//=================
// Add DFA grammar
//=================
bool cJulius::addGrammar( char *name, char *dictfile, char *dfafile, bool deleteAll )
{
WORD_INFO *winfo;
DFA_INFO *dfa;
RecogProcess *r;
boolean ret;
if ( ! m_recog ) return false;
r = m_recog->process_list;
// load grammar
switch( r->lmvar ) {
case LM_DFA_WORD:
winfo = word_info_new();
dfa = NULL;
ret = loadGrammar( winfo, NULL, dictfile, NULL, r );
if ( ! ret ) {
word_info_free(winfo);
return false;
}
break;
case LM_DFA_GRAMMAR:
winfo = word_info_new();
dfa = dfa_info_new();
ret = loadGrammar( winfo, dfa, dictfile, dfafile, r );
if ( ! ret ) {
word_info_free(winfo);
dfa_info_free(dfa);
return false;
}
}
if ( deleteAll ) {
/* delete all existing grammars */
multigram_delete_all(r->lm);
}
/* register the new grammar to multi-gram tree */
multigram_add(dfa, winfo, name, r->lm);
/* need to rebuild the global lexicon */
/* tell engine to update at requested timing */
schedule_grammar_update(m_recog);
/* make sure this process will be activated */
r->active = 1;
PostMessage(getWindow(), WM_JULIUS, JEVENT_GRAM_UPDATE, 0L);
return true;
}
/**
* <JA>
* @brief 単語辞書をファイルから読み込んでセットアップする.
*
* 辞書上のモノフォン表記からトライフォンへの計算は init_voca() で
* 読み込み時に行われる. このため，辞書読み込み時には，認識で使用する
* 予定のHMM情報を与える必要がある.
*
* N-gram 使用時は，文頭無音単語およぶ文末無音単語をここで設定する.
* また，"-iwspword" 指定時は，ポーズ単語を辞書の最後に挿入する.
*
* </JA>
* <EN>
* @brief Read in word dictionary from a file and setup for recognition.
*
* Monophone-to-triphone conversion will be performed inside init_voca().
* So, an HMM definition data that will be used with the LM should also be
* specified as an argument.
*
* When reading dictionary for N-gram, sentence head silence word and
* tail silence word will be determined in this function. Also,
* when an option "-iwspword" is specified, this will insert a pause
* word at the last of the given dictionary.
*
* </EN>
*
* @param lmconf [in] LM configuration variables
* @param hmminfo [in] HMM definition of each phone in dictionary, for
* phone checking and monophone-to-triphone conversion.
*
* @return the newly created word dictionary structure, or NULL on failure.
*
*/
static WORD_INFO *
initialize_dict(JCONF_LM *lmconf, HTK_HMM_INFO *hmminfo)
{
WORD_INFO *winfo;
JCONF_LM_NAMELIST *nl;
char buf[MAXLINELEN];
int n;
/* allocate new word dictionary */
winfo = word_info_new();
/* read in dictinary from file */
if ( !
#ifdef MONOTREE
/* leave winfo monophone for 1st pass lexicon tree */
init_voca(winfo, lmconf->dictfilename, hmminfo, TRUE, lmconf->forcedict_flag)
#else
init_voca(winfo, lmconf->dictfilename, hmminfo, FALSE, lmconf->forcedict_flag)
#endif
) {
jlog("ERROR: m_fusion: failed to read dictionary, terminated\n");
word_info_free(winfo);
return NULL;
}
/* load additional entries */
for (nl = lmconf->additional_dict_files; nl; nl=nl->next) {
FILE *fp;
if ((fp = fopen(nl->name, "rb")) == NULL) {
jlog("ERROR: m_fusion: failed to open %s\n",nl->name);
word_info_free(winfo);
return NULL;
}
n = winfo->num;
while (getl_fp(buf, MAXLINELEN, fp) != NULL) {
if (voca_load_line(buf, winfo, hmminfo) == FALSE) break;
}
if (voca_load_end(winfo) == FALSE) {
if (lmconf->forcedict_flag) {
jlog("Warning: m_fusion: the error words above are ignored\n");
} else {
jlog("ERROR: m_fusion: error in reading dictionary %s\n", nl->name);
fclose(fp);
word_info_free(winfo);
return NULL;
}
}
if (fclose(fp) == -1) {
jlog("ERROR: m_fusion: failed to close %s\n", nl->name);
word_info_free(winfo);
return NULL;
}
jlog("STAT: + additional dictionary: %s (%d words)\n", nl->name, winfo->num - n);
}
n = winfo->num;
for (nl = lmconf->additional_dict_entries; nl; nl=nl->next) {
if (voca_load_line(nl->name, winfo, hmminfo) == FALSE) {
jlog("ERROR: m_fusion: failed to set entry: %s\n", nl->name);
}
}
if (lmconf->additional_dict_entries) {
if (voca_load_end(winfo) == FALSE) {
jlog("ERROR: m_fusion: failed to read additinoal word entry\n");
