voca_load_htkdict.c

julius version 4.12.about sound recognition.

  return(ret);
}

/** 
 * Append a single entry to the existing word dictionary.
 * 
 * @param entry [in] dictionary entry string to be appended.
 * @param winfo [out] pointer to word dictionary to append the data.
 * @param hmminfo [in] HTK %HMM definition data.  if NULL, phonemes are ignored.
 * @param ignore_tri_conv [in] TRUE if triphone conversion is ignored
 * 
 * @return TRUE on success, FALSE on any error word.
 */
boolean
voca_append_htkdict(char *entry, WORD_INFO *winfo, HTK_HMM_INFO *hmminfo, boolean ignore_tri_conv)
{
  voca_load_line(entry, winfo, hmminfo);
  return(voca_load_end(winfo));
}

/** 
 * Sub function to Add a dictionary entry line to the word dictionary.
 * 
 * @param buf [i/o] buffer to hold the input string, will be modified in this function
 * @param vnum_p [in] current number of words in @a winfo
 * @param linenum [in] current line number of the input
 * @param winfo [out] pointer to word dictionary to append the data.
 * @param hmminfo [in] HTK %HMM definition data.  if NULL, phonemes are ignored.
 * @param do_conv [in] TRUE if performing triphone conversion
 * @param ok_flag [out] will be set to FALSE if an error occured for this input.
 * 
 * @return FALSE if buf == "DICEND", else TRUE will be returned.
 */
boolean
voca_load_htkdict_line(char *buf, WORD_ID *vnum_p, int linenum, WORD_INFO *winfo, HTK_HMM_INFO *hmminfo, boolean do_conv, boolean *ok_flag)
{
  char *ptmp, *lp = NULL, *p;
  static char cbuf[MAX_HMMNAME_LEN];
  static HMM_Logical **tmpwseq = NULL;
  static int tmpmaxlen;
  int len;
  HMM_Logical *tmplg;
  boolean pok;
  int vnum;

  vnum = *vnum_p;

  if (strmatch(buf, "DICEND")) return FALSE;

  /* allocate temporal work area for the first call */
  if (tmpwseq == NULL) {
    tmpmaxlen = PHONEMELEN_STEP;
    tmpwseq = (HMM_Logical **)mymalloc(sizeof(HMM_Logical *) * tmpmaxlen);
  }

  /* backup whole line for debug output */
  strcpy(bufbak, buf);
  
  /* GrammarEntry */
  if ((ptmp = mystrtok_quote(buf, " \t\n")) == NULL) {
    jlog("Error: voca_load_htkdict: line %d: corrupted data:\n> %s\n", linenum, bufbak);
    winfo->errnum++;
    *ok_flag = FALSE;
    return TRUE;
  }
  winfo->wname[vnum] = strcpy((char *)mybmalloc2(strlen(ptmp)+1, &(winfo->mroot)), ptmp);

  /* just move pointer to next token */
  if ((ptmp = mystrtok_movetonext(NULL, " \t\n")) == NULL) {
    jlog("Error: voca_load_htkdict: line %d: corrupted data:\n> %s\n", linenum, bufbak);
    winfo->errnum++;
    *ok_flag = FALSE;
    return TRUE;
  }
#ifdef CLASS_NGRAM
  winfo->cprob[vnum] = 0.0;	/* prob = 1.0, logprob = 0.0 */
#endif
  
  if (ptmp[0] == '@') {		/* class N-gram prob */
#ifdef CLASS_NGRAM
    /* word probability within the class (for class N-gram) */
    /* format: classname @classprob wordname [output] phoneseq */
    /* classname equals to wname, and wordname will be omitted */
    /* format: @%f (log scale) */
    /* if "@" not found or "@0", it means class == word */
    if ((ptmp = mystrtok(NULL, " \t\n")) == NULL) {
      jlog("Error: voca_load_htkdict: line %d: corrupted data:\n> %s\n", linenum, bufbak);
      winfo->errnum++;
      *ok_flag = FALSE;
      return TRUE;
    }
    if (ptmp[1] == '\0') {	/* space between '@' and figures */
      jlog("Error: voca_load_htkdict: line %d: value after '@' missing, maybe wrong space?\n> %s\n", linenum, bufbak);
      winfo->errnum++;
      *ok_flag = FALSE;
      return TRUE;
    }
    winfo->cprob[vnum] = atof(&(ptmp[1]));
    if (winfo->cprob[vnum] != 0.0) winfo->cwnum++;
    /* read next word entry (just skip them) */
    if ((ptmp = mystrtok(NULL, " \t\n")) == NULL) {
      jlog("Error: voca_load_htkdict: line %d: corrupted data:\n> %s\n", linenum,bufbak);
      winfo->errnum++;
      *ok_flag = FALSE;
      return TRUE;
    }
    /* move to the next word entry */
    if ((ptmp = mystrtok_movetonext(NULL, " \t\n")) == NULL) {
      jlog("Error: voca_load_htkdict: line %d: corrupted data:\n> %s\n", linenum, bufbak);
      winfo->errnum++;
      *ok_flag = FALSE;
      return TRUE;
    }
#else  /* ~CLASS_NGRAM */
    jlog("Error: voca_load_htkdict: line %d: cannot handle in-class word probability\n> %s\n", linenum, ptmp, bufbak);
    winfo->errnum++;
    *ok_flag = FALSE;
    return TRUE;
#endif /* CLASS_NGRAM */
  }

