1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
|
/* -*- c-basic-offset: 4; indent-tabs-mode: nil -*- */
/* ====================================================================
* Copyright (c) 1999-2007 Carnegie Mellon University. All rights
* reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the
* distribution.
*
* This work was supported in part by funding from the Defense Advanced
* Research Projects Agency and the National Science Foundation of the
* United States of America, and the CMU Sphinx Speech Consortium.
*
* THIS SOFTWARE IS PROVIDED BY CARNEGIE MELLON UNIVERSITY ``AS IS'' AND
* ANY EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
* THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL CARNEGIE MELLON UNIVERSITY
* NOR ITS EMPLOYEES BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
* ====================================================================
*
*/
/*
* \file lm3g_model.c Core Sphinx 3-gram code used in
* DMP/DMP32/ARPA (for now) model code.
*
* Author: A cast of thousands, probably.
*/
#include <string.h>
#include <assert.h>
#include <limits.h>
#include "sphinxbase/listelem_alloc.h"
#include "sphinxbase/ckd_alloc.h"
#include "sphinxbase/err.h"
#include "lm3g_model.h"
void
lm3g_tginfo_free(ngram_model_t *base, lm3g_model_t *lm3g)
{
if (lm3g->tginfo == NULL)
return;
listelem_alloc_free(lm3g->le);
ckd_free(lm3g->tginfo);
}
void
lm3g_tginfo_reset(ngram_model_t *base, lm3g_model_t *lm3g)
{
if (lm3g->tginfo == NULL)
return;
listelem_alloc_free(lm3g->le);
memset(lm3g->tginfo, 0, base->n_counts[0] * sizeof(tginfo_t *));
lm3g->le = listelem_alloc_init(sizeof(tginfo_t));
}
void
lm3g_apply_weights(ngram_model_t *base,
lm3g_model_t *lm3g,
float32 lw, float32 wip, float32 uw)
{
int32 log_wip, log_uw, log_uniform_weight;
int i;
/* Precalculate some log values we will like. */
log_wip = logmath_log(base->lmath, wip);
log_uw = logmath_log(base->lmath, uw);
log_uniform_weight = logmath_log(base->lmath, 1.0 - uw);
for (i = 0; i < base->n_counts[0]; ++i) {
int32 prob1, bo_wt, n_used;
/* Backoff weights just get scaled by the lw. */
bo_wt = (int32)(lm3g->unigrams[i].bo_wt1.l / base->lw);
/* Unscaling unigram probs is a bit more complicated, so punt
* it back to the general code. */
prob1 = ngram_ng_prob(base, i, NULL, 0, &n_used);
/* Now compute the new scaled probabilities. */
lm3g->unigrams[i].bo_wt1.l = (int32)(bo_wt * lw);
if (strcmp(base->word_str[i], "<s>") == 0) { /* FIXME: configurable start_sym */
/* Apply language weight and WIP */
lm3g->unigrams[i].prob1.l = (int32)(prob1 * lw) + log_wip;
}
else {
/* Interpolate unigram probability with uniform. */
prob1 += log_uw;
prob1 = logmath_add(base->lmath, prob1, base->log_uniform + log_uniform_weight);
/* Apply language weight and WIP */
lm3g->unigrams[i].prob1.l = (int32)(prob1 * lw) + log_wip;
}
}
for (i = 0; i < lm3g->n_prob2; ++i) {
int32 prob2;
/* Can't just punt this back to general code since it is quantized. */
prob2 = (int32)((lm3g->prob2[i].l - base->log_wip) / base->lw);
lm3g->prob2[i].l = (int32)(prob2 * lw) + log_wip;
}
if (base->n > 2) {
for (i = 0; i < lm3g->n_bo_wt2; ++i) {
lm3g->bo_wt2[i].l = (int32)(lm3g->bo_wt2[i].l / base->lw * lw);
}
for (i = 0; i < lm3g->n_prob3; i++) {
int32 prob3;
/* Can't just punt this back to general code since it is quantized. */
prob3 = (int32)((lm3g->prob3[i].l - base->log_wip) / base->lw);
