sumom****@users*****
sumom****@users*****
2008年 12月 12日 (金) 16:14:58 JST
Index: julius4/libjulius/src/beam.c
diff -u julius4/libjulius/src/beam.c:1.10 julius4/libjulius/src/beam.c:1.11
--- julius4/libjulius/src/beam.c:1.10 Fri Nov 14 13:12:12 2008
+++ julius4/libjulius/src/beam.c Fri Dec 12 16:14:57 2008
@@ -42,7 +42,7 @@
* @author Akinobu LEE
* @date Tue Feb 22 17:00:45 2005
*
- * $Revision: 1.10 $
+ * $Revision: 1.11 $
*
*/
/*
@@ -583,6 +583,7 @@
LOGPROB maxscore;
int i;
TRELLIS_ATOM **idx;
+ int num;
if (r->lmvar != LM_DFA_WORD) return;
@@ -629,13 +630,14 @@
/* more than one candidate is requested */
/* get actual number of candidates to output */
- r->result.sentnum = r->config->output.output_hypo_maxnum;
- if (r->result.sentnum > bt->num[last_time]) {
- r->result.sentnum = bt->num[last_time];
+ num = r->config->output.output_hypo_maxnum;
+ if (num > bt->num[last_time]) {
+ num = bt->num[last_time];
}
/* prepare result storage */
- r->result.sent = (Sentence *)mymalloc(sizeof(Sentence)* r->result.sentnum);
+ result_sentence_malloc(r, num);
+ r->result.sentnum = num;
/* sort by score */
idx = (TRELLIS_ATOM **)mymalloc(sizeof(TRELLIS_ATOM *)*bt->num[last_time]);
@@ -662,7 +664,6 @@
} else {
s->gram_id = 0;
}
- s->align.filled = FALSE;
}
/* free work area for sort */
free(idx);
@@ -670,7 +671,7 @@
} else { /* only max is needed */
/* prepare result storage */
- r->result.sent = (Sentence *)mymalloc(sizeof(Sentence));
+ result_sentence_malloc(r, 1);
r->result.sentnum = 1;
s = &(r->result.sent[0]);
s->word_num = 1;
@@ -686,11 +687,11 @@
} else {
s->gram_id = 0;
}
- s->align.filled = FALSE;
}
/* copy as 1st pass result */
memcpy(&(r->result.pass1), &(r->result.sent[0]), sizeof(Sentence));
+ r->result.pass1.align = NULL;
//callback_exec(CALLBACK_RESULT, r);
//free(r->result.sent);
Index: julius4/libjulius/src/jfunc.c
diff -u julius4/libjulius/src/jfunc.c:1.4 julius4/libjulius/src/jfunc.c:1.5
--- julius4/libjulius/src/jfunc.c:1.4 Thu Sep 25 14:00:06 2008
+++ julius4/libjulius/src/jfunc.c Fri Dec 12 16:14:57 2008
@@ -19,7 +19,7 @@
* @author Akinobu Lee
* @date Wed Aug 8 15:04:28 2007
*
- * $Revision: 1.4 $
+ * $Revision: 1.5 $
*
*/
/*
@@ -1484,8 +1484,10 @@
LOGPROB max_score;
PROCESS_AM *am;
MFCCCalc *mfcc;
+ SentenceAlign *align;
s = &(r->result.sent[0]);
+ align = result_align_new();
max_score = LOG_ZERO;
@@ -1507,10 +1509,10 @@
return;
}
outprob_prepare(&(r->am->hmmwrk), mfcc->param->samplenum);
- word_align(s->word, s->word_num, mfcc->param, s, r);
- printf("%f: %f\n", alpha, s->align.allscore);
- if (max_score < s->align.allscore) {
- max_score = s->align.allscore;
+ word_align(s->word, s->word_num, mfcc->param, align, r);
+ printf("%f: %f\n", alpha, align->allscore);
+ if (max_score < align->allscore) {
+ max_score = align->allscore;
max_alpha = alpha;
}
}
@@ -1523,6 +1525,8 @@
printf("------------ end VTLN -------------\n");
+ result_align_free(align);
+
}
#endif
Index: julius4/libjulius/src/recogmain.c
diff -u julius4/libjulius/src/recogmain.c:1.9 julius4/libjulius/src/recogmain.c:1.10
--- julius4/libjulius/src/recogmain.c:1.9 Tue Nov 18 17:46:59 2008
+++ julius4/libjulius/src/recogmain.c Fri Dec 12 16:14:57 2008
@@ -12,7 +12,7 @@
* @author Akinobu Lee
* @date Wed Aug 8 14:53:53 2007
*
- * $Revision: 1.9 $
+ * $Revision: 1.10 $
*
*/
@@ -192,6 +192,67 @@
/**
* <EN>
+ * allocate storage of recognition alignment results.