  /* OutputString */
  switch(ptmp[0]) {
  case '[':			/* not transparent word */
    winfo->is_transparent[vnum] = FALSE;
    ptmp = mystrtok_quotation(NULL, " \t\n", '[', ']', 0);
    break;
  case '{':			/* transparent word */
    winfo->is_transparent[vnum] = TRUE;
    ptmp = mystrtok_quotation(NULL, " \t\n", '{', '}', 0);
    break;
  default:
#if 1
    /* ALLOW no entry for output */
    /* same as wname is used */
    winfo->is_transparent[vnum] = FALSE;
    ptmp = winfo->wname[vnum];
#else
    /* error */
    jlog("Error: voca_load_htkdict: line %d: missing output string??\n> %s\n", linenum, bufbak);
    winfo->errnum++;
    *ok_flag = FALSE;
    return TRUE;
#endif
  }
  if (ptmp == NULL) {
    jlog("Error: voca_load_htkdict: line %d: corrupted data:\n> %s\n", linenum, bufbak);
    winfo->errnum++;
    *ok_flag = FALSE;
    return TRUE;
  }
  winfo->woutput[vnum] = strcpy((char *)mybmalloc2(strlen(ptmp)+1, &(winfo->mroot)), ptmp);
    
  /* phoneme sequence */
  if (hmminfo == NULL) {
    /* don't read */
    winfo->wseq[vnum] = NULL;
    winfo->wlen[vnum] = 0;
  } else {

    /* store converted phone sequence to temporal bufffer */
    len = 0;
      
    if (do_conv) {
      /* convert phoneme to triphone expression (word-internal) */
      cycle_triphone(NULL);
      if ((lp = mystrtok(NULL, " \t\n")) == NULL) {
	jlog("Error: voca_load_htkdict: line %d: word %s has no phoneme:\n> %s\n", linenum, winfo->wname[vnum], bufbak);
	winfo->errnum++;
	*ok_flag = FALSE;
	return TRUE;
      }
      cycle_triphone(lp);
    }

    pok = TRUE;
    for (;;) {
      if (do_conv) {
/*	if (lp != NULL) jlog(" %d%s",len,lp);*/
	if (lp != NULL) lp = mystrtok(NULL, " \t\n");
	if (lp != NULL) p = cycle_triphone(lp);
	else p = cycle_triphone_flush();
      } else {
	p = mystrtok(NULL, " \t\n");
      }
      if (p == NULL) break;

      /* both defined/pseudo phone is allowed */
      tmplg = htk_hmmdata_lookup_logical(hmminfo, p);
      if (tmplg == NULL) {
	/* not found */
	if (do_conv) {
	  /* both defined or pseudo phone are not found */
	  if (len == 0 && lp == NULL) {
	    jlog("Error: voca_load_htkdict: line %d: triphone \"*-%s+*\" or monophone \"%s\" not found\n", linenum, p, p);
	    snprintf(cbuf,MAX_HMMNAME_LEN,"*-%s+* or monophone %s", p, p);
	  } else if (len == 0) {
	    jlog("Error: voca_load_htkdict: line %d: triphone \"*-%s\" or biphone \"%s\" not found\n", linenum, p, p);
	    snprintf(cbuf,MAX_HMMNAME_LEN,"*-%s or biphone %s", p, p);
	  } else if (lp == NULL) {
	    jlog("Error: voca_load_htkdict: line %d: triphone \"%s+*\" or biphone \"%s\" not found\n", linenum, p, p);
	    snprintf(cbuf,MAX_HMMNAME_LEN,"%s+* or biphone %s", p, p);
	  } else {
	    jlog("Error: voca_load_htkdict: line %d: triphone \"%s\" not found\n", linenum, p);
	    snprintf(cbuf,MAX_HMMNAME_LEN,"%s", p);
	  }
	} else {
	  jlog("Error: voca_load_htkdict: line %d: phone \"%s\" not found\n", linenum, p);
	  snprintf(cbuf, MAX_HMMNAME_LEN, "%s", p);
	}
	add_to_error(winfo, cbuf);
	pok = FALSE;
      } else {
	/* found */
	if (len >= tmpmaxlen) {
	  /* expand wseq area by PHONEMELEN_STEP */
	  tmpmaxlen += PHONEMELEN_STEP;
	  tmpwseq = (HMM_Logical **)myrealloc(tmpwseq, sizeof(HMM_Logical *) * tmpmaxlen);
	}
	/* store to temporal buffer */
	tmpwseq[len] = tmplg;
      }
      len++;
    }
    if (!pok) {			/* error in phoneme */
      jlog("Error: voca_load_htkdict: the line content was: %s\n", bufbak);
      winfo->errnum++;
      *ok_flag = FALSE;
      return TRUE;
    }
    if (len == 0) {
      jlog("Error: voca_load_htkdict: line %d: no phone specified:\n> %s\n", linenum, bufbak);
      winfo->errnum++;
      *ok_flag = FALSE;
      return TRUE;
    }
    /* store to winfo */
    winfo->wseq[vnum] = (HMM_Logical **)mybmalloc2(sizeof(HMM_Logical *) * len, &(winfo->mroot));
    memcpy(winfo->wseq[vnum], tmpwseq, sizeof(HMM_Logical *) * len);
    winfo->wlen[vnum] = len;
  }

