multi-gram.c

julius version 4.12.about sound recognition.

/**
 * @file   multi-gram.c
 * 
 * <JA>
 * @brief  千急脱矢恕の瓷妄
 *
 * このファイルには·千急脱矢恕の粕み哈みと瓷妄を乖う簇眶が崔まれています. 
 * これらの簇眶は·矢恕ファイルの粕み哈み·および称硷デ〖タの
 * セットアップを乖います. 
 *
 * 剩眶矢恕の票箕千急に滦炳しています. 剩眶の矢恕を办刨に粕み哈んで·
 * 事误に千急を乖えます. また·モジュ〖ルモ〖ドでは·クライアントから
 * 千急悸乖面に矢恕を瓢弄に纳裁ˇ猴近したり·办婶尸の矢恕を痰跟步ˇ
 * 铜跟步したりできます. また涂えられた改」の矢恕ごとに千急冯蔡を
 * 叫すことができます. 
 *
 * 涂えられた∈剩眶の∷矢恕は办つのグロ〖バル矢恕として冯圭され,
 * 矢恕の粕み哈みや猴近などの觉轮恃构を乖ったとき·构糠されます. 
 * 冯圭された菇矢惮搂 (DFA) が global_dfa に·胳酌辑今が global_winfo に
 * それぞれロ〖カルに呈羌されます. これらは努磊なタイミングで
 * multigram_build() が钙び叫されたときに·global.h 柒の络拌恃眶 dfa
 * および winfo にコピ〖され·千急借妄において蝗脱されるようになります. 
 * </JA>
 * 
 * <EN>
 * @brief  Management of Recognition grammars
 *
 * This file contains functions to read and manage recognition grammar.
 * These function read in grammar and dictionary, and setup data for
 * recognition.
 *
 * Recognition with multiple grammars are supported.  Julian can read
 * several grammars specified at startup time, and perform recognition
 * with those grammars simultaneously.  In module mode, you can add /
 * delete / activate / deactivate each grammar while performing recognition,
 * and also can output optimum results for each grammar.
 *
 * Internally, the given grammars are composed to a single Global Grammar.
 * The global grammar will be updated whenever a new grammar has been read
 * or deleted.  The syntax rule (DFA) of the global grammar will be stored
 * at global_dfa, and the corresponding dictionary will be at global_winfo
 * locally, independent of the decoding timing.  After that, multigram_build()
 * will be called to make the prepared global grammar to be used in the
 * actual recognition process, by copying the grammar and the dictionary
 * to the global variable dfa and winfo.
 * 
 * @author Akinobu Lee
 * @date   Sat Jun 18 23:45:18 2005
 *
 * $Revision: 1.8 $
 * 
 */
/*
 * 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>

/// For debug: define to enable grammar update messages to stdout
#define MDEBUG

/** 
 * <JA>
 * @brief  グロ〖バル矢恕から腾菇陇步辑今を菇蜜する. 
 *
 * 涂えられた矢恕で千急を乖うために·千急借妄インスタンスが附哼积つ
 * グロ〖バル矢恕から腾菇陇步辑今を∈浩∷菇蜜します. また· 
 * 弹瓢箕にビ〖ム升が汤绩弄に回绩されていない眷圭やフルサ〖チの眷圭·
 * ビ〖ム升の浩肋年も乖います.
 * 
 * @param r [i/o] 千急借妄インスタンス
 * </JA>
 * <EN>
 * @brief  Build tree lexicon from global grammar.
 *
 * This function will re-construct the tree lexicon using the global grammar
 * in the recognition process instance.  If the beam width was not explicitly
 * specified on startup, the the beam width will be guessed
 * according to the size of the new lexicon.
 * 
 * @param r [i/o] recognition process instance
 * </EN>
 */
static boolean
multigram_rebuild_wchmm(RecogProcess *r)
{
  boolean ret;