+ *
+ * @return the new pointer
+ * </EN>
+ * <JA>
+ * アラインメント結果の格納場所を確保
+ *
+ * @return 確保された領域へのポインタ
+ * </JA>
+ *
+ * @callgraph
+ * @callergraph
+ *
+ */
+SentenceAlign *
+result_align_new()
+{
+ SentenceAlign *new;
+ new = (SentenceAlign *)mymalloc(sizeof(SentenceAlign));
+ new->w = NULL;
+ new->ph = NULL;
+ new->loc = NULL;
+ new->begin_frame = NULL;
+ new->end_frame = NULL;
+ new->avgscore = NULL;
+ new->is_iwsp = NULL;
+ new->next = NULL;
+ return new;
+}
+
+/**
+ * <EN>
+ * free storage of recognition alignment results.
+ *
+ * @param a [i/o] alignment data to be released
+ * </EN>
+ * <JA>
+ * アラインメント結果の格納場所を確保
+ *
+ * @param a [i/o] 解放されるアラインメントデータ
+ * </JA>
+ *
+ * @callgraph
+ * @callergraph
+ *
+ */
+void
+result_align_free(SentenceAlign *a)
+{
+ if (a->w) free(a->w);
+ if (a->ph) free(a->ph);
+ if (a->loc) free(a->loc);
+ if (a->begin_frame) free(a->begin_frame);
+ if (a->end_frame) free(a->end_frame);
+ if (a->avgscore) free(a->avgscore);
+ if (a->is_iwsp) free(a->is_iwsp);
+ free(a);
+}
+
+/**
+ * <EN>
* Allocate storage of recognition results.
* </EN>
* <JA>
@@ -210,16 +271,7 @@
{
int i;
r->result.sent = (Sentence *)mymalloc(sizeof(Sentence) * num);
- for(i=0;i<num;i++) {
- r->result.sent[i].align.filled = FALSE;
- r->result.sent[i].align.w = NULL;
- r->result.sent[i].align.ph = NULL;
- r->result.sent[i].align.loc = NULL;
- r->result.sent[i].align.begin_frame = NULL;
- r->result.sent[i].align.end_frame = NULL;
- r->result.sent[i].align.avgscore = NULL;
- r->result.sent[i].align.is_iwsp = NULL;
- }
+ for(i=0;i<num;i++) r->result.sent[i].align = NULL;
r->result.sentnum = 0;
}
@@ -240,15 +292,15 @@
result_sentence_free(RecogProcess *r)
{
int i;
+ SentenceAlign *a, *atmp;
if (r->result.sent) {
for(i=0;i<r->result.sentnum;i++) {
- if (r->result.sent[i].align.w) free(r->result.sent[i].align.w);
- if (r->result.sent[i].align.ph) free(r->result.sent[i].align.ph);
- if (r->result.sent[i].align.loc) free(r->result.sent[i].align.loc);
- if (r->result.sent[i].align.begin_frame) free(r->result.sent[i].align.begin_frame);
- if (r->result.sent[i].align.end_frame) free(r->result.sent[i].align.end_frame);
- if (r->result.sent[i].align.avgscore) free(r->result.sent[i].align.avgscore);
- if (r->result.sent[i].align.is_iwsp) free(r->result.sent[i].align.is_iwsp);
+ a = r->result.sent[i].align;
+ while(a) {
+ atmp = a->next;
+ result_align_free(a);
+ a = atmp;
+ }
}
free(r->result.sent);
r->result.sent = NULL;
Index: julius4/libjulius/src/word_align.c
diff -u julius4/libjulius/src/word_align.c:1.4 julius4/libjulius/src/word_align.c:1.5
--- julius4/libjulius/src/word_align.c:1.4 Sun Nov 16 21:28:04 2008
+++ julius4/libjulius/src/word_align.c Fri Dec 12 16:14:57 2008
@@ -33,7 +33,7 @@
* @author Akinobu Lee
* @date Sat Sep 24 16:09:46 2005
*
- * $Revision: 1.4 $
+ * $Revision: 1.5 $
*
*/
/*
@@ -170,7 +170,7 @@
* @param wnum [in] @a words の長さ
* @param param [in] 入力特徴パラメータ列
* @param per_what [in] 単語・音素・状態のどの単位でアラインメントを取るかを指定
- * @param s [out] アラインメント結果を格納するSentence構造体
+ * @param align [out] アラインメント結果を格納するSentence構造体
* @param r [i/o] 認識処理インスタンス
* </JA>
* <EN>
@@ -185,7 +185,7 @@
* </EN>
*/
static void
-do_align(WORD_ID *words, short wnum, HTK_Param *param, int per_what, Sentence *s, RecogProcess *r)
+do_align(WORD_ID *words, short wnum, HTK_Param *param, int per_what, SentenceAlign *align, RecogProcess *r)
