dfa_decode.c

julius version 4.12.about sound recognition.

/**
 * @file   dfa_decode.c
 * 
 * <JA>
 * @brief  淡揭矢恕に答づく肌帽胳徒卢∈妈２パス∷
 *
 * 涂えられた簿棱に滦して·DFA 矢恕惧儡鲁材墙な肌帽胳の礁圭を疯年する. 
 * ただし悸狠には, 鸥倡傅簿棱の徒卢される幌眉フレ〖ム件收の帽胳トレリス
 * 惧に荒っている帽胳のみが鸥倡される. 
 *
 * 矢恕扩腆ではショ〖トポ〖ズは帽胳として淡揭し·そのショ〖トポ〖ズ帽胳の
 * 叫附疤弥を矢恕で回年する. ただし·悸狠の掐蜗ではその鳞年した疤弥に
 * 涩ずしもポ〖ズが掐らないため·帽胳鸥倡においては·
 * 肌帽胳礁圭にショ〖トポ〖ズがある眷圭は·さらにその肌の帽胳礁圭まで斧て
 * 肌帽胳礁圭に崔める. 悸狠にそこにショ〖トポ〖ズが赁掐されるかどうかは·
 * search_bestfirst_main.c で尉荚のスコアを孺秤して冉们する. 
 *
 * 矢恕を脱いる千急借妄インスタンスでは·dfa_firstwords(), dfa_nextwords(),
 * dfa_acceptable(), dfa_eosscore() が妈2パスのメイン簇眶 wchmm_fbs() から
 * 蝗脱される. なお N-gram を脱いる千急借妄インスタンスでは·
 * 洛わりに ngram_decode.c 柒の簇眶が蝗われる. 
 * </JA>
 * 
 * <EN>
 * @brief  Grammar-based word prediction (2nd pass)
 *
 * Given a part-of-sentence hypothesis, these function determine a set of next
 * words allowed to be connected by the grammar.  Actually, only words in the
 * word trellis, which exist around the estimated word-end frame will be
 * expanded.
 *
 * When using DFA grammar, the possible (short) pause insertion point
 * should be explicitly specified in grammar, by defining "short-pause
 * word" in vocabulary and write its appearance in grammar.  Since a
 * short pause will not always appear on the specified point, Julius
 * will consider the skipping of such short pause word for the next
 * word prediction in these functions.  Whether short pause was
 * actually inserted or not in the user input will be determined by
 * score in search_bestfirst_main.c.
 *
 * In recognition process instance with DFA grammar, dfa_firstwords(),
 * dfa_nextwords(), dfa_acceptable() and dfa_eosscore() will be called
 * from main search function wchmm_fbs().  When using N-gram, on the
 * other hand, the corresponding functions in ngram_decode.c will be
 * used instead.  </EN>
 * 
 * @author Akinobu LEE
 * @date   Mon Mar  7 15:31:00 2005
 *
 * $Revision: 1.2 $
 * 
 */
/*
 * Copyright (c) 1991-2007 Kawahara Lab., Kyoto University
 * Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and Technology
 * Copyright (c) 2005-2007 Julius project team, Nagoya Institute of Technology
 * All rights reserved
 */

#include <julius/julius.h>

/** 
 * <JA>
 * 矢恕にしたがって·矢片に儡鲁しうる帽胳を呵介の徒卢帽胳凡として手す. 
 * 
 * @param nw [out] 肌帽胳礁圭の呈羌黎へのポインタ
 * @param peseqlen [in] 掐蜗フレ〖ム墓
 * @param maxnw [in] @a nw の钓される呵络墓
 * @param r [in] 千急プロセスインスタンス
 * 
 * @return 徒卢された帽胳眶 (ただしエラ〖箕は -1 を手す)
 * </JA>
 * <EN>
 * Return initial word set from grammar.
 * 
 * @param nw [out] pointer to hold the resulting next word set
 * @param peseqlen [in] input frame length
 * @param maxnw [in] maximum number of words that can be set in @a nw
 * @param r [in] recognition process instance
 * 
 * @return the number of predicted words, or -1 on error.
 * </EN>
 *
 * @callgraph
 * @callergraph
 * 
 */
int
dfa_firstwords(NEXTWORD **nw, int peseqlen, int maxnw, RecogProcess *r)
{
  DFA_INFO *dfa;
  DFA_ARC *arc;
  MULTIGRAM *m;
  int s, sb, se;
  int cate, iw, ns;
  int num = 0;

