search_bestfirst_v1.c

julius version 4.12.about sound recognition.

/**
 * @file   search_bestfirst_v1.c
 * 
 * <JA>
 * @brief  妈2パスのViterbi遍换および簿棱スコア纷换 (光庐惹)
 *
 * ここでは·妈2パスにおいて玫瑚面の簿棱のViterbiスコアの构糠遍换·
 * 肌帽胳とのトレリス儡鲁·および簿棱のスコア纷换を乖う簇眶が年盗されて
 * います. 
 *
 * 帽胳儡鲁婶の帽胳粗不燎茨董巴赂拉は·光庐な backscan アルゴリズムに
 * よって乖われます. このファイルで年盗されている簇眶は·config.h において
 * PASS2_STRICT_IWCD が undef であるときに蝗脱されます. 嫡に惧淡が define
 * されているときは·search_bestfirst_v2.c の簇眶が脱いられます. 
 *
 * Backscan アルゴリズムでは·デコ〖ディングの光庐步のため·
 * 肌帽胳とその涟の帽胳の儡鲁爬について·≈帽胳粗不燎コンテキストの觅变借妄∽
 * を乖ないます¨
 * 
 *  -# 糠簿棱の栏喇(next_word())では·肌帽胳の呵稿の不燎の宝コンテキスト
 *     のみが雇胃される. 
 *  -# その帽胳粗の窗链な不燎茨董巴赂拉は·その簿棱がいったんスタックに
 *     掐った稿もう办刨 POP されたときに scan_word() にて猖めて纷换する. 
 *
 * 簿棱栏喇箕にはすべての栏喇簿棱に滦して巴赂纷换を乖なず·あとでスコアが
 * 光く POP された簿棱についてのみ浩纷换を乖ないます. このため借妄が
 * 光庐步されますが·簿棱スコア纷换(next_word())において肌帽胳儡鲁婶尸の
 * 茨董巴赂拉が雇胃されないので, 玫瑚面のスコアに疙汗が栏じる眷圭があります. 
 *
 * 悸刘について:
 * 
 *  -# next_word() では·肌帽胳の呵稿の不燎のみを宝コンテキスト(=鸥倡傅
 *     帽胳の呵介の不燎)を雇胃して恃步させ·トレリス儡鲁爬の叫蜗澄唯を滇める. 
 *  -# scan_word() では·糠帽胳婶尸ともう１つ涟の帽胳の呵介の不燎を恃步
 *     させ·scan する. そのため糠帽胳婶尸だけでなく·そのもう办不燎涟まで
 *     scan の滦据となる. この "1-phoneme backscan" を乖なうため,
 *     称簿棱ノ〖ドは呵姜HMM觉轮の涟羹きスコア (NODEにおける g[]) だけでなく·
 *     その backscan 倡幌爬(もう１つ涟の帽胳の呵介の不燎の缄涟)のスコア
 *     も瘦赂しておく涩妥がある (NODE における g_prev[]). 
 *
 * なお·１不燎のみからなる帽胳では backscan 倡幌爬と帽胳董肠が脚なることを
 * 雇胃する涩妥があるため·悸刘はもう警し剩花になる. 
 * </JA>
 * 
 * <EN>
 * @brief  Viterbi path update and scoring on the second pass (fast version)
 *
 * This file has functions for score calculations on the 2nd pass.
 * It includes Viterbi path update calculation of a hypothesis, calculations
 * of scores and word trellis connection at word expansion.
 * 
 * The cross-word triphone will be computed not at word expansion time,
 * but at later pop up for rapid decoding. This is called "backscan"
 * altgorithm. These functions are enabled when PASS2_STRICT_IWCD is
 * UNDEFINED in config.h.  If defined, "nextscan" functions in
 * search_bestfirst_v2.c are used instead.
 *
 * Here we use "delayed cross-word context handling" method
 * for connection of next word and last word of the hypothesis for
 * speeding up decoding:
 *
 *  -# Only right context of the tail phone in the next word is considered
 *     when generating a new hypothesis (next_word()).
 *     
 *  -# The whole context dependency will be fully computed when the
 *     hypothesis is once pushed to stack and later popped in scan_word().
 *
 * This method avoid computing full context-dependency handling for all
 * generated hypothesis in next_word(), and only re-compute it after
 * primising ones are popped from stack later.  This speeds up decoding.
 * But since the context dependency is not considered in the total hypothesis
 * score (computed in next_word()).
 *
 * The actual implimentation:
 * 
 *  -# In nextword(), the tail phone in the new word is modified considering
 *     the right context (= head phone in the last word of source hypothesis),
 *     and the outprob on the connection point between backtrellis and forward
 *     trellis is computed using the triphone.
 *     
 *  -# In scan_word(), not only the new word but also the head phone in the
 *     previous word should be modified and re-scanned.
 *     To realize this '1-phoneme backscan' procedure, hypothesis nodes
 *     have to keep forward scores not only at the last HMM state (g[] in
 *     NODE), but also at the backscan restart point (= before the head
 *     phone in the previous word, g_prev[] in NODE).
 *
 * Note that the actual implementation becomes a little more complicated
 * to handle 1-phoneme words...
 * </EN>
 * 
 * @author Akinobu Lee
 * @date   Sun Sep 11 23:54:53 2005
 *
 * $Revision: 1.4 $
 * 
 */
/*
 * 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>

