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
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
|
/*
* Copyright (C) 2008-2009 SVOX AG, Baslerstr. 30, 8048 Zuerich, Switzerland
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/**
* @file picokdt.h
*
* knowledge handling for decision trees
*
* Copyright (C) 2008-2009 SVOX AG, Baslerstr. 30, 8048 Zuerich, Switzerland
* All rights reserved.
*
* History:
* - 2009-04-20 -- initial version
*
*/
#ifndef PICOKDT_H_
#define PICOKDT_H_
#include "picoos.h"
#include "picoknow.h"
#ifdef __cplusplus
extern "C" {
#endif
#if 0
}
#endif
/* ************************************************************/
/*
Several specialized decision trees kb are provided by this
knowledge handling module:
- Part of speech prediction decision tree: ...kdt_PosP
- Part of speech disambiguation decision tree: ...kdt_PosD
- Grapheme-to-phoneme decision tree: ...kdt_G2P
- Phrasing decision tree: ...kdt_PHR
- Accentuation decision tree: ...kdt_ACC
these 5 tree types may be unified in the future to a single type
- Phono-acoustical model trees: ...kdt_PAM
(actually 11 trees, but all have the same characteristics and
are instances of the same class)
*/
/* ************************************************************/
/* ************************************************************/
/* defines and functions to create specialized kb, */
/* to be used by picorsrc only */
/* ************************************************************/
typedef enum {
PICOKDT_KDTTYPE_POSP,
PICOKDT_KDTTYPE_POSD,
PICOKDT_KDTTYPE_G2P,
PICOKDT_KDTTYPE_PHR,
PICOKDT_KDTTYPE_ACC,
PICOKDT_KDTTYPE_PAM
} picokdt_kdttype_t;
pico_status_t picokdt_specializeDtKnowledgeBase(picoknow_KnowledgeBase this,
picoos_Common common,
const picokdt_kdttype_t type);
/* ************************************************************/
/* decision tree types (opaque) and get Tree functions */
/* ************************************************************/
/* decision tree types */
typedef struct picokdt_dtposp * picokdt_DtPosP;
typedef struct picokdt_dtposd * picokdt_DtPosD;
typedef struct picokdt_dtg2p * picokdt_DtG2P;
typedef struct picokdt_dtphr * picokdt_DtPHR;
typedef struct picokdt_dtacc * picokdt_DtACC;
typedef struct picokdt_dtpam * picokdt_DtPAM;
/* return kb decision tree for usage in PU */
picokdt_DtPosP picokdt_getDtPosP(picoknow_KnowledgeBase this);
picokdt_DtPosD picokdt_getDtPosD(picoknow_KnowledgeBase this);
picokdt_DtG2P picokdt_getDtG2P (picoknow_KnowledgeBase this);
picokdt_DtPHR picokdt_getDtPHR (picoknow_KnowledgeBase this);
picokdt_DtACC picokdt_getDtACC (picoknow_KnowledgeBase this);
picokdt_DtPAM picokdt_getDtPAM (picoknow_KnowledgeBase this);
/* number of attributes (= input vector size) for each tree type */
typedef enum {
PICOKDT_NRATT_POSP = 12,
PICOKDT_NRATT_POSD = 7,
PICOKDT_NRATT_G2P = 16,
PICOKDT_NRATT_PHR = 8,
PICOKDT_NRATT_ACC = 13,
PICOKDT_NRATT_PAM = 60
} kdt_nratt_t;
/* ************************************************************/
/* decision tree classification result type */
/* ************************************************************/
typedef struct {
picoos_uint8 set; /* TRUE if class set, FALSE otherwise */
picoos_uint16 class;
} picokdt_classify_result_t;
/* maximum number of output values the tree output is mapped to */
#define PICOKDT_MAXSIZE_OUTVEC 8
typedef struct {
picoos_uint8 nr; /* 0 if no class set, nr of values set otherwise */
picoos_uint16 classvec[PICOKDT_MAXSIZE_OUTVEC];
} picokdt_classify_vecresult_t;
/* ************************************************************/
/* decision tree functions */
/* ************************************************************/
/* constructInVec:
for every tree type there is a constructInVec function to construct
the size-optimized input vector for the tree using the input map
tables that are part of the decistion tree knowledge base. The
constructed input vector is stored in the tree object (this->invec
and this->inveclen) and will be used in the following call to the
classify function.
classify:
for every tree type there is a classify function to apply the
decision tree to the previously constructed input vector. The
size-optimized, encoded output is stored in the tree object
(this->outval) and will be used in the following call to the
decompose function. Where needed (hitory attribute) the direct tree
output is returned by the classify function in a variable.
decomposeOutClass:
for every tree type there is a decompose function to decompose the
size-optimized, encoded tree output and map it to the outside the
tree usable class value.