{
HMM_Logical **phones; /* phoneme sequence */
boolean *has_sp; /* whether phone can follow short pause */
@@ -277,53 +277,51 @@
allscore = viterbi_segment(shmm, param, r->wchmm->hmmwrk, hmminfo->multipath, end_state, end_num, &id_seq, &end_frame, &end_score, &rlen);
/* store result to s */
- s->align.num = rlen;
- s->align.unittype = per_what;
- s->align.begin_frame = (int *)mymalloc(sizeof(int) * rlen);
- s->align.end_frame = (int *)mymalloc(sizeof(int) * rlen);
- s->align.avgscore = (LOGPROB *)mymalloc(sizeof(LOGPROB) * rlen);
+ align->num = rlen;
+ align->unittype = per_what;
+ align->begin_frame = (int *)mymalloc(sizeof(int) * rlen);
+ align->end_frame = (int *)mymalloc(sizeof(int) * rlen);
+ align->avgscore = (LOGPROB *)mymalloc(sizeof(LOGPROB) * rlen);
for(i=0;i<rlen;i++) {
- s->align.begin_frame[i] = (i == 0) ? 0 : end_frame[i-1] + 1;
- s->align.end_frame[i] = end_frame[i];
- s->align.avgscore[i] = end_score[i];
+ align->begin_frame[i] = (i == 0) ? 0 : end_frame[i-1] + 1;
+ align->end_frame[i] = end_frame[i];
+ align->avgscore[i] = end_score[i];
}
switch(per_what) {
case PER_WORD:
- s->align.w = (WORD_ID *)mymalloc(sizeof(WORD_ID) * rlen);
+ align->w = (WORD_ID *)mymalloc(sizeof(WORD_ID) * rlen);
for(i=0;i<rlen;i++) {
- s->align.w[i] = words[id_seq[i]];
+ align->w[i] = words[id_seq[i]];
}
break;
case PER_PHONEME:
- s->align.ph = (HMM_Logical **)mymalloc(sizeof(HMM_Logical *) * rlen);
+ align->ph = (HMM_Logical **)mymalloc(sizeof(HMM_Logical *) * rlen);
for(i=0;i<rlen;i++) {
- s->align.ph[i] = phones[id_seq[i]];
+ align->ph[i] = phones[id_seq[i]];
}
break;
case PER_STATE:
- s->align.ph = (HMM_Logical **)mymalloc(sizeof(HMM_Logical *) * rlen);
- s->align.loc = (short *)mymalloc(sizeof(short) * rlen);
- if (hmminfo->multipath) s->align.is_iwsp = (boolean *)mymalloc(sizeof(boolean) * rlen);
+ align->ph = (HMM_Logical **)mymalloc(sizeof(HMM_Logical *) * rlen);
+ align->loc = (short *)mymalloc(sizeof(short) * rlen);
+ if (hmminfo->multipath) align->is_iwsp = (boolean *)mymalloc(sizeof(boolean) * rlen);
for(i=0;i<rlen;i++) {
- s->align.ph[i] = phones[phloc[id_seq[i]]];
+ align->ph[i] = phones[phloc[id_seq[i]]];
if (hmminfo->multipath) {
if (enable_iwsp && stloc[id_seq[i]] > end_num) {
- s->align.loc[i] = stloc[id_seq[i]] - end_num;
- s->align.is_iwsp[i] = TRUE;
+ align->loc[i] = stloc[id_seq[i]] - end_num;
+ align->is_iwsp[i] = TRUE;
} else {
- s->align.loc[i] = stloc[id_seq[i]];
- s->align.is_iwsp[i] = FALSE;
+ align->loc[i] = stloc[id_seq[i]];
+ align->is_iwsp[i] = FALSE;
}
} else {
- s->align.loc[i] = stloc[id_seq[i]];
+ align->loc[i] = stloc[id_seq[i]];
}
}
break;
}
- s->align.allscore = allscore;
-
- s->align.filled = TRUE;
+ align->allscore = allscore;
free_hmm(shmm);
free(id_seq);
@@ -353,7 +351,7 @@
* @param words [in] 単語列
* @param wnum [in] @a words の単語数
* @param param [in] 入力特徴ベクトル列
- * @param s [out] アラインメント結果を格納するSentence構造体
+ * @param align [out] アラインメント結果を格納するSentence構造体
* @param r [i/o] 認識処理インスタンス
* </JA>
* <EN>
@@ -362,16 +360,16 @@
* @param words [in] word sequence
* @param wnum [in] length of @a words
* @param param [in] input parameter vectors
- * @param s [out] Sentence data area to store the alignment result
+ * @param align [out] Sentence data area to store the alignment result
* @param r [i/o] recognition process instance
* </EN>
* @callgraph