#ifndef PASS2_STRICT_IWCD

#undef TCD			///< Define if want triphone debug messages

/**********************************************************************/
/************ 簿棱ノ〖ドの答塑拎侯                         ************/
/************ Basic functions for hypothesis node handling ************/
/**********************************************************************/

#undef STOCKER_DEBUG

#ifdef STOCKER_DEBUG
static int stocked_num = 0;
static int reused_num = 0;
static int new_num = 0;
static int request_num = 0;
#endif

/** 
 * <JA>
 * 簿棱ノ〖ドを悸狠にメモリ惧から豺庶する. 
 * 
 * @param node [in] 簿棱ノ〖ド
 * </JA>
 * <EN>
 * Free a hypothesis node actually.
 * 
 * @param node [in] hypothesis node
 * </EN>
 */
static void
free_node_exec(NODE *node)
{
  if (node == NULL) return;

  free(node->g);
  if (node->g_prev != NULL) free(node->g_prev);
#ifdef GRAPHOUT_PRECISE_BOUNDARY
  if (node->region->graphout) {
    free(node->wordend_frame);
    free(node->wordend_gscore);
  }
#endif

  free(node);
}

/** 
 * <JA>
 * 簿棱ノ〖ドの网脱を姜位してリサイクル脱にストックする
 * 
 * @param node [in] 簿棱ノ〖ド
 * </JA>
 * <EN>
 * Stock an unused hypothesis node for recycle.
 * 
 * @param node [in] hypothesis node
 * </EN>
 *
 * @callgraph
 * @callergraph
 */
void
free_node(NODE *node)
{
  if (node == NULL) return;

  if (node->region->graphout) {
    if (node->prevgraph != NULL && node->prevgraph->saved == FALSE) {
      wordgraph_free(node->prevgraph);
    }
  }

  /* save to stocker */
  node->next = node->region->pass2.stocker_root;
  node->region->pass2.stocker_root = node;

#ifdef STOCKER_DEBUG
  stocked_num++;
#endif
}

/** 
 * <JA>
 * リサイクル脱ノ〖ド呈羌杆を鄂にする.
 *
 * @param s [in] stack decoding work area
 * 
 * </JA>
 * <EN>
 * Clear the node stocker for recycle.
 *
 * @param s [in] stack decoding work area
 * 
 * </EN>
 * @callgraph
 * @callergraph
 */
void
clear_stocker(StackDecode *s)
{
  NODE *node, *tmp;
  node = s->stocker_root;
  while(node) {
    tmp = node->next;
    free_node_exec(node);
    node = tmp;
  }
  s->stocker_root = NULL;

#ifdef STOCKER_DEBUG
  jlog("DEBUG: %d times requested, %d times newly allocated, %d times reused\n", request_num, new_num, reused_num);
  stocked_num = 0;
  reused_num = 0;
  new_num = 0;
  request_num = 0;
#endif
}

/** 
 * <JA>
 * 簿棱をコピ〖する. 
 * 
 * @param dst [out] コピ〖黎の簿棱
 * @param src [in] コピ〖傅の簿棱
 * 
 * @return @a dst を手す. 
 * </JA>
 * <EN>
 * Copy the content of node to another.
 * 
 * @param dst [out] target hypothesis
 * @param src [in] source hypothesis
 * 
 * @return the value of @a dst.
 * </EN>
 * @callgraph
 * @callergraph
 */
NODE *
cpy_node(NODE *dst, NODE *src)
{
  int peseqlen;

  peseqlen = src->region->peseqlen;
  
  dst->next = src->next;
  dst->prev = src->prev;
  memcpy(dst->g, src->g, sizeof(LOGPROB) * peseqlen);
  memcpy(dst->seq, src->seq, sizeof(WORD_ID) * MAXSEQNUM);
#ifdef CM_SEARCH
#ifdef CM_MULTIPLE_ALPHA
  {
    int w;
    for(w=0;w<src->seqnum;w++) {
      memcpy(dst->cmscore[w], src->cmscore[w], sizeof(LOGPROB) * src->region->config->annotate.cm_alpha_num);
    }
  }     
#else
  memcpy(dst->cmscore, src->cmscore, sizeof(LOGPROB) * MAXSEQNUM);
#endif
#endif /* CM_SEARCH */
  dst->seqnum = src->seqnum;
  dst->score = src->score;
  dst->bestt = src->bestt;
  dst->estimated_next_t = src->estimated_next_t;
  dst->endflag = src->endflag;
  dst->state = src->state;
  dst->tre = src->tre;
  if (src->g_prev != NULL) { // ccd_flag == TRUE
    memcpy(dst->g_prev, src->g_prev, sizeof(LOGPROB) * peseqlen);
    dst->last_ph = src->last_ph;
    dst->last_ph_sp_attached = src->last_ph_sp_attached;
    dst->lscore = src->lscore;
  }
  dst->totallscore = src->totallscore;
  dst->final_g = src->final_g;
#ifdef VISUALIZE
  dst->popnode = src->popnode;
#endif

