               if (hmm->transP==i->owner->transP && EquivHMM(hmm,i->owner)) break;
             if (i!=NULL) {
                hmm->hook=i->owner;
                i->owner->nUse++;


-- 
------------------------------------------------
 Heiga ZEN     (in Japanese pronunciation)
 Byung Ha CHUN (in Korean pronunciation)

 Department of Computer Science and Engineering
 Nagoya Institute of Technology
 Gokiso-cho, Showa-ku, Nagoya 466-8555 Japan

 http://www.sp.nitech.ac.jp/~zen
------------------------------------------------