  /* re-build wchmm */
  if (r->wchmm != NULL) {
    wchmm_free(r->wchmm);
  }
  r->wchmm = wchmm_new();
  r->wchmm->lmtype = r->lmtype;
  r->wchmm->lmvar  = r->lmvar;
  r->wchmm->ccd_flag = r->ccd_flag;
  r->wchmm->category_tree = TRUE;
  r->wchmm->hmmwrk = &(r->am->hmmwrk);
  /* assign models */
  r->wchmm->dfa = r->lm->dfa;
  r->wchmm->winfo = r->lm->winfo;
  r->wchmm->hmminfo = r->am->hmminfo;
  if (r->wchmm->category_tree) {
    if (r->config->pass1.old_tree_function_flag) {
      ret = build_wchmm(r->wchmm, r->lm->config);
    } else {
      ret = build_wchmm2(r->wchmm, r->lm->config);
    }
  } else {
    ret = build_wchmm2(r->wchmm, r->lm->config);
  }

  /* 弹瓢箕 -check でチェックモ〖ドへ */
  if (r->config->sw.wchmm_check_flag) {
    wchmm_check_interactive(r->wchmm);
  }
  
  if (ret == FALSE) {
    jlog("ERROR: multi-gram: failed to build (global) lexicon tree for recognition\n");
    return FALSE;
  }
  
  /* guess beam width from models, when not specified */
  r->trellis_beam_width = set_beam_width(r->wchmm, r->config->pass1.specified_trellis_beam_width);
  switch(r->config->pass1.specified_trellis_beam_width) {
  case 0:
    jlog("STAT: multi-gram: beam width set to %d (full) by lexicon change\n", r->trellis_beam_width);
    break;
  case -1:
    jlog("STAT: multi-gram: beam width set to %d (guess) by lexicon change\n", r->trellis_beam_width);
  }

  /* re-allocate factoring cache for the tree lexicon*/
  /* for n-gram only?? */
  //max_successor_cache_free(recog->wchmm);
  //max_successor_cache_init(recog->wchmm);

  /* finished! */

  return TRUE;
}

/** 
 * <EN>
 * @brief  Check for global grammar and (re-)build tree lexicon if needed.
 * 
 * If any modification of the global grammar has been occured, 
 * the tree lexicons and some other data for recognition will be re-constructed
 * from the updated global grammar.
 * </EN>
 * <JA>
 * @brief  グロ〖バル矢恕を拇べ·涩妥があれば腾菇陇步辑今を∈浩∷菇蜜する. 
 * 
 * グロ〖バル辑今に恃构があれば·その构糠されたグロ〖バル
 * 辑今から腾菇陇步辑今などの不兰千急脱デ〖タ菇陇を浩菇蜜する. 
 * 
 * </JA>
 * 
 * @param r [in] recognition process instance
 * 
 * @return TRUE on success, FALSE on error.
 *
 * @callgraph
 * @callergraph
 * @ingroup grammar
 * 
 */
boolean
multigram_build(RecogProcess *r)
{
  if (r->lm->winfo != NULL) {
    /* re-build tree lexicon for recognition process */
    if (multigram_rebuild_wchmm(r) == FALSE) {
      jlog("ERROR: multi-gram: failed to re-build tree lexicon\n");
      return FALSE;
    }
#ifdef MDEBUG
    jlog("STAT: wchmm (re)build completed\n");
#endif
  }
  return(TRUE);
}