*/
/* ************************************************************/
/* decision tree defines */
/* ************************************************************/
/* to construct the input vectors several hard-coded values are used
to handle attributes that, at the given position, are outside the
context. */
/* graph attributes: values to be used if the graph attribute is
outside the grapheme string (ie. word) */
#define PICOKDT_OUTSIDEGRAPH_DEFCH (picoos_uint8)'\x30' /* ascii "0" */
#define PICOKDT_OUTSIDEGRAPH_DEFSTR (picoos_uint8 *)"\x30" /* ascii "0" */
#define PICOKDT_OUTSIDEGRAPH_DEFLEN 1
/* graph attributes (special case for g2p): values to be used if the
graph attribute is directly outside the grapheme string (ie. at the
word boundary word). Use PICOKDT_OUTSIDEGRAPH_DEF* if further
outside. */
#define PICOKDT_OUTSIDEGRAPH_EOW_DEFCH (picoos_uint8)'\x31' /* ascii "1" */
#define PICOKDT_OUTSIDEGRAPH_EOW_DEFSTR (picoos_uint8 *)"\x31" /* ascii "1" */
#define PICOKDT_OUTSIDEGRAPH_EOW_DEFLEN 1
/* byte and word type attributes: value to be used if a byte or word
attribute is outside the context, e.g. for POS */
#define PICOKDT_EPSILON 7
/* byte and word type attributes: for attribute with history info a
'zero' value is needed when starting the sequence of predictions.
Use the following value to initialize history. Note that the direct
tree outputs (not mapped with output map table) of previous
predictions need to be used when constructing the input vector for
a following prediction. This direct tree output will then be mapped
together with the rest of the input vector by the input map
table. */
#define PICOKDT_HISTORY_ZERO 30000
/* ************************************************************/
/* decision tree POS prediction (PosP) functions */
/* ************************************************************/
/* construct a POS prediction input vector
tree input vector: 0-3 prefix UTF8 graphemes
4-9 suffex UTF8 graphemes
10 special grapheme existence flag (TRUE/FALSE)
11 number of graphemes
graph: the grapheme string of the word for wich POS will be predicted
graphlen: length of graph in number of bytes
specgraphflag: existence of a special grapheme boolean
returns: TRUE if okay, FALSE otherwise
note: use PICOKDT_OUTSIDEGRAPH* for att values outside context
*/
picoos_uint8 picokdt_dtPosPconstructInVec(const picokdt_DtPosP this,
const picoos_uint8 *graph,
const picoos_uint16 graphlen,
const picoos_uint8 specgraphflag);
/* classify a previously constructed input vector using tree 'this'
returns: TRUE if okay, FALSE otherwise
*/
picoos_uint8 picokdt_dtPosPclassify(const picokdt_DtPosP this);
/* decompose the tree output and return the class in dtres
dtres: POS or POSgroup ID classification result
returns: TRUE if okay, FALSE otherwise
*/
picoos_uint8 picokdt_dtPosPdecomposeOutClass(const picokdt_DtPosP this,
picokdt_classify_result_t *dtres);
/* ************************************************************/
/* decision tree POS disambiguation (PosD) functions */
/* ************************************************************/
/* construct a POS disambiguation input vector (run in left-to-right mode)
tree input vector: 0-2 POS or POSgroup for each of the three previous words
3 POSgroup for current word
4-6 POS or POSgroup (can be history) for each of
the three following words
pre3 - pre1: POSgroup or POS for the previous three words
src: POSgroup of current word (if unique POS no posdisa possible)
fol1 - fol3: POS or history for the following three words (the more
complicated the better... :-( NEEDS TO BE uint16
ishist1-ishist3: flag to indicate if fol1-3 are predicted tree
output values (history) or the HISTORY_ZERO (TRUE)
or an already unambiguous POS (FALSE)
returns: TRUE if okay, FALSE otherwise
note: use PICOKDT_EPSILON for att values outside context,
if POS in fol* unique use this POS instead of real
