multi-gram.c

julius version 4.12.about sound recognition.

 * <JA>
 * 矢恕を铜跟步する. ここでは肌搀构糠箕に
 * 瓤鼻されるようにマ〖クをつけるのみである. 
 * 
 * @param gid [in] 铜跟步したい矢恕の ID
 * @param lm [i/o] 咐胳借妄インスタンス
 * </JA>
 * <EN>
 * Activate a grammar in the grammar list.  The specified grammar
 * will only be marked as to be activated in the next grammar update timing.
 * 
 * @param gid [in] grammar ID to be activated
 * @param lm [i/o] LM processing instance
 *
 * @return 0 on success, -1 on error (when specified grammar not found),
 * of 1 if already active
 * </EN>
 * @callgraph
 * @callergraph
 * @ingroup grammar
 */
int
multigram_activate(int gid, PROCESS_LM *lm)	/* only mark */
{
  MULTIGRAM *m;
  int ret;

  for(m=lm->grammars;m;m=m->next) {
    if (m->id == gid) {
      if (m->hook & MULTIGRAM_DEACTIVATE) {
	ret = 0;
	m->hook &= ~(MULTIGRAM_DEACTIVATE);
	m->hook |= MULTIGRAM_ACTIVATE;
	jlog("STAT: Gram #%d %s: marked active, superceding deactivate\n", m->id, m->name);
      } else {
	if (m->hook & MULTIGRAM_ACTIVATE) {
	  jlog("STAT: Gram #%d %s: already marked active\n", m->id, m->name);
	  ret = 1;
	} else {
	  ret = 0;
	  m->hook |= MULTIGRAM_ACTIVATE;
	  jlog("STAT: Gram #%d %s: marked activate\n", m->id, m->name);
	}
      }
      break;
    }
  }
  if (! m) {
    jlog("WARNING: Gram #%d: not found, activation ignored\n", gid);
    ret = -1;
  }

  return(ret);
}

/** 
 * <JA>
 * 矢恕を痰跟步する. 痰跟步された矢恕は
 * 千急において簿棱鸥倡されない. これによって·グロ〖バル辑今を
 * 浩菇蜜することなく·办箕弄に改」の矢恕をON/OFFできる. 痰跟步した
 * 矢恕は multigram_activate() で浩び铜跟步できる. なおここでは
 * 肌搀の矢恕构糠タイミングで瓤鼻されるようにマ〖クをつけるのみである. 
 * 
 * @param gid [in] 痰跟步したい矢恕のID
 * @param lm [i/o] 咐胳借妄インスタンス
 * </JA>
 * <EN>
 * Deactivate a grammar in the grammar list.  The words of the de-activated
 * grammar will not be expanded in the recognition process.  This feature
 * enables rapid switching of grammars without re-building tree lexicon.
 * The de-activated grammar will again be activated by calling
 * multigram_activate().
 * 
 * @param gid [in] grammar ID to be de-activated
 * @param lm [i/o] LM processing instance
 * 
 * @return 0 on success, -1 on error (when specified grammar not found),
 * of 1 if already inactive
 * </EN>
 * @callgraph
 * @callergraph
 * @ingroup grammar
 */
int
multigram_deactivate(int gid, PROCESS_LM *lm)	/* only mark */
{
  MULTIGRAM *m;
  int ret;

  for(m=lm->grammars;m;m=m->next) {
    if (m->id == gid) {
      if (m->hook & MULTIGRAM_ACTIVATE) {
	ret = 0;
	m->hook &= ~(MULTIGRAM_ACTIVATE);
	m->hook |= MULTIGRAM_DEACTIVATE;
	jlog("STAT: Gram #%d %s: marked deactivate, superceding activate\n", m->id, m->name);
      } else {
	if (m->hook & MULTIGRAM_DEACTIVATE) {
	  jlog("STAT: Gram #%d %s: already marked deactivate\n", m->id, m->name);
	  ret = 1;
	} else {
	  ret = 0;
	  m->hook |= MULTIGRAM_DEACTIVATE;
	  jlog("STAT: Gram #%d %s: marked deactivate\n", m->id, m->name);
	}
      }
      break;
    }
  }
  if (! m) {
    jlog("WARNING: - Gram #%d: not found, deactivation ignored\n", gid);
    ret = -1;
  }