/** 
 * <JA>
 * @brief  グロ〖バル矢恕の琐萨に矢恕を纳裁する. 
 *
 * もとの矢恕菇陇挛には·グロ〖バル矢恕のどの疤弥にその矢恕が纳裁
 * されたか·そのカテゴリ戎规と辑今戎规の认跋が淡峡される. 
 * 
 * @param gdfa [i/o] 冯圭黎の矢恕のDFA攫鼠
 * @param gwinfo [i/o] 冯圭黎の矢恕の辑今攫鼠
 * @param m [i/o] 冯圭する矢恕攫鼠. 
 * </JA>
 * <EN>
 * @brief  Append a grammar to the tail of global grammar.
 *
 * The location of the grammar in the global grammar (categories and words)
 * will be stored to the grammar structure for later access.
 * 
 * @param gdfa [i/o] DFA information of the global grammar
 * @param gwinfo [i/o] Dictionary information of the global grammar
 * @param m [i/o] New grammar information to be installed.
 * </EN>
 */
static boolean
multigram_append_to_global(DFA_INFO *gdfa, WORD_INFO *gwinfo, MULTIGRAM *m)
{
  /* the new grammar 'm' will be appended to the last of gdfa and gwinfo */
  m->state_begin = gdfa->state_num;	/* initial state ID */
  m->cate_begin = gdfa->term_num;	/* initial terminal ID */
  m->word_begin = gwinfo->num;	/* initial word ID */
  
  /* append category ID and node number of src DFA */
  /* Julius allow multiple initial states: connect each initial node
     is not necesarry. */
  dfa_append(gdfa, m->dfa, m->state_begin, m->cate_begin);
  /* append words of src vocabulary to global winfo */
  if (voca_append(gwinfo, m->winfo, m->cate_begin, m->word_begin) == FALSE) {
    return FALSE;
  }
  /* append category->word mapping table */
  terminfo_append(&(gdfa->term), &(m->dfa->term), m->cate_begin, m->word_begin);
  /* append catergory-pair information */
  /* pause has already been considered on m->dfa, so just append here */
  if (cpair_append(gdfa, m->dfa, m->cate_begin) == FALSE) {
    return FALSE;
  }
  /* re-set noise entry by merging */
  if (dfa_pause_word_append(gdfa, m->dfa, m->cate_begin) == FALSE) {
    return FALSE;
  }

  jlog("STAT: Gram #%d %s: installed\n", m->id, m->name);

  return TRUE;
}

/** 
 * <JA>
 * 糠たな矢恕を·矢恕リストに纳裁する.
 * 附哼インスタンスが瘦积している矢恕のリストは lm->grammars に瘦赂される. 
 * 纳裁した矢恕には·newbie, active のフラグがセットされ·肌搀の
 * 矢恕构糠チェック箕に构糠滦据となる. 
 * 
 * @param dfa [in] 纳裁判峡する矢恕のDFA攫鼠
 * @param winfo [in] 纳裁判峡する矢恕の辑今攫鼠
 * @param name [in] 纳裁判峡する矢恕の叹疚
 * @param lm [i/o] 咐胳借妄インスタンス
 *
 * @return 矢恕IDを手す. 
 * </JA>
 * <EN>
 * Add a new grammar to the current list of grammars.
 * The list of grammars which the LM instance keeps currently is
 * at lm->grammars.
 * The new grammar is flaged at "newbie" and "active", to be treated
 * properly at the next grammar update check.
 * 
 * @param dfa [in] DFA information of the new grammar.
 * @param winfo [in] dictionary information of the new grammar.
 * @param name [in] name string of the new grammar.
 * @param lm [i/o] LM processing instance
 *
 * @return the new grammar ID for the given grammar.
 * </EN>
 *
 * @callgraph
 * @callergraph
 * @ingroup grammar
 */
int
multigram_add(DFA_INFO *dfa, WORD_INFO *winfo, char *name, PROCESS_LM *lm)
{
  MULTIGRAM *new;

  /* allocate new gram */
  new = (MULTIGRAM *)mymalloc(sizeof(MULTIGRAM));
  if (name != NULL) {
    strncpy(new->name, name, MAXGRAMNAMELEN);
  } else {
    strncpy(new->name, "(no name)", MAXGRAMNAMELEN);
  }

  new->id = lm->gram_maxid;
  new->dfa = dfa;
  new->winfo = winfo;
  new->hook = MULTIGRAM_DEFAULT;
  new->newbie = TRUE;		/* need to setup */
  new->active = TRUE;		/* default: active */

