1 /** 2 * @file vocabulary.h 3 * 4 * <JA> 5 * @brief ñ�켭��ι�¤����� 6 * 7 * ���Υե������ǧ�����Ѥ�����ñ�켭���������ޤ���ñ�켭���ñ����ɤߡ� 8 * ����ʸ�����������¾�ˡ�ʸ�γ��ϡ���λñ���Ʃ��ñ�������ݻ����ޤ��� 9 * 10 * N-gram�˽и�������äμ���� NGRAM_INFO �˳�Ǽ���졤����ǧ����ñ�켭��Ȥ� 11 * ���̤���뤳�Ȥ���դ��Ʋ�������ñ�켭����N-gram�θ��äؤΥޥåԥ��� 12 * WORD_INFO ��� wton[] �ˤ�äƹԤʤ��ޤ����ޤ�DFA�ξ�硤wton �� 13 * ����ñ�줬°����DFA_INFO ��Υ��ƥ����ֹ��ޤߤޤ��� 14 * </JA> 15 * <EN> 16 * @brief Word dictionary for recognition 17 * 18 * This file defines data structure for word dictionary used in recognition. 19 * It stores word's string, output string, phoneme sequence, transparency. 20 * Beginning-of-sentence word and End-of-sentence word guessed from runtime 21 * environment is also stored here. 22 * 23 * Please note that the N-gram vocabulary is stored in NGRAM_INFO and it 24 * can differ from this word dictionary. The reference from the word 25 * dictionary to a N-gram vocabulary is done by wton[] member in WORD_INFO. 26 * When used with DFA, the wton[] holds a category number to which each word 27 * belongs. 28 * </EN> 29 * 30 * @author Akinobu LEE 31 * @date Sat Feb 12 12:38:13 2005 32 * 33 * $Revision: 1.5 $ 34 * 35 */ 36 /* 37 * Copyright (c) 1991-2007 Kawahara Lab., Kyoto University 38 * Copyright (c) 2000-2005 Shikano Lab., Nara Institute of Science and Technology 39 * Copyright (c) 2005-2007 Julius project team, Nagoya Institute of Technology 40 * All rights reserved 41 */ 42 43 #ifndef __SENT_VOCA_H__ 44 #define __SENT_VOCA_H__ 45 46 #include <sent/stddefs.h> 47 #include <sent/htk_hmm.h> 48 49 /// Memory allocation step in number of words when loading a word dictionary 50 #define MAXWSTEP 4000 51 52 /// Word dictionary structure to hold vocabulary 53 typedef struct { 54 WORD_ID maxnum; ///< Allocated number of word space 55 WORD_ID num; ///< Number of words 56 WORD_ID errnum; ///< Number of error words that were skipped when reading dictionary 57 WORD_ID linenum; ///< Current line number while loading 58 boolean do_conv; ///< TRUE if needs conversion while loading 59 boolean ok_flag; ///< FALSE if any error occur while loading 60 61 unsigned char *wlen; ///< Number of phonemes for each word [wid] 62 63 char **wname; ///< Word name string for each word [wid]. With DFA, it's category ID. With N-gram, it's N-gram entry name. 64 char **woutput; ///< Word output string that will be output as recognition result for each word [wid] 65 HMM_Logical ***wseq; ///< Phone sequence of each word [wid][0..wlen[wid]-1] 66 WORD_ID *wton; ///< Reference to N-gram/category ID of each word ID [wid] 67 #ifdef CLASS_NGRAM 68 LOGPROB *cprob; ///< Class probability of each word [wid] 69 WORD_ID cwnum; ///< Number of words whose class prob is specified (just for a statistic) 70 #endif 71 WORD_ID head_silwid; ///< Word ID of beginning-of-sentence silence 72 WORD_ID tail_silwid; ///< Word ID of end-of-sentence silence 73 short maxwn; ///< Maximum number of %HMM states per word (statistic) 74 short maxwlen; ///< Maximum number of phones