word_info_free(winfo);
return NULL;
}
jlog("STAT: + additional entries: %d words\n", winfo->num - n);
}
if (lmconf->lmtype == LM_PROB) {
/* if necessary, append a IW-sp word to the dict if "-iwspword" specified */
if (lmconf->enable_iwspword) {
if (
#ifdef MONOTREE
voca_append_htkdict(lmconf->iwspentry, winfo, hmminfo, TRUE)
#else
voca_append_htkdict(lmconf->iwspentry, winfo, hmminfo, FALSE)
#endif
== FALSE) {
jlog("ERROR: m_fusion: failed to make IW-sp word entry \"%s\"\n", lmconf->iwspentry);
word_info_free(winfo);
return NULL;
} else {
jlog("STAT: 1 IW-sp word entry added\n");
}
}
/* set {head,tail}_silwid */
winfo->head_silwid = voca_lookup_wid(lmconf->head_silname, winfo);
if (winfo->head_silwid == WORD_INVALID) { /* not exist */
jlog("ERROR: m_fusion: head sil word \"%s\" not exist in voca\n", lmconf->head_silname);
word_info_free(winfo);
return NULL;
}
winfo->tail_silwid = voca_lookup_wid(lmconf->tail_silname, winfo);
if (winfo->tail_silwid == WORD_INVALID) { /* not exist */
jlog("ERROR: m_fusion: tail sil word \"%s\" not exist in voca\n", lmconf->tail_silname);
word_info_free(winfo);
return NULL;
}
}
return(winfo);
}
/**
* <EN>
* @brief Reload dictionaries.
*
* This function reload dictionaries from file.
*
* This function works only for N-gram LM.
* It also re-create all the recognition process, even if it is grammar-based.
*
* @param recog [i/o] engine instance
* @param lm [i/o] LM instance to reload
*
* @return TRUE on success, or FALSE on error.
*
*/
boolean
j_reload_adddict(Recog *recog, PROCESS_LM *lm)
{
RecogProcess *p, *ptmp;
JCONF_SEARCH *sh;
PROCESS_AM *am, *atmp;
jlog("STAT: *** reloading (additional) dictionary of LM%02d %s\n", lm->config->id, lm->config->name);
/* free current dictionary */
if (lm->winfo) word_info_free(lm->winfo);
if (lm->grammars) multigram_free_all(lm->grammars);
if (lm->dfa) dfa_info_free(lm->dfa);
/* free all current process instanfces */
p = recog->process_list;
while(p) {
ptmp = p->next;
j_recogprocess_free(p);
p = ptmp;
}
recog->process_list = NULL;
/* reload dictionary */
if (lm->lmtype == LM_PROB) {
if ((lm->winfo = initialize_dict(lm->config, lm->am->hmminfo)) == NULL) {
jlog("ERROR: m_fusion: failed to reload dictionary\n");
return FALSE;
}
if (lm->config->ngram_filename_lr_arpa || lm->config->ngram_filename_rl_arpa || lm->config->ngram_filename) {
/* re-map dict item to N-gram entry */
if (make_voca_ref(lm->ngram, lm->winfo) == FALSE) {
jlog("ERROR: m_fusion: failed to map words in additional dictionary to N-gram\n");
return FALSE;
}
#if 0
/* post-fix EOS / BOS uni prob for SRILM */
fix_uniprob_srilm(ngram, winfo);
#endif
}
}
#if 0
/* grammar case: not tested */
if (lm->lmtype == LM_DFA) {
if (lm->config->dfa_filename != NULL && lm->config->dictfilename != NULL) {
/* here add grammar specified by "-dfa" and "-v" to grammar list */
multigram_add_gramlist(lm->config->dfa_filename, lm->config->dictfilename, lm->config, LM_DFA_GRAMMAR);
}
/* load all the specified grammars */
if (multigram_load_all_gramlist(lm) == FALSE) {
jlog("ERROR: m_fusion: some error occured in reading grammars\n");
return FALSE;
}
/* setup for later wchmm building */
multigram_update(lm);
/* the whole lexicon will be forced to built in the boot sequence,
so reset the global modification flag here */
lm->global_modified = FALSE;
}
#endif
/* re-create all recognition process instance */
for(sh=recog->jconf->search_root;sh;sh=sh->next) {
if (j_launch_recognition_instance(recog, sh) == FALSE) {
jlog("ERROR: m_fusion: failed to re-start recognizer instance \"%s\"\n", sh->name);
return FALSE;
}
}
/* the created process will be live=FALSE, active = 1, so
the new recognition instance is dead now but
will be made live at next session */
/* tell engine to update */
recog->process_want_reload = TRUE;
jlog("STAT: *** LM%02d %s additional dictionary reloaded\n", lm->config->id, lm->config->name);
return TRUE;
}