  if (src->region->graphout) {
#ifdef GRAPHOUT_PRECISE_BOUNDARY
    memcpy(dst->wordend_frame, src->wordend_frame, sizeof(short) * peseqlen);
    memcpy(dst->wordend_gscore, src->wordend_gscore, sizeof(LOGPROB) * peseqlen);
#endif
    dst->prevgraph = src->prevgraph;
    dst->lastcontext = src->lastcontext;
#ifndef GRAPHOUT_PRECISE_BOUNDARY
    dst->tail_g_score = src->tail_g_score;
#endif
  }

  return(dst);
}

/** 
 * <JA>
 * 糠たな簿棱ノ〖ドを充り烧ける. もし呈羌杆に笆涟活脱されなくなった
 * ノ〖ドがある眷圭はそれを浩网脱する. なければ糠たに充り烧ける.
 *
 * @param r [in] 千急借妄インスタンス
 * 
 * @return 糠たに充り烧けられた簿棱ノ〖ドへのポインタを手す. 
 * </JA>
 * <EN>
 * Allocate a new hypothesis node.  If the node stocker is not empty,
 * the one in the stocker is re-used.  Otherwise, allocate as new.
 *
 * @param r [in] recognition process instance
 * 
 * @return pointer to the newly allocated node.
 * </EN>
 * @callgraph
 * @callergraph
 */
NODE *
newnode(RecogProcess *r)
{
  NODE *tmp;
  int i;
  int peseqlen;

  peseqlen = r->peseqlen;

#ifdef STOCKER_DEBUG
  request_num++;
#endif
  if ((tmp = r->pass2.stocker_root) != NULL) {
    /* re-use ones in the stocker */
    r->pass2.stocker_root = tmp->next;
#ifdef STOCKER_DEBUG
    stocked_num--;
    reused_num++;
#endif
  } else {
    /* allocate new */
    tmp =(NODE *)mymalloc(sizeof(NODE));
    tmp->g = (LOGPROB *)mymalloc(sizeof(LOGPROB) * peseqlen);
    if (r->ccd_flag) {
      tmp->g_prev = (LOGPROB *)mymalloc(sizeof(LOGPROB) * peseqlen);
    } else {
      tmp->g_prev = NULL;
    }

#ifdef GRAPHOUT_PRECISE_BOUNDARY
    if (r->graphout) {
      tmp->wordend_frame = (short *)mymalloc(sizeof(short) * peseqlen);
      tmp->wordend_gscore = (LOGPROB *)mymalloc(sizeof(LOGPROB) * peseqlen);
    }
#endif
#ifdef STOCKER_DEBUG
    new_num++;
#endif
  }

  /* clear the data */
  /*bzero(tmp,sizeof(NODE));*/
  tmp->next=NULL;
  tmp->prev=NULL;
  tmp->last_ph = NULL;
  tmp->last_ph_sp_attached = FALSE;
  if (r->ccd_flag) {
    if (r->lmtype == LM_PROB) {
      tmp->lscore = LOG_ZERO;
      tmp->totallscore = LOG_ZERO;
    } else if (r->lmtype == LM_DFA) {
      tmp->lscore = 0.0;
      tmp->totallscore = 0.0;
    }
  }
  tmp->endflag = FALSE;
  tmp->seqnum = 0;
  for(i=0;i<peseqlen;i++) {
    tmp->g[i] = LOG_ZERO;
  }
  if (r->ccd_flag) {
    for(i=0;i<peseqlen;i++) {
      tmp->g_prev[i] = LOG_ZERO;
    }
  }
  tmp->final_g = LOG_ZERO;
#ifdef VISUALIZE
  tmp->popnode = NULL;
#endif
  tmp->tre = NULL;

  if (r->graphout) {
    tmp->prevgraph = NULL;
    tmp->lastcontext = NULL;
  }

  tmp->region = r;

  return(tmp);
}


/**********************************************************************/
/************ 涟羹きトレリス鸥倡と锑刨纷换              ***************/
/************ Expand trellis and update forward viterbi ***************/
/**********************************************************************/

/** 
 * <JA>
 * 1帽胳尸のトレリス纷换脱のワ〖クエリアを澄瘦.
 * 
 * @param r [in] 千急借妄インスタンス
 * 
 * </JA>
 * <EN>
 * Allocate work area for trellis computation of a word.
 * 
 * @param r [in] recognition process instance
 * 
 * </EN>
 * @callgraph
 * @callergraph
 */
void
malloc_wordtrellis(RecogProcess *r)
{
  int maxwn;
  StackDecode *dwrk;