1 /* -*- c-file-style: "linux" -*- */
2 /* ====================================================================
3 * Copyright (c) 2005 Carnegie Mellon University. All rights
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
18 * This work was supported in part by funding from the Defense Advanced
19 * Research Projects Agency and the National Science Foundation of the
20 * United States of America, and the CMU Sphinx Speech Consortium.
22 * THIS SOFTWARE IS PROVIDED BY CARNEGIE MELLON UNIVERSITY ``AS IS'' AND
23 * ANY EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
24 * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
25 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL CARNEGIE MELLON UNIVERSITY
26 * NOR ITS EMPLOYEES BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 * ====================================================================
40 * Binary format model definition files, with support for
41 * heterogeneous topologies and variable-size N-phones
43 * @author David Huggins-Daines <dhuggins@cs.cmu.edu>
45 #ifndef __BIN_MDEF_H__
46 #define __BIN_MDEF_H__
53 #endif /* __cplusplus */
55 /* SphinxBase headers. */
56 #include <sphinxbase/mmio.h>
57 #include <sphinxbase/cmd_ln.h>
61 #define BIN_MDEF_FORMAT_VERSION 1
62 /* Little-endian machines will write "BMDF" to disk, big-endian ones "FDMB". */
63 #define BIN_MDEF_NATIVE_ENDIAN 0x46444d42 /* 'BMDF' in little-endian order */
64 #define BIN_MDEF_OTHER_ENDIAN 0x424d4446 /* 'BMDF' in big-endian order */
66 #define ATTRIBUTE_PACKED __attribute__((packed))
68 #define ATTRIBUTE_PACKED
72 * Phone entry (on-disk, 12 bytes)
74 typedef struct mdef_entry_s mdef_entry_t;
76 int32 ssid; /**< Senone sequence ID */
77 int32 tmat; /**< Transition matrix ID */
78 /* FIXME: is any of this actually necessary? */
80 /**< CI phone information - attributes (just "filler" for now) */
85 /**< CD phone information - context info. */
88 int8 ctx[3]; /**< quintphones will require hacking */
94 * Invalid senone sequence ID (limited to 16 bits for PocketSphinx).
96 #define BAD_SSID 0xffff
98 * Invalid senone ID (limited to 16 bits for PocketSphinx).
100 #define BAD_SENID 0xffff
103 * Node in CD phone tree (on-disk, 8 bytes).
105 typedef struct cd_tree_s cd_tree_t;
107 int16 ctx; /**< Context (word position or CI phone) */
108 int16 n_down; /**< Number of children (0 for leafnode) */
110 int32 pid; /**< Phone ID (leafnode) */
111 int32 down; /**< Next level of the tree (offset from start of cd_trees) */
116 * Model definition structure (in-memory).
118 typedef struct bin_mdef_s bin_mdef_t;
121 int32 n_ciphone; /**< Number of base (CI) phones */
122 int32 n_phone; /**< Number of base (CI) phones + (CD) triphones */
123 int32 n_emit_state; /**< Number of emitting states per phone (0 for heterogeneous) */
124 int32 n_ci_sen; /**< Number of CI senones; these are the first */
125 int32 n_sen; /**< Number of senones (CI+CD) */
126 int32 n_tmat; /**< Number of transition matrices */
127 int32 n_sseq; /**< Number of unique senone sequences */
128 int32 n_ctx; /**< Number of phones of context */
129 int32 n_cd_tree; /**< Number of nodes in cd_tree (below) */
130 int16 sil; /**< CI phone ID for silence */
132 mmio_file_t *filemap;/**< File map for this file (if any) */
133 char **ciname; /**< CI phone names */
134 cd_tree_t *cd_tree; /**< Tree mapping CD phones to phone IDs */
135 mdef_entry_t *phone; /**< All phone structures */
136 uint16 **sseq; /**< Unique senone sequences (2D array built at load time) */
137 uint8 *sseq_len; /**< Number of states in each sseq (NULL for homogeneous) */