  dfa = r->lm->dfa;

  for (m = r->lm->grammars; m; m = m->next) {
    if (m->active) {
      sb = m->state_begin;
      se = sb + m->dfa->state_num;
      for(s=sb;s<se;s++) {
	if ((dfa->st[s].status & INITIAL_S) != 0) { /* from initial state */
	  for (arc = dfa->st[s].arc; arc; arc = arc->next) {	/* for all arc */
	    cate = arc->label;	/* category ID */
	    ns = arc->to_state;	/* next DFA state ID */
	    /* all words within the category is expanded */
	    for (iw=0;iw<dfa->term.wnum[cate];iw++) {
	      nw[num]->id = dfa->term.tw[cate][iw]; /* word ID */
	      nw[num]->next_state = ns; /* next state */
	      nw[num]->can_insert_sp = FALSE; /* short pause should not inserted before this word */
	      nw[num]->lscore = r->config->lmp.penalty2;
	      num++;
	      if (num >= maxnw) return -1; /* buffer overflow */
	    }
	  }
	}
      }
    }
  }

  return num;
}

/** 
 * <JA>
 * 婶尸矢簿棱に滦して·矢恕に骄って肌に儡鲁しうる帽胳凡を手す. 
 * 
 * @param hypo [in] 鸥倡傅の婶尸矢簿棱
 * @param nw [out] 肌帽胳礁圭の呈羌黎へのポインタ
 * @param maxnw [in] @a nw の钓される呵络墓
 * @param r [in] 千急プロセスインスタンス
 * 
 * @return 徒卢された帽胳眶 (ただしエラ〖箕は -1 を手す)
 * </JA>
 * <EN>
 * Given a part-of-sentence hypothesis, returns the next word set defined
 * by DFA grammar.
 * 
 * @param hypo [in] the source part-of-sentene hypothesis
 * @param nw [out] pointer to hold the resulting next word set
 * @param maxnw [in] maximum number of words that can be set in @a nw
 * @param r [in] recognition process instance
 * 
 * @return the number of predicted words, or -1 on error.
 * </EN>
 *
 * @callgraph
 * @callergraph
 * 
 */
int
dfa_nextwords(NODE *hypo, NEXTWORD **nw, int maxnw, RecogProcess *r)
{
  DFA_INFO *dfa;
  DFA_ARC *arc, *arc2;
  int iw,cate,ns,cate2,ns2;
  int num = 0;

  dfa = r->lm->dfa;

  /* hypo->state: current DFA state ID */
  for (arc = dfa->st[hypo->state].arc; arc; arc = arc->next) {/* for all arc */
    cate = arc->label;
    ns = arc->to_state;
    if (dfa->is_sp[cate]) {	/* short pause */
      /* expand one more next (not expand the short pause word itself) */
      for (arc2 = dfa->st[ns].arc; arc2; arc2 = arc2->next) {
	cate2 = arc2->label;
	ns2 = arc2->to_state;
	for (iw=0;iw<dfa->term.wnum[cate2];iw++) {
	  nw[num]->id = dfa->term.tw[cate2][iw];
	  nw[num]->next_state = ns2;
	  nw[num]->can_insert_sp = TRUE;
	  nw[num]->lscore = r->config->lmp.penalty2;
	  num++;
	  if (num >= maxnw) return -1; /* buffer overflow */
	}
      }
    } else {			/* not short pause */
      /* all words within the category is expanded */
      for (iw=0;iw<dfa->term.wnum[cate];iw++) {
	nw[num]->id = dfa->term.tw[cate][iw];
	nw[num]->next_state = ns;
	nw[num]->can_insert_sp = FALSE;

	nw[num]->lscore = r->config->lmp.penalty2;
	num++;
	if (num >= maxnw) return -1; /* buffer overflow */
      }
    }
  }
  return num;
}


