regtree-mllr-diag-gmm.cc
16.3 KB
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
// transform/regtree-mllr-diag-gmm.cc
// Copyright 2009-2011 Saarland University; Jan Silovsky
// See ../../COPYING for clarification regarding multiple authors
//
// 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
//
// THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED
// WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE,
// MERCHANTABLITY OR NON-INFRINGEMENT.
// See the Apache 2 License for the specific language governing permissions and
// limitations under the License.
#include <utility>
using std::pair;
#include <vector>
using std::vector;
#include "transform/regtree-mllr-diag-gmm.h"
namespace kaldi {
void RegtreeMllrDiagGmm::Init(int32 num_xforms, int32 dim) {
if (num_xforms == 0) { // empty transform
xform_matrices_.clear();
dim_ = 0; // non-zero dimension is meaningless with empty transform
num_xforms_ = 0;
bclass2xforms_.clear();
} else {
KALDI_ASSERT(dim != 0); // if not empty, dim = 0 is meaningless
dim_ = dim;
num_xforms_ = num_xforms;
xform_matrices_.resize(num_xforms);
vector< Matrix<BaseFloat> >::iterator xform_itr = xform_matrices_.begin(),
xform_itr_end = xform_matrices_.end();
for (; xform_itr != xform_itr_end; ++xform_itr) {
xform_itr->Resize(dim, dim+1);
xform_itr->SetUnit();
}
}
}
void RegtreeMllrDiagGmm::SetUnit() {
vector< Matrix<BaseFloat> >::iterator xform_itr = xform_matrices_.begin(),
xform_itr_end = xform_matrices_.end();
for (; xform_itr != xform_itr_end; ++xform_itr) {
xform_itr->SetUnit();
}
}
void RegtreeMllrDiagGmm::TransformModel(const RegressionTree ®tree,
AmDiagGmm *am) {
KALDI_ASSERT(static_cast<int32>(bclass2xforms_.size()) ==
regtree.NumBaseclasses());
Vector<BaseFloat> extended_mean(dim_+1), xformed_mean(dim_);
for (int32 bclass_index = 0, num_bclasses = regtree.NumBaseclasses();
bclass_index < num_bclasses; ++bclass_index) {
int32 xform_index;
if ((xform_index = bclass2xforms_[bclass_index]) > -1) {
KALDI_ASSERT(xform_index < num_xforms_);
const vector< pair<int32, int32> > &bclass =
regtree.GetBaseclass(bclass_index);
for (vector< pair<int32, int32> >::const_iterator itr = bclass.begin(),
end = bclass.end(); itr != end; ++itr) {
SubVector<BaseFloat> tmp_mean(extended_mean.Range(0, dim_));
am->GetGaussianMean(itr->first, itr->second, &tmp_mean);
extended_mean(dim_) = 1.0;
xformed_mean.AddMatVec(1.0, xform_matrices_[xform_index], kNoTrans,
extended_mean, 0.0);
am->SetGaussianMean(itr->first, itr->second, xformed_mean);
} // end iterating over Gaussians in baseclass
} // else keep the means untransformed
} // end iterating over all baseclasses
am->ComputeGconsts();
}
void RegtreeMllrDiagGmm::GetTransformedMeans(const RegressionTree ®tree,
const AmDiagGmm &am,
int32 pdf_index,
MatrixBase<BaseFloat> *out) const {
KALDI_ASSERT(static_cast<int32>(bclass2xforms_.size()) ==
regtree.NumBaseclasses());
int32 num_gauss = am.GetPdf(pdf_index).NumGauss();
KALDI_ASSERT(out->NumRows() == num_gauss && out->NumCols() == dim_);
Vector<BaseFloat> extended_mean(dim_+1);
extended_mean(dim_) = 1.0;
for (int32 gauss_index = 0; gauss_index < num_gauss; gauss_index++) {
int32 bclass_index = regtree.Gauss2BaseclassId(pdf_index, gauss_index);
int32 xform_index = bclass2xforms_[bclass_index];
if (xform_index > -1) { // use a transform
KALDI_ASSERT(xform_index < num_xforms_);
SubVector<BaseFloat> tmp_mean(extended_mean.Range(0, dim_));
am.GetGaussianMean(pdf_index, gauss_index, &tmp_mean);
SubVector<BaseFloat> out_row(out->Row(gauss_index));