history, use reverse output mapping in these cases
*/
picoos_uint8 picokdt_dtPosDconstructInVec(const picokdt_DtPosD this,
const picoos_uint16 * input);
/* classify a previously constructed input vector using tree 'this'
treeout: direct tree output value
returns: TRUE if okay, FALSE otherwise
*/
picoos_uint8 picokdt_dtPosDclassify(const picokdt_DtPosD this,
picoos_uint16 *treeout);
/* decompose the tree output and return the class in dtres
dtres: POS classification result
returns: TRUE if okay, FALSE otherwise
*/
picoos_uint8 picokdt_dtPosDdecomposeOutClass(const picokdt_DtPosD this,
picokdt_classify_result_t *dtres);
/* convert (unique) POS index into corresponding tree output index */
picoos_uint8 picokdt_dtPosDreverseMapOutFixed(const picokdt_DtPosD this,
const picoos_uint16 inval,
picoos_uint16 *outval,
picoos_uint16 *outfallbackval);
/* ************************************************************/
/* decision tree grapheme-to-phoneme (G2P) functions */
/* ************************************************************/
/* construct a G2P input vector (run in right-to-left mode)
tree input vector: 0-8 the 4 previous, current, and 4 following graphemes
9 POS
10-11 vowel count and vowel ID
12 primary stress flag (TRUE/FALSE)
13-15 the three following phones predicted
graph: the grapheme string used to determine invec[0:8]
graphlen: length of graph in number of bytes
count: the grapheme number for which invec will be constructed [0..]
pos: the part of speech of the word
nrvow number of vowel-like graphemes in graph if vowel,
set to 0 otherwise
ordvow order of 'count' vowel in graph if vowel,
set to 0 otherwise
primstressflag: flag indicating if primary stress was already predicted
phonech1-3: the three following phon chunks predicted (right-to-left)
returns: TRUE if okay, FALSE otherwise
*/
picoos_uint8 picokdt_dtG2PconstructInVec(const picokdt_DtG2P this,
const picoos_uint8 *graph,
const picoos_uint16 graphlen,
const picoos_uint8 count,
const picoos_uint8 pos,
const picoos_uint8 nrvow,
const picoos_uint8 ordvow,
picoos_uint8 *primstressflag,
const picoos_uint16 phonech1,
const picoos_uint16 phonech2,
const picoos_uint16 phonech3);
/* classify a previously constructed input vector using tree 'this'
treeout: direct tree output value
returns: TRUE if okay, FALSE otherwise
*/
picoos_uint8 picokdt_dtG2Pclassify(const picokdt_DtG2P this,
picoos_uint16 *treeout);
/* decompose the tree output and return the class vector in dtvres
dtvres: phones vector classification result
returns: TRUE if okay, FALSE otherwise
*/
picoos_uint8 picokdt_dtG2PdecomposeOutClass(const picokdt_DtG2P this,
picokdt_classify_vecresult_t *dtvres);
/* ************************************************************/
/* decision tree phrasing (PHR) functions */
/* ************************************************************/
/* construct a PHR input vector (run in right-to-left mode)
tree input vector: 0-1 POS for each of the two previous words
2 POS for current word
3-4 POS for each of the two following words
5 nr words left
6 nr words right
7 nr syllables right
pre2 - pre1: POS for the previous two words
src: POS of current word
fol1 - fol2: POS for the following two words
nrwordspre: number of words left (previous) of current word
nrwordsfol: number of words right (following) of current word,
incl. current word, up to next BOUND (also
considering previously predicted PHR2/3)
nrsyllsfol: number of syllables right (following) of current word,
incl. syllables of current word, up to next BOUND
(also considering previously predicted PHR2/3)
returns: TRUE if okay, FALSE otherwise
note: use PICOKDT_EPSILON for att values outside context
*/
picoos_uint8 picokdt_dtPHRconstructInVec(const picokdt_DtPHR this,
const picoos_uint8 pre2,
const picoos_uint8 pre1,
const picoos_uint8 src,
const picoos_uint8 fol1,
const picoos_uint8 fol2,
const picoos_uint16 nrwordspre,
const picoos_uint16 nrwordsfol,