  /* the new grammar is now added to gramlist */
  new->next = lm->grammars;
  lm->grammars = new;

  jlog("STAT: Gram #%d %s registered\n", new->id, new->name);
  lm->gram_maxid++;

  return new->id;
}

/** 
 * <JA>
 * 矢恕を猴近する. 
 *
 * 矢恕リスト面のある矢恕について·猴近マ〖クを烧ける. 
 * 悸狠の猴近は multigram_exec_delete() で乖われる. 
 * 
 * @param delid [in] 猴近する矢恕の矢恕ID
 * @param lm [i/o] 咐胳借妄インスタンス
 * 
 * @return 奶撅箕 TRUE を手す. 回年されたIDの矢恕が痰い眷圭は FALSE を手す. 
 * </JA>
 * <EN>
 * Mark a grammar in the grammar list to be deleted at the next grammar update.
 * 
 * @param delid [in] grammar id to be deleted
 * @param lm [i/o] LM processing instance
 * 
 * @return TRUE on normal exit, or FALSE if the specified grammar is not found
 * in the grammar list.
 * </EN>
 * @callgraph
 * @callergraph
 * @ingroup grammar
 */
boolean
multigram_delete(int delid, PROCESS_LM *lm)
{
  MULTIGRAM *m;
  for(m=lm->grammars;m;m=m->next) {
    if (m->id == delid) {
      m->hook |= MULTIGRAM_DELETE;
      jlog("STAT: Gram #%d %s: marked delete\n", m->id, m->name);
      break;
    }
  }
  if (! m) {
    jlog("STAT: Gram #%d: not found\n", delid);
    return FALSE;
  }
  return TRUE;
}

/** 
 * <JA>
 * すべての矢恕を肌搀构糠箕に猴近するようマ〖クする.
 * 
 * @param lm [i/o] 咐胳借妄インスタンス
 * </JA>
 * <EN>
 * Mark all grammars to be deleted at next grammar update.
 * 
 * @param lm [i/o] LM processing instance
 * </EN>
 * @callgraph
 * @callergraph
 * @ingroup grammar
 */
void
multigram_delete_all(PROCESS_LM *lm)
{
  MULTIGRAM *m;
  for(m=lm->grammars;m;m=m->next) {
    m->hook |= MULTIGRAM_DELETE;
  }
}

/** 
 * <JA>
 * 猴近マ〖クのついた矢恕をリストから猴近する. 
 * 
 * @param lm [i/o] 咐胳借妄インスタンス
 * 
 * @return グロ〖バル矢恕の浩菇蜜が涩妥なときは TRUE を·稍涩妥なときは FALSE を手す. 
 * </JA>
 * <EN>
 * Purge grammars marked as delete.
 * 
 * @param lm [i/o] LM processing instance
 * 
 * @return TRUE if the global grammar must be re-constructed, or FALSE if not needed.
 * </EN>
 */
static boolean
multigram_exec_delete(PROCESS_LM *lm)
{
  MULTIGRAM *m, *mtmp, *mprev;
  boolean ret_flag = FALSE;

  /* exec delete */
  mprev = NULL;
  m = lm->grammars;
  while(m) {
    mtmp = m->next;
    if (m->hook & MULTIGRAM_DELETE) {
      /* if any grammar is deleted, we need to rebuild lexicons etc. */
      /* so tell it to the caller */
      if (! m->newbie) ret_flag = TRUE;
      if (m->dfa) dfa_info_free(m->dfa);
      word_info_free(m->winfo);
      jlog("STAT: Gram #%d %s: purged\n", m->id, m->name);
      free(m);
      if (mprev != NULL) {
	mprev->next = mtmp;
      } else {
	lm->grammars = mtmp;
      }
    } else {
      mprev = m;
    }
    m = mtmp;
  }

  return(ret_flag);
}

/**