/** 
 * <JA>
 * 婶尸矢簿棱が矢恕惧矢として呵姜(减妄材墙)觉轮にあるかどうかを手す. 
 * 
 * @param hypo [in] 婶尸矢簿棱
 * @param r [in] 千急プロセスインスタンス
 * 
 * @return 减妄材墙觉轮にあるとき TRUE 减妄稍材墙なとき FALSE
 * </JA>
 * <EN>
 * Return whether the hypothesis is currently on final state
 * 
 * @param hypo [in] sentence hypothesis
 * @param r [in] recognition process instance
 * 
 * @return TRUE when on final state, or FALSE if not acceptable.
 * </EN>
 *
 * @callgraph
 * @callergraph
 * 
 */
boolean
dfa_acceptable(NODE *hypo, RecogProcess *r)
{
  if (r->lm->dfa->st[hypo->state].status & ACCEPT_S) {
    return TRUE;
  } else {
    return FALSE;
  }
}

/* patch by kashima */
/** 
 * <JA>
 * 肌帽胳铬输がその夸年された儡鲁徒卢爬の涟稿の帽胳トレリス惧に
 * あるかどうかをチェックし·もしあればそのトレリス帽胳へのポインタをセット
 * する. なお呵锑の儡鲁爬はあとで疯まるので·ここでは呵努なトレリス帽胳
 * でなくてよい. 
 * 
 * @param nword [i/o] 肌帽胳铬输 (滦炳するトレリス帽胳へのポインタが
 * セットされる)
 * @param hypo [in] 鸥倡傅簿棱
 * @param r [in] 千急プロセスインスタンス
 * 
 * @return 帽胳トレリス惧の徒卢疤弥烧夺に肌帽胳が赂哼すれば TRUE·赂哼
 * しなければ FALSE を手す. 
 * </JA>
 * <EN>
 * Check if the given nextword exists in the word trellis around the
 * estimated connection time.  If exist, set the pointer to the corresponding
 * trellis word to the nextword.  Since the best connection time will be
 * re-computed later, it need not to be an optimal one.
 * 
 * @param nword [i/o] next word candidate (pointer to the found trellis word
 * will be set)
 * @param hypo [in] source part-of-sentence hypothesis
 * @param r [in] recognition process instance
 * 
 * @return TRUE if the nextword exists on the word trellis around the estimated
 * connection point, or FALSE if not exist.
 * </EN>
 *
 * @callgraph
 * @callergraph
 * 
 */
boolean
dfa_look_around(NEXTWORD *nword, NODE *hypo, RecogProcess *r)
{
  int t,tm;
  int i;
  WORD_ID w;
  BACKTRELLIS *bt;
  int lookup_range;

  bt = r->backtrellis;
  lookup_range = r->config->pass2.lookup_range;
  
  tm = hypo->estimated_next_t;	/* estimated connection time */

  /* look aound [tm-lookup_range..tm+lookup_range] frame */
  /* near the center is better:
     1. the first half (backward)   2. the second half (forward) */
  /* 1. backward */
  for(t = tm; t >= tm - lookup_range; t--) {
    if (t < 0) break;
     for (i=0;i<bt->num[t];i++) {
       w = (bt->rw[t][i])->wid;
       if(w == nword->id){	/* found */
         nword->tre = bt->rw[t][i];
         return TRUE;
       }
     }
  }
  /* 2. forward */
  for(t = tm + 1; t < tm + lookup_range; t++) {
    if (t > bt->framelen - 1) break;
    if (t >= hypo->bestt) break;
    for (i=0;i<bt->num[t];i++) {
      w = (bt->rw[t][i])->wid;
      if(w == nword->id){	/* found */
        nword->tre = bt->rw[t][i];
        return TRUE;
      }
    }
  }

  return FALSE;			/* not found */
}

/* end of file */