  return(ret);
}

/** 
 * <JA>
 * 矢恕の铜跟步ˇ痰跟步を悸乖する. 
 * 
 * @param lm [i/o] 咐胳借妄インスタンス
 * 
 * @return 痰跟から铜跟へ·あるいは铜跟から痰跟へ觉轮が恃步した矢恕が办つでも
 * あればTRUE, 觉轮が链く恃步しなかった眷圭は FALSE を手す. 
 * </JA>
 * <EN>
 * Execute (de)activation of grammars.
 * 
 * @param lm [i/o] LM processing instance
 * 
 * @return TRUE if at least one grammar has been changed, or FALSE if no
 * grammar has changed its status.
 * </EN>
 */
static boolean
multigram_exec_activate(PROCESS_LM *lm)
{
  MULTIGRAM *m;
  boolean modified;
  
  modified = FALSE;
  for(m=lm->grammars;m;m=m->next) {
    if (m->hook & MULTIGRAM_ACTIVATE) {
      m->hook &= ~(MULTIGRAM_ACTIVATE);
      if (!m->active) {
	jlog("STAT: Gram #%d %s: turn on active\n", m->id, m->name);
      }
      m->active = TRUE;
      modified = TRUE;
    } else if (m->hook & MULTIGRAM_DEACTIVATE) {
      m->hook &= ~(MULTIGRAM_DEACTIVATE);
      if (m->active) {
	jlog("STAT: Gram #%d %s: turn off inactive\n", m->id, m->name);
      }
      m->active = FALSE;
      modified = TRUE;
    }
  }
  return(modified);
}
 
/** 
 * <JA>
 * @brief  グロ〖バル矢恕の构糠
 * 
 * 涟搀钙叫しからの矢恕リストの恃构をチェックする. 
 * リスト面に猴近マ〖クがつけられた矢恕がある眷圭は·その矢恕を猴近し·
 * グロ〖バル辑今を浩菇蜜する. 糠たに纳裁された矢恕がある眷圭は·
 * その矢恕を附哼のグロ〖バル辑今の琐萨に纳裁する. 
 *
 * @param lm [i/o] 咐胳借妄インスタンス
 * 
 * @return 撅に TRUE を手す. 
 * </JA>
 * <EN>
 * @brief  Update  global grammar if needed.
 *
 * This function checks for any modification in the grammar list from
 * previous call, and update the global grammar if needed.
 *
 * If there are grammars marked to be deleted in the grammar list,
 * they will be actually deleted from memory.  Then the global grammar is
 * built from scratch using the rest grammars.
 * If there are new grammars, they are appended to the current global grammar.
 * 
 * @param lm [i/o] LM processing instance
 * 
 * @return TRUE when any of add/delete/active/inactive occurs, or FALSE if
 * nothing modified.
 * </EN>
 * @callgraph
 * @callergraph
 * @ingroup grammar
 */
boolean				/* return FALSE if no gram */
multigram_update(PROCESS_LM *lm)
{
  MULTIGRAM *m;
  boolean active_changed = FALSE;
  boolean rebuild_flag;