* @callergraph
*/
void
-word_align(WORD_ID *words, short wnum, HTK_Param *param, Sentence *s, RecogProcess *r)
+word_align(WORD_ID *words, short wnum, HTK_Param *param, SentenceAlign *align, RecogProcess *r)
{
- do_align(words, wnum, param, PER_WORD, s, r);
+ do_align(words, wnum, param, PER_WORD, align, r);
}
/**
@@ -381,7 +379,7 @@
* @param revwords [in] 単語列（逆順）
* @param wnum [in] @a revwords の単語数
* @param param [in] 入力特徴ベクトル列
- * @param s [out] アラインメント結果を格納するSentence構造体
+ * @param align [out] アラインメント結果を格納するSentence構造体
* @param r [i/o] 認識処理インスタンス
* </JA>
* <EN>
@@ -390,20 +388,20 @@
* @param revwords [in] word sequence in reversed direction
* @param wnum [in] length of @a revwords
* @param param [in] input parameter vectors
- * @param s [out] Sentence data area to store the alignment result
+ * @param align [out] Sentence data area to store the alignment result
* @param r [i/o] recognition process instance
* </EN>
* @callgraph
* @callergraph
*/
void
-word_rev_align(WORD_ID *revwords, short wnum, HTK_Param *param, Sentence *s, RecogProcess *r)
+word_rev_align(WORD_ID *revwords, short wnum, HTK_Param *param, SentenceAlign *align, RecogProcess *r)
{
WORD_ID *words; /* word sequence (true order) */
int w;
words = (WORD_ID *)mymalloc(sizeof(WORD_ID) * wnum);
for (w=0;w<wnum;w++) words[w] = revwords[wnum-w-1];
- do_align(words, wnum, param, PER_WORD, s, r);
+ do_align(words, wnum, param, PER_WORD, align, r);
free(words);
}
@@ -414,7 +412,7 @@
* @param words [in] 単語列
* @param num [in] @a words の単語数
* @param param [in] 入力特徴ベクトル列
- * @param s [out] アラインメント結果を格納するSentence構造体
+ * @param align [out] アラインメント結果を格納するSentence構造体
* @param r [i/o] 認識処理インスタンス
* </JA>
* <EN>
@@ -423,16 +421,16 @@
* @param words [in] word sequence
* @param num [in] length of @a words
* @param param [in] input parameter vectors
- * @param s [out] Sentence data area to store the alignment result
+ * @param align [out] Sentence data area to store the alignment result
* @param r [i/o] recognition process instance
* </EN>
* @callgraph
* @callergraph
*/
void
-phoneme_align(WORD_ID *words, short num, HTK_Param *param, Sentence *s, RecogProcess *r)
+phoneme_align(WORD_ID *words, short num, HTK_Param *param, SentenceAlign *align, RecogProcess *r)
{
- do_align(words, num, param, PER_PHONEME, s, r);
+ do_align(words, num, param, PER_PHONEME, align, r);
}
/**
@@ -442,7 +440,7 @@
* @param revwords [in] 単語列（逆順）
* @param num [in] @a revwords の単語数
* @param param [in] 入力特徴ベクトル列
- * @param s [out] アラインメント結果を格納するSentence構造体
+ * @param align [out] アラインメント結果を格納するSentence構造体
* @param r [i/o] 認識処理インスタンス
* </JA>
* <EN>
@@ -451,20 +449,20 @@
* @param revwords [in] word sequence in reversed direction
* @param num [in] length of @a revwords
* @param param [in] input parameter vectors
- * @param s [out] Sentence data area to store the alignment result
+ * @param align [out] Sentence data area to store the alignment result
* @param r [i/o] recognition process instance
* </EN>
* @callgraph
* @callergraph
*/
void
-phoneme_rev_align(WORD_ID *revwords, short num, HTK_Param *param, Sentence *s, RecogProcess *r)
+phoneme_rev_align(WORD_ID *revwords, short num, HTK_Param *param, SentenceAlign *align, RecogProcess *r)
{
WORD_ID *words; /* word sequence (true order) */
int p;
words = (WORD_ID *)mymalloc(sizeof(WORD_ID) * num);
for (p=0;p<num;p++) words[p] = revwords[num-p-1];