out_row.AddMatVec(1.0, xform_matrices_[xform_index], kNoTrans,
extended_mean, 0.0);
} else { // Copy untransformed mean
SubVector<BaseFloat> out_row(out->Row(gauss_index));
am.GetGaussianMean(pdf_index, gauss_index, &out_row);
}
}
}
void RegtreeMllrDiagGmm::Write(std::ostream &out, bool binary) const {
WriteToken(out, binary, "<MLLRXFORM>");
WriteToken(out, binary, "<NUMXFORMS>");
WriteBasicType(out, binary, num_xforms_);
WriteToken(out, binary, "<DIMENSION>");
WriteBasicType(out, binary, dim_);
vector< Matrix<BaseFloat> >::const_iterator xform_itr =
xform_matrices_.begin(), xform_itr_end = xform_matrices_.end();
for (; xform_itr != xform_itr_end; ++xform_itr) {
WriteToken(out, binary, "<XFORM>");
xform_itr->Write(out, binary);
}
WriteToken(out, binary, "<BCLASS2XFORMS>");
WriteIntegerVector(out, binary, bclass2xforms_);
WriteToken(out, binary, "</MLLRXFORM>");
}
void RegtreeMllrDiagGmm::Read(std::istream &in, bool binary) {
ExpectToken(in, binary, "<MLLRXFORM>");
ExpectToken(in, binary, "<NUMXFORMS>");
ReadBasicType(in, binary, &num_xforms_);
ExpectToken(in, binary, "<DIMENSION>");
ReadBasicType(in, binary, &dim_);
KALDI_ASSERT(num_xforms_ >= 0 && dim_ >= 0); // can be 0 for empty xform
xform_matrices_.resize(num_xforms_);
vector< Matrix<BaseFloat> >::iterator xform_itr = xform_matrices_.begin(),
xform_itr_end = xform_matrices_.end();
for (; xform_itr != xform_itr_end; ++xform_itr) {
ExpectToken(in, binary, "<XFORM>");
xform_itr->Read(in, binary);
KALDI_ASSERT(xform_itr->NumRows() == (xform_itr->NumCols() - 1)
&& xform_itr->NumRows() == dim_);
}
ExpectToken(in, binary, "<BCLASS2XFORMS>");
ReadIntegerVector(in, binary, &bclass2xforms_);
ExpectToken(in, binary, "</MLLRXFORM>");
}
// ************************************************************************
void RegtreeMllrDiagGmmAccs::Init(int32 num_bclass, int32 dim) {
if (num_bclass == 0) { // empty stats
DeletePointers(&baseclass_stats_);
baseclass_stats_.clear();
num_baseclasses_ = 0;
dim_ = 0; // non-zero dimension is meaningless in empty stats
} else {
KALDI_ASSERT(dim != 0); // if not empty, dim = 0 is meaningless
num_baseclasses_ = num_bclass;
dim_ = dim;
baseclass_stats_.resize(num_baseclasses_);
for (vector<AffineXformStats*>::iterator it = baseclass_stats_.begin(),
end = baseclass_stats_.end(); it != end; ++it) {
*it = new AffineXformStats();
(*it)->Init(dim_, dim_);
}
}
}
void RegtreeMllrDiagGmmAccs::SetZero() {
for (vector<AffineXformStats*>::iterator it = baseclass_stats_.begin(),
end = baseclass_stats_.end(); it != end; ++it) {
(*it)->SetZero();
}
}
BaseFloat RegtreeMllrDiagGmmAccs::AccumulateForGmm(
const RegressionTree ®tree, const AmDiagGmm &am,
const VectorBase<BaseFloat> &data, int32 pdf_index, BaseFloat weight) {
const DiagGmm &pdf = am.GetPdf(pdf_index);
int32 num_comp = static_cast<int32>(pdf.NumGauss());
Vector<BaseFloat> posterior(num_comp);
BaseFloat loglike = pdf.ComponentPosteriors(data, &posterior);
posterior.Scale(weight);
Vector<double> posterior_d(posterior);
Vector<double> data_d(data);
Vector<double> inv_var_x(dim_);
Vector<double> extended_mean(dim_+1);
SpMatrix<double> mean_scatter(dim_+1);
for (int32 m = 0; m < num_comp; m++) {
unsigned bclass = regtree.Gauss2BaseclassId(pdf_index, m);
inv_var_x.CopyFromVec(pdf.inv_vars().Row(m));
inv_var_x.MulElements(data_d);
// Using SubVector to stop compiler warning
SubVector<double> tmp_mean(extended_mean, 0, dim_);
pdf.GetComponentMean(m, &tmp_mean); // modifies extended_mean
extended_mean(dim_) = 1.0;
mean_scatter.SetZero();
mean_scatter.AddVec2(1.0, extended_mean);