139 /* These two are not stored on disk, but are generated at load time. */
140 int16 *cd2cisen; /**< Parent CI-senone id for each senone */
141 int16 *sen2cimap; /**< Parent CI-phone for each senone (CI or CD) */
143 /** Allocation mode for this object. */
144 enum { BIN_MDEF_FROM_TEXT, BIN_MDEF_IN_MEMORY, BIN_MDEF_ON_DISK } alloc_mode;
147 #define bin_mdef_is_fillerphone(m,p) (((p) < (m)->n_ciphone) \
148 ? (m)->phone[p].info.ci.filler \
149 : (m)->phone[(m)->phone[p].info.cd.ctx[0]].info.ci.filler)
150 #define bin_mdef_is_ciphone(m,p) ((p) < (m)->n_ciphone)
151 #define bin_mdef_n_ciphone(m) ((m)->n_ciphone)
152 #define bin_mdef_n_phone(m) ((m)->n_phone)
153 #define bin_mdef_n_sseq(m) ((m)->n_sseq)
154 #define bin_mdef_n_emit_state(m) ((m)->n_emit_state)
155 #define bin_mdef_n_emit_state_phone(m,p) ((m)->n_emit_state ? (m)->n_emit_state \
156 : (m)->sseq_len[(m)->phone[p].ssid])
157 #define bin_mdef_n_sen(m) ((m)->n_sen)
158 #define bin_mdef_n_tmat(m) ((m)->n_tmat)
159 #define bin_mdef_pid2ssid(m,p) ((m)->phone[p].ssid)
160 #define bin_mdef_pid2tmatid(m,p) ((m)->phone[p].tmat)
161 #define bin_mdef_silphone(m) ((m)->sil)
162 #define bin_mdef_sen2cimap(m,s) ((m)->sen2cimap[s])
163 #define bin_mdef_sseq2sen(m,ss,pos) ((m)->sseq[ss][pos])
164 #define bin_mdef_pid2ci(m,p) (((p) < (m)->n_ciphone) ? (p) \
165 : (m)->phone[p].info.cd.ctx[0])
168 * Read a binary mdef from a file.
170 bin_mdef_t *bin_mdef_read(cmd_ln_t *config, const char *filename);
172 * Read a text mdef from a file (creating an in-memory binary mdef).
174 bin_mdef_t *bin_mdef_read_text(cmd_ln_t *config, const char *filename);
176 * Write a binary mdef to a file.
178 int bin_mdef_write(bin_mdef_t *m, const char *filename);
180 * Write a binary mdef to a text file.
182 int bin_mdef_write_text(bin_mdef_t *m, const char *filename);
184 * Retain a pointer to a bin_mdef_t.
186 bin_mdef_t *bin_mdef_retain(bin_mdef_t *m);
188 * Release a pointer to a binary mdef.
190 int bin_mdef_free(bin_mdef_t *m);
193 * Context-independent phone lookup.
194 * @return phone id for ciphone.
196 int bin_mdef_ciphone_id(bin_mdef_t *m, /**< In: Model structure being queried */
197 const char *ciphone); /**< In: ciphone for which id wanted */
200 * Case-insensitive context-independent phone lookup.
201 * @return phone id for ciphone.
203 int bin_mdef_ciphone_id_nocase(bin_mdef_t *m, /**< In: Model structure being queried */
204 const char *ciphone); /**< In: ciphone for which id wanted */
206 /* Return value: READ-ONLY ciphone string name for the given ciphone id */
207 const char *bin_mdef_ciphone_str(bin_mdef_t *m, /**< In: Model structure being queried */
208 int32 ci); /**< In: ciphone id for which name wanted */
210 /* Return value: phone id for the given constituents if found, else -1 */
211 int bin_mdef_phone_id(bin_mdef_t *m, /**< In: Model structure being queried */
212 int32 b, /**< In: base ciphone id */
213 int32 l, /**< In: left context ciphone id */
214 int32 r, /**< In: right context ciphone id */
215 int32 pos); /**< In: Word position */
217 /* Look up a phone id, backing off to other word positions. */
218 int bin_mdef_phone_id_nearest(bin_mdef_t * m, int32 b,
219 int32 l, int32 r, int32 pos);
222 * Create a phone string for the given phone (base or triphone) id in the given buf.
224 * @return 0 if successful, -1 if error.
226 int bin_mdef_phone_str(bin_mdef_t *m, /**< In: Model structure being queried */
227 int pid, /**< In: phone id being queried */
228 char *buf); /**< Out: On return, buf has the string */
232 #endif /* __cplusplus */
234 #endif /* __BIN_MDEF_H__ */