  if (lm->lmvar == LM_DFA_GRAMMAR) {
    /* setup additional grammar info of new ones */
    for(m=lm->grammars;m;m=m->next) {
      if (m->newbie) {
	jlog("STAT: Gram #%d %s: new grammar loaded, now mash it up for recognition\n", m->id, m->name);
	/* map dict item to dfa terminal symbols */
	if (make_dfa_voca_ref(m->dfa, m->winfo) == FALSE) {
	  jlog("ERROR: failed to map dict <-> DFA. This grammar will be deleted\n");
	  /* mark as to be deleted */
	  m->hook |= MULTIGRAM_DELETE;
	  continue;
	} 
	/* set dfa->sp_id and dfa->is_sp */
	dfa_find_pause_word(m->dfa, m->winfo, lm->am->hmminfo);
	/* build catergory-pair information */
	jlog("STAT: Gram #%d %s: extracting category-pair constraint for the 1st pass\n", m->id, m->name);
	if (extract_cpair(m->dfa) == FALSE) {
	  jlog("ERROR: failed to extract category pair. This grammar will be deleted\n");
	  /* mark as to be deleted */
	  m->hook |= MULTIGRAM_DELETE;
	}
      }
    }
  }

  rebuild_flag = FALSE;
  /* delete grammars marked as "delete" */
  if (multigram_exec_delete(lm)) { /* some built grammars deleted */
    rebuild_flag = TRUE;	/* needs rebuilding global grammar */
  }
  /* find modified grammar */
  for(m=lm->grammars;m;m=m->next) {
    if (m->hook & MULTIGRAM_MODIFIED) {
      rebuild_flag = TRUE;	/* needs rebuilding global grammar */
      m->hook &= ~(MULTIGRAM_MODIFIED);
    }
  }

  if (rebuild_flag) {
    /* rebuild global grammar from scratch (including new) */
    /* active status not changed here (inactive grammar will also included) */
    /* activate/deactivate hook will be handled later, so just keep it here */
#ifdef MDEBUG
    jlog("STAT: re-build whole global grammar...\n");
#endif
    /* free old if not yet */
    if (lm->dfa != NULL) {
      dfa_info_free(lm->dfa);
      lm->dfa = NULL;
    }
    if (lm->winfo != NULL) {
      word_info_free(lm->winfo);
      lm->winfo = NULL;
    }
    /* concatinate all existing grammars to global */
    for(m=lm->grammars;m;m=m->next) {
      if (lm->lmvar == LM_DFA_GRAMMAR && lm->dfa == NULL) {
	lm->dfa = dfa_info_new();
	dfa_state_init(lm->dfa);
      }
      if (lm->winfo == NULL) {
	lm->winfo = word_info_new();
	winfo_init(lm->winfo);
      }
      if (m->newbie) m->newbie = FALSE;
      if (lm->lmvar == LM_DFA_WORD) {
	/* just append dictionaty */
	m->word_begin = lm->winfo->num;
	if (voca_append(lm->winfo, m->winfo, m->id, m->word_begin) == FALSE) {
	  jlog("ERROR: multi-gram: failed to add dictionary #%d to recognition network\n", m->id);
	  /* mark as delete */
	  m->hook |= MULTIGRAM_DELETE;
	}
      } else {
	if (multigram_append_to_global(lm->dfa, lm->winfo, m) == FALSE) {
	  jlog("ERROR: multi-gram: failed to add grammar #%d to recognition network\n", m->id);
	  /* mark as delete */
	  m->hook |= MULTIGRAM_DELETE;
	}
      }
    }
    /* delete the error grammars if exist */
    if (multigram_exec_delete(lm)) {
      jlog("ERROR: errorous grammar deleted\n");
    }
    lm->global_modified = TRUE;
  } else {			/* global not need changed by the deletion */
    /* append only new grammars */
    for(m=lm->grammars;m;m=m->next) {
      if (m->newbie) {
	if (lm->lmvar == LM_DFA_GRAMMAR && lm->dfa == NULL) {
	  lm->dfa = dfa_info_new();
	  dfa_state_init(lm->dfa);
	}
	if (lm->winfo == NULL) {
	  lm->winfo = word_info_new();
	  winfo_init(lm->winfo);
	}
	if (m->newbie) m->newbie = FALSE;
	if (lm->lmvar == LM_DFA_WORD) {
	  /* just append dictionaty */
	  m->word_begin = lm->winfo->num;
	  if (voca_append(lm->winfo, m->winfo, m->id, m->word_begin) == FALSE) {
	    jlog("ERROR: multi-gram: failed to add dictionary #%d to recognition network\n", m->id);
	    /* mark as delete */
	    m->hook |= MULTIGRAM_DELETE;
	  }
	} else {
	  if (multigram_append_to_global(lm->dfa, lm->winfo, m) == FALSE) {
	    jlog("ERROR: multi-gram: failed to add grammar #%d to recognition network\n", m->id);
	    /* mark as delete */
	    m->hook |= MULTIGRAM_DELETE;
	  }
	}
	lm->global_modified = TRUE;
      }
    }
  }