baseclass_stats_[bclass]->beta_ += posterior_d(m);
baseclass_stats_[bclass]->K_.AddVecVec(posterior_d(m), inv_var_x,
extended_mean);
vector< SpMatrix<double> > &G = baseclass_stats_[bclass]->G_;
for (int32 d = 0; d < dim_; d++)
G[d].AddSp((posterior_d(m) * pdf.inv_vars()(m, d)), mean_scatter);
}
return loglike;
}
void RegtreeMllrDiagGmmAccs::AccumulateForGaussian(
const RegressionTree ®tree, const AmDiagGmm &am,
const VectorBase<BaseFloat> &data, int32 pdf_index, int32 gauss_index,
BaseFloat weight) {
const DiagGmm &pdf = am.GetPdf(pdf_index);
Vector<double> data_d(data);
Vector<double> inv_var_x(dim_);
Vector<double> extended_mean(dim_+1);
double weight_d = static_cast<double>(weight);
unsigned bclass = regtree.Gauss2BaseclassId(pdf_index, gauss_index);
inv_var_x.CopyFromVec(pdf.inv_vars().Row(gauss_index));
inv_var_x.MulElements(data_d);
// Using SubVector to stop compiler warning
SubVector<double> tmp_mean(extended_mean, 0, dim_);
pdf.GetComponentMean(gauss_index, &tmp_mean); // modifies extended_mean
extended_mean(dim_) = 1.0;
SpMatrix<double> mean_scatter(dim_+1);
mean_scatter.AddVec2(1.0, extended_mean);
baseclass_stats_[bclass]->beta_ += weight_d;
baseclass_stats_[bclass]->K_.AddVecVec(weight_d, inv_var_x, extended_mean);
vector< SpMatrix<double> > &G = baseclass_stats_[bclass]->G_;
for (int32 d = 0; d < dim_; d++)
G[d].AddSp((weight_d * pdf.inv_vars()(gauss_index, d)), mean_scatter);
}
void RegtreeMllrDiagGmmAccs::Write(std::ostream &out, bool binary) const {
WriteToken(out, binary, "<MLLRACCS>");
WriteToken(out, binary, "<NUMBASECLASSES>");
WriteBasicType(out, binary, num_baseclasses_);
WriteToken(out, binary, "<DIMENSION>");
WriteBasicType(out, binary, dim_);
WriteToken(out, binary, "<STATS>");
vector<AffineXformStats*>::const_iterator itr = baseclass_stats_.begin(),
end = baseclass_stats_.end();
for ( ; itr != end; ++itr)
(*itr)->Write(out, binary);
WriteToken(out, binary, "</MLLRACCS>");
}
void RegtreeMllrDiagGmmAccs::Read(std::istream &in, bool binary, bool add) {
ExpectToken(in, binary, "<MLLRACCS>");
ExpectToken(in, binary, "<NUMBASECLASSES>");
ReadBasicType(in, binary, &num_baseclasses_);
ExpectToken(in, binary, "<DIMENSION>");
ReadBasicType(in, binary, &dim_);
KALDI_ASSERT(num_baseclasses_ > 0 && dim_ > 0);
baseclass_stats_.resize(num_baseclasses_);
ExpectToken(in, binary, "<STATS>");
vector<AffineXformStats*>::iterator itr = baseclass_stats_.begin(),
end = baseclass_stats_.end();
for ( ; itr != end; ++itr) {
*itr = new AffineXformStats();
(*itr)->Init(dim_, dim_);
(*itr)->Read(in, binary, add);
}
ExpectToken(in, binary, "</MLLRACCS>");
}
static void ComputeMllrMatrix(const Matrix<double> &K,
const vector< SpMatrix<double> > &G,
Matrix<BaseFloat> *out) {
int32 dim = G.size();
Matrix<double> tmp_out(dim, dim+1);
for (int32 d = 0; d < dim; d++) {
if (G[d].Cond() > 1.0e+9) {
KALDI_WARN << "Dim " << d << ": Badly conditioned stats. Setting MLLR "
<< "transform to unit.";
tmp_out.SetUnit();
break;
}
SpMatrix<double> inv_g(G[d]);
// KALDI_LOG << "Dim " << d << ": G: max = " << inv_g.Max() << ", min = "
// << inv_g.Min() << ", log det = " << inv_g.LogDet(NULL)
// << ", cond = " << inv_g.Cond();
inv_g.Invert();
// KALDI_LOG << "Inv G: max = " << inv_g.Max() << ", min = " << inv_g.Min()
// << ", log det = " << inv_g.LogDet(NULL) << ", cond = "
// << inv_g.Cond();
tmp_out.Row(d).AddSpVec(1.0, inv_g, K.Row(d), 0.0);
}
out->CopyFromMat(tmp_out, kNoTrans);
}
static BaseFloat MllrAuxFunction(const Matrix<BaseFloat> &xform,
const AffineXformStats &stats) {
