[hts-users:01873] Re: bug fix in HGen.c for model-level alignment

["Heiga Zen (Byung Ha CHUN)" <heiga.zen@xxxxxxxxxxxxxxxxx>]

Hi,

Matt Shannon wrote:

> There's a small bug when using model-level alignments during synthesis.
>  The model durations get floored rather than rounded, which can lead to
> quite a difference in a long utterance (at least, noticeable enough that
> I noticed it while graphing a short utterance!)
>
> Diff follows:
>
> ./HTKLib/HGen.c 763
> -            modeldur = (int) (label->end -
> label->start)/genInfo->frameRate;
> +            modeldur = (int) ((label->end -
> label->start)/genInfo->frameRate + 0.5);

A similar but different bug has already been reported by Alexis Moinet 
(hts-users:01538).  This bug was originally reported for hts_engine but 
it also existed in HGen.  To fix this problem, I made a patch code and 
asked Keiichiro & HTS-WG members to test it (I didn't have enough time 
to test it at that time).  We finished testing it several months ago, 
but forgot to report it to the ML.  I think the attached patch code also 
solves your bug as well as Alexis's one.  Can you please test it and 
report whether it works or not?

Best regards,

Heiga ZEN (Byung Ha CHUN)

--
--------------------------
Heiga ZEN (Byung Ha CHUN)
Speech Technology Group
Cambridge Research Lab
Toshiba Research Europe
phone: +44 1223 436975


______________________________________________________________________
This email has been scanned by the MessageLabs Email Security System.
For more information please visit http://www.messagelabs.com/email 
______________________________________________________________________
--- HGen.c      2008-05-30 07:19:14.565302000 +0000
+++ HGen.c      2008-07-22 18:16:47.626796000 +0000
@@ -676,7 +676,7 @@
 }
 
 /* CountDurStat: count duration statistics */
-static void CountDurStat (Vector mean, Vector ivar, float *sum, float *sqr, IntVec sindex)
+static void CountDurStat (DVector mean, DVector ivar, double *sum, double *sqr, IntVec sindex)
 {
    int i,j;
    
@@ -693,8 +693,8 @@
 static void SetStateDurations (GenInfo *genInfo)
 {
    int i,j,k,s,cnt,nStates,modeldur,start=0,tframe=0;
-   float sum,sqr,dur,rho=0.0,diff=0.0;
-   Vector *mean, *ivar;
+   double sum,sqr,dur,rho=0.0,diff=0.0;
+   DVector *mean, *ivar;
    Label *label;
    HLink dm;
    
@@ -704,9 +704,9 @@
       HError(9999,"SetStateDurations: Cannot use model-level alignments in random duration generation");
    
    /* state duration statistics storage */
-   if ((mean = (Vector *) New(genInfo->genMem, genInfo->labseqlen*sizeof(Vector)))== NULL)
+   if ((mean = (DVector *) New(genInfo->genMem, genInfo->labseqlen*sizeof(DVector)))== NULL)
       HError(9905,"SetStateDurations: Cannot allocate memory for mean");
-   if ((ivar = (Vector *) New(genInfo->genMem, genInfo->labseqlen*sizeof(Vector)))== NULL)
+   if ((ivar = (DVector *) New(genInfo->genMem, genInfo->labseqlen*sizeof(DVector)))== NULL)
       HError(9905,"SetStateDurations: Cannot allocate memory for inverse variance");
    mean--; ivar--;
    
@@ -718,17 +718,17 @@
       
       /* # of states in the i-th model */
       nStates = genInfo->hmm[i]->numStates-2;
-      mean[i] = CreateVector(genInfo->genMem, nStates);
-      ivar[i] = CreateVector(genInfo->genMem, nStates);
+      mean[i] = CreateDVector(genInfo->genMem, nStates);
+      ivar[i] = CreateDVector(genInfo->genMem, nStates);
       
       /* set statistics of the i-th state */
       for (s=cnt=1; s<=genInfo->dset->swidth[0]; s++) {
          for (k=1; k<=genInfo->dset->swidth[s]; k++,cnt++) {
             mean[i][cnt] = dm->svec[2].info->pdf[s].info->spdf.cpdf[1].mpdf->mean[k];
             switch(dm->svec[2].info->pdf[s].info->spdf.cpdf[1].mpdf->ckind) {
-            case DIAGC:    ivar[i][cnt] = 1.0 / dm->svec[2].info->pdf[s].info->spdf.cpdf[1].mpdf->cov.var[k]; break;
-            case INVDIAGC: ivar[i][cnt] = dm->svec[2].info->pdf[s].info->spdf.cpdf[1].mpdf->cov.var[k]; break;
-            case FULLC:    ivar[i][cnt] = dm->svec[2].info->pdf[s].info->spdf.cpdf[1].mpdf->cov.inv[k][k]; break;
+            case DIAGC:    ivar[i][cnt] = 1.0 / (double) dm->svec[2].info->pdf[s].info->spdf.cpdf[1].mpdf->cov.var[k]; break;
+            case INVDIAGC: ivar[i][cnt] = (double) dm->svec[2].info->pdf[s].info->spdf.cpdf[1].mpdf->cov.var[k]; break;
+            case FULLC:    ivar[i][cnt] = (double) dm->svec[2].info->pdf[s].info->spdf.cpdf[1].mpdf->cov.inv[k][k]; break;
             }
          }
       }
@@ -760,10 +760,9 @@
             rho = 0.0;
          }
          else {  /* model-level alignment of the i-th label is specified */
-            modeldur = (int) (label->end - label->start)/genInfo->frameRate;
             sum = sqr = 0.0;
             CountDurStat(mean[i], ivar[i], &sum, &sqr, genInfo->sindex[i]);
-            rho = (modeldur-sum)/sqr;
+            rho = (((label->end-label->start)/genInfo->frameRate)-sum)/sqr;
          }
       }
       
@@ -779,7 +778,7 @@
          if (genInfo->durations[i][j]<1)
             genInfo->durations[i][j] = 1;
 
-         diff += dur-(float)genInfo->durations[i][j];
+         diff += dur-(double)genInfo->durations[i][j];
          tframe += genInfo->durations[i][j];
          modeldur += genInfo->durations[i][j];
       }
@@ -801,7 +800,7 @@
 static void GetLabStateDurations (GenInfo *genInfo)
 {
    int i,j,k,tframe=0;
-   float diff=0.0;
+   double diff=0.0;
    Label *label;
    
    /* get state durations from given label sequence */ 
@@ -813,7 +812,7 @@
       
       /* get state duration from label */
       genInfo->durations[j][k] = (int)((label->end - label->start)/genInfo->frameRate+diff+0.5);
-      diff += (label->end - label->start)/genInfo->frameRate - (float)genInfo->durations[j][k];
+      diff += (label->end - label->start)/genInfo->frameRate - (double)genInfo->durations[j][k];
       
       /* count total frame */
       tframe += genInfo->durations[j][k++];