  /* process activate/deactivate hook */
  active_changed = multigram_exec_activate(lm);

  if (lm->global_modified) {		/* if global lexicon has changed */
    /* now global grammar info has been updated */
    /* check if no grammar */
    if (lm->lmvar == LM_DFA_GRAMMAR) {
      if (lm->dfa == NULL || lm->winfo == NULL) {
	if (lm->dfa != NULL) {
	  dfa_info_free(lm->dfa);
	  lm->dfa = NULL;
	}
	if (lm->winfo != NULL) {
	  word_info_free(lm->winfo);
	  lm->winfo = NULL;
	}
      }
    }
#ifdef MDEBUG
    jlog("STAT: grammar update completed\n");
#endif
  }

  if (lm->global_modified || active_changed) {
    return (TRUE);
  }

  return FALSE;
}

/** 
 * <JA>
 * dfaファイルとdictファイルを粕み哈んで矢恕リストに纳裁する. 
 * 
 * @param dfa_file [in] dfa ファイル叹
 * @param dict_file [in] dict ファイル叹
 * @param lm [i/o] 咐胳借妄インスタンス
 * </JA>
 * <EN>
 * Add grammar to the grammar list specified by dfa file and dict file.
 * 
 * @param dfa_file [in] dfa file name
 * @param dict_file [in] dict file name
 * @param lm [i/o] LM processing instance
 * </EN>
 */
static boolean
multigram_read_file_and_add(char *dfa_file, char *dict_file, PROCESS_LM *lm)
{
  WORD_INFO *new_winfo;
  DFA_INFO *new_dfa;
  char buf[MAXGRAMNAMELEN], *p, *q;
  boolean ret;

  if (dfa_file != NULL) {
    jlog("STAT: reading [%s] and [%s]...\n", dfa_file, dict_file);
  } else {
    jlog("STAT: reading [%s]...\n", dict_file);
  }
  
  /* read dict*/
  new_winfo = word_info_new();

  if (lm->lmvar == LM_DFA_GRAMMAR) {
    ret = init_voca(new_winfo, dict_file, lm->am->hmminfo, 
#ifdef MONOTREE
		    TRUE,
#else 
		    FALSE,
#endif
		    lm->config->forcedict_flag);
    if ( ! ret ) {
      jlog("ERROR: failed to read dictionary \"%s\"\n", dict_file);
      word_info_free(new_winfo);
      return FALSE;
    }
  } else if (lm->lmvar == LM_DFA_WORD) {
    ret = init_wordlist(new_winfo, dict_file, lm->am->hmminfo, 
			lm->config->wordrecog_head_silence_model_name,
			lm->config->wordrecog_tail_silence_model_name,
			(lm->config->wordrecog_silence_context_name[0] == '\0') ? NULL : lm->config->wordrecog_silence_context_name,
			lm->config->forcedict_flag);
    if ( ! ret ) {
      jlog("ERROR: failed to read word list \"%s\"\n", dict_file);
      word_info_free(new_winfo);
      return FALSE;
    }
  }

  new_dfa = NULL;
  if (lm->lmvar == LM_DFA_GRAMMAR) {
    /* read dfa */
    new_dfa = dfa_info_new();
    if (init_dfa(new_dfa, dfa_file) == FALSE) {
      jlog("ERROR: multi-gram: error in reading DFA\n");
      word_info_free(new_winfo);
      dfa_info_free(new_dfa);
      return FALSE;
    }
  }