const picoos_uint16 nrsyllsfol);
/* classify a previously constructed input vector using tree 'this'
returns: TRUE if okay, FALSE otherwise
*/
picoos_uint8 picokdt_dtPHRclassify(const picokdt_DtPHR this);
/* decompose the tree output and return the class vector in dtres
dtres: phrasing classification result
returns: TRUE if okay, FALSE otherwise
*/
picoos_uint8 picokdt_dtPHRdecomposeOutClass(const picokdt_DtPHR this,
picokdt_classify_result_t *dtres);
/* ************************************************************/
/* decision tree accentuation (ACC) functions */
/* ************************************************************/
/* construct an ACC input vector (run in right-to-left mode)
tree input vector: 0-1 POS for each of the two previous words
2 POS for current word
3-4 POS for each of the two following words
5-6 history values (already predicted following)
7 nr words left (previous) to any bound
8 nr syllables left to any bound
9 nr words right (following) to any bound
10 nr syllables right to any bound
11 nr words right to predicted "1" prominence (foot)
12 nr syllables right to predicted "1" prominence (foot)
pre2 - pre1: POS for the previous two words
src: POS of current word
fol1 - fol2: POS for the following two words
hist1 - hist2: previously predicted ACC values
nrwordspre: number of words left (previous) of current word
nrsyllspre: number of syllables left (previous) of current word,
incl. initial non-prim stress syllables of current word
nrwordsfol: number of words right (following) of current word,
incl. current word, up to next BOUND (any strength != 0)
nrsyllsfol: number of syllables right (following) of current word,
incl. syllables of current word starting with prim. stress
syllable
footwordsfol: nr of words to the following prominence '1'
footsyllspre: nr of syllables to the previous prominence '1'
returns: TRUE if okay, FALSE otherwise
note: use PICOKDT_EPSILON for att 0-4 values outside context
*/
picoos_uint8 picokdt_dtACCconstructInVec(const picokdt_DtACC this,
const picoos_uint8 pre2,
const picoos_uint8 pre1,
const picoos_uint8 src,
const picoos_uint8 fol1,
const picoos_uint8 fol2,
const picoos_uint16 hist1,
const picoos_uint16 hist2,
const picoos_uint16 nrwordspre,
const picoos_uint16 nrsyllspre,
const picoos_uint16 nrwordsfol,
const picoos_uint16 nrsyllsfol,
const picoos_uint16 footwordsfol,
const picoos_uint16 footsyllsfol);
/* classify a previously constructed input vector using tree 'this'
treeout: direct tree output value
returns: TRUE if okay, FALSE otherwise
*/
picoos_uint8 picokdt_dtACCclassify(const picokdt_DtACC this,
picoos_uint16 *treeout);
/* decompose the tree output and return the class vector in dtres
dtres: phrasing classification result
returns: TRUE if okay, FALSE otherwise
*/
picoos_uint8 picokdt_dtACCdecomposeOutClass(const picokdt_DtACC this,
picokdt_classify_result_t *dtres);
/* ************************************************************/
/* decision tree phono-acoustical model (PAM) functions */
/* ************************************************************/
/* construct a Pam input vector and store the tree-specific encoded
input vector in the tree object.
vec: tree input vector, 60 single-byte-sized attributes
veclen: length of vec in number of bytes
returns: TRUE if okay, FALSE otherwise
*/
picoos_uint8 picokdt_dtPAMconstructInVec(const picokdt_DtPAM this,
const picoos_uint8 *vec,
const picoos_uint8 veclen);
/* classify a previously constructed input vector using tree 'this'
returns: TRUE if okay, FALSE otherwise
*/
picoos_uint8 picokdt_dtPAMclassify(const picokdt_DtPAM this);
/* decompose the tree output and return the class in dtres
dtres: phones vector classification result
returns: TRUE if okay, FALSE otherwise
*/
picoos_uint8 picokdt_dtPAMdecomposeOutClass(const picokdt_DtPAM this,
picokdt_classify_result_t *dtres);
#ifdef __cplusplus
}
#endif
#endif /*PICOKDT_H_*/
|