- do_align(words, num, param, PER_PHONEME, s, r);
+ do_align(words, num, param, PER_PHONEME, align, r);
free(words);
}
@@ -475,7 +473,7 @@
* @param words [in] 単語列
* @param num [in] @a words の単語数
* @param param [in] 入力特徴ベクトル列
- * @param s [out] アラインメント結果を格納するSentence構造体
+ * @param align [out] アラインメント結果を格納するSentence構造体
* @param r [i/o] 認識処理インスタンス
* </JA>
* <EN>
@@ -484,16 +482,16 @@
* @param words [in] word sequence
* @param num [in] length of @a words
* @param param [in] input parameter vectors
- * @param s [out] Sentence data area to store the alignment result
+ * @param align [out] Sentence data area to store the alignment result
* @param r [i/o] recognition process instance
* </EN>
* @callgraph
* @callergraph
*/
void
-state_align(WORD_ID *words, short num, HTK_Param *param, Sentence *s, RecogProcess *r)
+state_align(WORD_ID *words, short num, HTK_Param *param, SentenceAlign *align, RecogProcess *r)
{
- do_align(words, num, param, PER_STATE, s, r);
+ do_align(words, num, param, PER_STATE, align, r);
}
/**
@@ -503,7 +501,7 @@
* @param revwords [in] 単語列（逆順）
* @param num [in] @a revwords の単語数
* @param param [in] 入力特徴ベクトル列
- * @param s [out] アラインメント結果を格納するSentence構造体
+ * @param align [out] アラインメント結果を格納するSentence構造体
* @param r [i/o] 認識処理インスタンス
* </JA>
* <EN>
@@ -512,20 +510,20 @@
* @param revwords [in] word sequence in reversed direction
* @param num [in] length of @a revwords
* @param param [in] input parameter vectors
- * @param s [out] Sentence data area to store the alignment result
+ * @param align [out] Sentence data area to store the alignment result
* @param r [i/o] recognition process instance
* </EN>
* @callgraph
* @callergraph
*/
void
-state_rev_align(WORD_ID *revwords, short num, HTK_Param *param, Sentence *s, RecogProcess *r)
+state_rev_align(WORD_ID *revwords, short num, HTK_Param *param, SentenceAlign *align, RecogProcess *r)
{
WORD_ID *words; /* word sequence (true order) */
int p;
words = (WORD_ID *)mymalloc(sizeof(WORD_ID) * num);
for (p=0;p<num;p++) words[p] = revwords[num-p-1];
- do_align(words, num, param, PER_STATE, s, r);
+ do_align(words, num, param, PER_STATE, align, r);
free(words);
}
@@ -550,16 +548,32 @@
{
int n;
Sentence *s;
+ SentenceAlign *now, *prev;
for(n = 0; n < r->result.sentnum; n++) {
s = &(r->result.sent[n]);
- /* do forced alignment if needed */
- if (r->config->annotate.align_result_word_flag)
- word_align(s->word, s->word_num, param, s, r);
- if (r->config->annotate.align_result_phoneme_flag)
- phoneme_align(s->word, s->word_num, param, s, r);
- if (r->config->annotate.align_result_state_flag)
- state_align(s->word, s->word_num, param, s, r);
+ /* do forced alignment if needed */
+ if (r->config->annotate.align_result_word_flag) {
+ now = result_align_new();
+ word_align(s->word, s->word_num, param, now, r);
+ if (s->align == NULL) s->align = now;
+ else prev->next = now;
+ prev = now;
+ }
+ if (r->config->annotate.align_result_phoneme_flag) {
+ now = result_align_new();
+ phoneme_align(s->word, s->word_num, param, now, r);
+ if (s->align == NULL) s->align = now;
+ else prev->next = now;
+ prev = now;
+ }
+ if (r->config->annotate.align_result_state_flag) {
+ now = result_align_new();
+ state_align(s->word, s->word_num, param, now, r);
+ if (s->align == NULL) s->align = now;
+ else prev->next = now;
+ prev = now;
+ }
}
}