in a word (statistic) 75 int totalstatenum; ///< Total number of HMM states 76 int totalmodelnum; ///< Total number of models (phonemes) 77 int totaltransnum; ///< Total number of state transitions 78 boolean *is_transparent; ///< TRUE if the word can be treated as transparent [wid] 79 APATNODE *errph_root; ///< Root node of index tree for gathering error %HMM name appeared when reading the dictionary 80 BMALLOC_BASE *mroot; ///< Pointer for block memory allocation 81 } WORD_INFO; 82 83 WORD_INFO *word_info_new(); 84 void word_info_free(WORD_INFO *winfo); 85 void winfo_init(WORD_INFO *winfo); 86 boolean winfo_expand(WORD_INFO *winfo); 87 boolean init_voca(WORD_INFO *winfo, char *filename, HTK_HMM_INFO *hmminfo, boolean, boolean); 88 boolean init_wordlist(WORD_INFO *winfo, char *filename, HTK_HMM_INFO *hmminfo, char *headphone, char *tailphone, char *contextphone, boolean force_dict); 89 void voca_set_stats(WORD_INFO *winfo); 90 void voca_load_start(WORD_INFO *winfo, HTK_HMM_INFO *hmminfo, boolean ignore_tri_conv); 91 boolean voca_load_line(char *buf, WORD_INFO *winfo, HTK_HMM_INFO *hmminfo); 92 boolean voca_load_end(WORD_INFO *winfo); 93 boolean voca_load_htkdict(FILE *, WORD_INFO *, HTK_HMM_INFO *, boolean); 94 boolean voca_load_htkdict_fp(FILE *, WORD_INFO *, HTK_HMM_INFO *, boolean); 95 boolean voca_append_htkdict(char *entry, WORD_INFO *winfo, HTK_HMM_INFO *hmminfo, boolean ignore_tri_conv); 96 boolean voca_append(WORD_INFO *dstinfo, WORD_INFO *srcinfo, int coffset, int woffset); 97 boolean voca_load_htkdict_line(char *buf, WORD_ID *vnum, int linenum, WORD_INFO *winfo, HTK_HMM_INFO *hmminfo, boolean do_conv, boolean *ok_flag); 98 99 boolean voca_load_word_line(char *buf, WORD_INFO *winfo, HTK_HMM_INFO *hmminfo, char *headphone, char *tailpohone, char *contextphone); 100 boolean voca_load_wordlist(FILE *fp, WORD_INFO *winfo, HTK_HMM_INFO *hmminfo, char *headphone, char *tailphone, char *contextphone); 101 boolean voca_load_wordlist_fp(FILE *fp, WORD_INFO *winfo, HTK_HMM_INFO *hmminfo, char *headphone, char *tailphone, char *contextphone); 102 boolean voca_load_wordlist_line(char *buf, WORD_ID *vnum, int linenum, WORD_INFO *winfo, HTK_HMM_INFO *hmminfo, boolean do_conv, boolean *ok_flag, char *headphone, char *tailphone, char *contextphone); 103 boolean voca_mono2tri(WORD_INFO *winfo, HTK_HMM_INFO *hmminfo); 104 WORD_ID voca_lookup_wid(char *, WORD_INFO *); 105 WORD_ID *new_str2wordseq(WORD_INFO *, char *, int *); 106 char *cycle_triphone(char *p); 107 char *cycle_triphone_flush(); 108 void print_voca_info(FILE *fp, WORD_INFO *); 109 void put_voca(FILE *fp, WORD_INFO *winfo, WORD_ID wid); 110 111 /* HMMList check */ 112 boolean make_base_phone(HTK_HMM_INFO *hmminfo, WORD_INFO *winfo); 113 void print_phone_info(FILE *fp, HTK_HMM_INFO *hmminfo); 114 void print_all_basephone_detail(HMM_basephone *base); 115 void print_all_basephone_name(HMM_basephone *base); 116 void test_interword_triphone(HTK_HMM_INFO *hmminfo, WORD_INFO *winfo); 117 118 #endif /* __SENT_VOCA_H__ */ 119