lm3g->prob3[i].l = (int32)(prob3 * lw) + log_wip;
}
}
/* Store updated values in the model. */
base->log_wip = log_wip;
base->log_uw = log_uw;
base->log_uniform_weight = log_uniform_weight;
base->lw = lw;
}
int32
lm3g_add_ug(ngram_model_t *base,
lm3g_model_t *lm3g, int32 wid, int32 lweight)
{
int32 score;
/* This would be very bad if this happened! */
assert(!NGRAM_IS_CLASSWID(wid));
/* Reallocate unigram array. */
lm3g->unigrams = ckd_realloc(lm3g->unigrams,
sizeof(*lm3g->unigrams) * base->n_1g_alloc);
memset(lm3g->unigrams + base->n_counts[0], 0,
(base->n_1g_alloc - base->n_counts[0]) * sizeof(*lm3g->unigrams));
/* Reallocate tginfo array. */
lm3g->tginfo = ckd_realloc(lm3g->tginfo,
sizeof(*lm3g->tginfo) * base->n_1g_alloc);
memset(lm3g->tginfo + base->n_counts[0], 0,
(base->n_1g_alloc - base->n_counts[0]) * sizeof(*lm3g->tginfo));
/* FIXME: we really ought to update base->log_uniform *and*
* renormalize all the other unigrams. This is really slow, so I
* will probably just provide a function to renormalize after
* adding unigrams, for anyone who really cares. */
/* This could be simplified but then we couldn't do it in logmath */
score = lweight + base->log_uniform + base->log_uw;
score = logmath_add(base->lmath, score,
base->log_uniform + base->log_uniform_weight);
lm3g->unigrams[wid].prob1.l = score;
/* This unigram by definition doesn't participate in any bigrams,
* so its backoff weight and bigram pointer are both undefined. */
lm3g->unigrams[wid].bo_wt1.l = 0;
lm3g->unigrams[wid].bigrams = 0;
/* Finally, increase the unigram count */
++base->n_counts[0];
/* FIXME: Note that this can actually be quite bogus due to the
* presence of class words. If wid falls outside the unigram
* count, increase it to compensate, at the cost of no longer
* really knowing how many unigrams we have :( */
if (wid >= base->n_counts[0])
base->n_counts[0] = wid + 1;
return score;
}
#define INITIAL_SORTED_ENTRIES MAX_UINT16
void
init_sorted_list(sorted_list_t * l)
{
l->list = ckd_calloc(INITIAL_SORTED_ENTRIES, sizeof(sorted_entry_t));
l->list[0].val.l = INT_MIN;
l->list[0].lower = 0;
l->list[0].higher = 0;
l->free = 1;
l->size = INITIAL_SORTED_ENTRIES;
}
void
free_sorted_list(sorted_list_t * l)
{
free(l->list);
}
lmprob_t *
vals_in_sorted_list(sorted_list_t * l)
{
lmprob_t *vals;
int32 i;
vals = ckd_calloc(l->free, sizeof(lmprob_t));
for (i = 0; i < l->free; i++)
vals[i] = l->list[i].val;
return (vals);
}
int32
sorted_id(sorted_list_t * l, int32 *val)
{
int32 i = 0;
for (;;) {
if (*val == l->list[i].val.l)
return (i);
if (*val < l->list[i].val.l) {
if (l->list[i].lower == 0) {
if (l->free >= l->size) {
int newsize = l->size + INITIAL_SORTED_ENTRIES;
l->list = ckd_realloc(l->list, sizeof(sorted_entry_t) * newsize);
memset(l->list + l->size,
0, INITIAL_SORTED_ENTRIES * sizeof(sorted_entry_t));
l->size = newsize;
}
l->list[i].lower = l->free;
(l->free)++;
i = l->list[i].lower;
l->list[i].val.l = *val;
return (i);
}
else
i = l->list[i].lower;
}
else {
if (l->list[i].higher == 0) {
if (l->free >= l->size) {
int newsize = l->size + INITIAL_SORTED_ENTRIES;
l->list = ckd_realloc(l->list, sizeof(sorted_entry_t) * newsize);
memset(l->list + l->size,
0, INITIAL_SORTED_ENTRIES * sizeof(sorted_entry_t));
l->size = newsize;
}
l->list[i].higher = l->free;
(l->free)++;
i = l->list[i].higher;
l->list[i].val.l = *val;
return (i);
}
else
i = l->list[i].higher;
}
}
}
|