  vnum++;

  *vnum_p = vnum;
  
  return(TRUE);
}

/** 
 * @brief Convert whole words in word dictionary to word-internal triphone.
 *
 * Normally triphone conversion will be performed directly when reading
 * dictionary file.  This function is for post conversion only.
 * 
 * @param winfo [i/o] word dictionary information
 * @param hmminfo [in] HTK %HMM definition
 * 
 * @return TRUE on success, FALSE on failure.
 */
boolean
voca_mono2tri(WORD_INFO *winfo, HTK_HMM_INFO *hmminfo)
{
  WORD_ID w;
  int ph;
  char *p;
  HMM_Logical *tmplg;
  boolean ok_flag = TRUE;
  
  for (w=0;w<winfo->num;w++) {
    cycle_triphone(NULL);
    cycle_triphone(winfo->wseq[w][0]->name);

    for (ph = 0; ph < winfo->wlen[w] ; ph++) {
      if (ph == winfo->wlen[w] - 1) {
	p = cycle_triphone_flush();
      } else {
	p = cycle_triphone(winfo->wseq[w][ph + 1]->name);
      }
      if ((tmplg = htk_hmmdata_lookup_logical(hmminfo, p)) == NULL) {
	jlog("Error: voca_load_htkdict: word \"%s[%s]\"(id=%d): HMM \"%s\" not found\n", winfo->wname[w], winfo->woutput[w], w, p);
	ok_flag = FALSE;
	continue;
      }
      winfo->wseq[w][ph] = tmplg;
    }
  }
  return (ok_flag);
}

/** 
 * Append one word dictionary to other, for multiple grammar handling.
 * Assumes that the same %HMM definition is used on both word dictionary.
 * 
 * @param dstinfo [i/o] word dictionary
 * @param srcinfo [in] word dictionary to be appended to @a dst
 * @param coffset [in] category id offset in @a dst where the new data
 * should be stored
 * @param woffset [in] word id offset in @a dst where the new data
 * should be stored
 */
boolean
voca_append(WORD_INFO *dstinfo, WORD_INFO *srcinfo, int coffset, int woffset)
{
  WORD_ID n, w;
  int i;

  n = woffset;
  while (n >= dstinfo->maxnum) {
    if (winfo_expand(dstinfo) == FALSE) return FALSE;
  }
  for(w=0;w<srcinfo->num;w++) {
    /* copy data */
    dstinfo->wlen[n] = srcinfo->wlen[w];
    if (srcinfo->wname[w]) dstinfo->wname[n] = strcpy((char *)mybmalloc2(strlen(srcinfo->wname[w])+1, &(dstinfo->mroot)), srcinfo->wname[w]);
    if (srcinfo->woutput[w]) dstinfo->woutput[n] = strcpy((char *)mybmalloc2(strlen(srcinfo->woutput[w])+1, &(dstinfo->mroot)), srcinfo->woutput[w]);
    if (srcinfo->wlen[w] > 0) dstinfo->wseq[n] = (HMM_Logical **)mybmalloc2(sizeof(HMM_Logical *) * srcinfo->wlen[w], &(dstinfo->mroot));
    for(i=0;i<srcinfo->wlen[w];i++) {
      dstinfo->wseq[n][i] = srcinfo->wseq[w][i];
    }
    dstinfo->is_transparent[n] = srcinfo->is_transparent[w];
    /* offset category ID by coffset */
    dstinfo->wton[n] = srcinfo->wton[w] + coffset;
    
    n++;
    if (n >= dstinfo->maxnum) {
      if (winfo_expand(dstinfo) == FALSE) return FALSE;
    }

  }
  dstinfo->num = n;

  /* compute maxwn */
  voca_set_stats(dstinfo);

  return TRUE;
}