int32 dim = stats.G_.size();
Matrix<double> xform_d(xform);
Vector<double> xform_row_g(dim + 1);
SubMatrix<double> A(xform_d, 0, dim, 0, dim);
double obj = TraceMatMat(xform_d, stats.K_, kTrans);
for (int32 d = 0; d < dim; d++) {
xform_row_g.AddSpVec(1.0, stats.G_[d], xform_d.Row(d), 0.0);
obj -= 0.5 * VecVec(xform_row_g, xform_d.Row(d));
}
return obj;
}
void RegtreeMllrDiagGmmAccs::Update(const RegressionTree ®tree,
const RegtreeMllrOptions &opts,
RegtreeMllrDiagGmm *out_mllr,
BaseFloat *auxf_impr,
BaseFloat *t) const {
BaseFloat tot_auxf_impr = 0, tot_t = 0;
Matrix<BaseFloat> xform_mat(dim_, dim_ + 1);
if (opts.use_regtree) { // estimate transforms using a regression tree
vector<AffineXformStats*> regclass_stats;
vector<int32> base2regclass;
bool update_xforms = regtree.GatherStats(baseclass_stats_, opts.min_count,
&base2regclass, ®class_stats);
out_mllr->set_bclass2xforms(base2regclass);
// If update_xforms == true, none should be negative, else all should be -1
if (update_xforms) {
out_mllr->Init(regclass_stats.size(), dim_);
for (int32 rclass_index = 0, num_rclass = regclass_stats.size();
rclass_index < num_rclass; ++rclass_index) {
KALDI_ASSERT(regclass_stats[rclass_index]->beta_ >= opts.min_count);
xform_mat.SetUnit();
BaseFloat obj_old = MllrAuxFunction(xform_mat,
*(regclass_stats[rclass_index]));
ComputeMllrMatrix(regclass_stats[rclass_index]->K_,
regclass_stats[rclass_index]->G_, &xform_mat);
out_mllr->SetParameters(xform_mat, rclass_index);
BaseFloat obj_new = MllrAuxFunction(xform_mat,
*(regclass_stats[rclass_index]));
KALDI_LOG << "MLLR: regclass " << (rclass_index)
<< ": Objective function impr per frame is "
<< ((obj_new - obj_old)/regclass_stats[rclass_index]->beta_)
<< " over " << regclass_stats[rclass_index]->beta_
<< " frames.";
KALDI_ASSERT(obj_new >= obj_old - (std::abs(obj_new)+std::abs(obj_old))*1.0e-05);
tot_t += regclass_stats[rclass_index]->beta_;
tot_auxf_impr += obj_new - obj_old;
}
} else {
out_mllr->Init(1, dim_); // Use a unit transform at the root.
}
DeletePointers(®class_stats);
// end of estimation using regression tree
} else { // estimate 1 transform per baseclass (if enough count)
out_mllr->Init(num_baseclasses_, dim_);
vector<int32> base2xforms(num_baseclasses_, -1);
for (int32 bclass_index = 0; bclass_index < num_baseclasses_;
++bclass_index) {
if (baseclass_stats_[bclass_index]->beta_ > opts.min_count) {
base2xforms[bclass_index] = bclass_index;
xform_mat.SetUnit();
BaseFloat obj_old = MllrAuxFunction(xform_mat,
*(baseclass_stats_[bclass_index]));
ComputeMllrMatrix(baseclass_stats_[bclass_index]->K_,
baseclass_stats_[bclass_index]->G_, &xform_mat);
out_mllr->SetParameters(xform_mat, bclass_index);
BaseFloat obj_new = MllrAuxFunction(xform_mat,
*(baseclass_stats_[bclass_index]));
KALDI_LOG << "MLLR: base-class " << (bclass_index)
<< ": Auxiliary function impr per frame is "
<< ((obj_new-obj_old)/baseclass_stats_[bclass_index]->beta_);
KALDI_ASSERT(obj_new >= obj_old - (std::abs(obj_new)+std::abs(obj_old))*1.0e-05);
tot_t += baseclass_stats_[bclass_index]->beta_;
tot_auxf_impr += obj_new - obj_old;
} else {
KALDI_WARN << "For baseclass " << (bclass_index) << " count = "
<< (baseclass_stats_[bclass_index]->beta_) << " < "
<< opts.min_count << ": not updating MLLR";
tot_t += baseclass_stats_[bclass_index]->beta_;
}
} // end looping over all baseclasses
out_mllr->set_bclass2xforms(base2xforms);
} // end of estimating one transform per baseclass
if (auxf_impr != NULL) *auxf_impr = tot_auxf_impr;
if (t != NULL) *t = tot_t;
}
} // namespace kaldi