full-gmm-normal.cc
4.14 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
// gmm/full-gmm-normal.cc
// Copyright 2009-2011 Microsoft Corporation; Saarland University;
// Yanmin Qian
// Univ. Erlangen-Nuremberg, Korbinian Riedhammer
// 2013 Johns Hopkins University (author: Daniel Povey)
// 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 <algorithm>
#include <limits>
#include <string>
#include <vector>
#include "gmm/full-gmm-normal.h"
#include "gmm/full-gmm.h"
namespace kaldi {
void FullGmmNormal::Resize(int32 nmix, int32 dim) {
KALDI_ASSERT(nmix > 0 && dim > 0);
if (weights_.Dim() != nmix)
weights_.Resize(nmix);
if (means_.NumRows() != nmix ||
means_.NumCols() != dim)
means_.Resize(nmix, dim);
if (vars_.size() != nmix)
vars_.resize(nmix);
for (int32 i = 0; i < nmix; i++) {
if (vars_[i].NumRows() != nmix ||
vars_[i].NumCols() != dim) {
vars_[i].Resize(dim);
}
}
}
void FullGmmNormal::CopyFromFullGmm(const FullGmm &fullgmm) {
/// resize the variables to fit the gmm
size_t dim = fullgmm.Dim();
size_t num_gauss = fullgmm.NumGauss();
Resize(num_gauss, dim);
/// copy weights
weights_.CopyFromVec(fullgmm.weights_);
/// we need to split the natural components for each gaussian
Vector<double> mean_times_invcovar(dim);
for (size_t i = 0; i < num_gauss; i++) {
// copy and invert (inverse) covariance matrix
vars_[i].CopyFromSp(fullgmm.inv_covars_[i]);
vars_[i].InvertDouble();
// multiply the (mean x icov) by (cov) to get the means back
mean_times_invcovar.CopyFromVec(fullgmm.means_invcovars_.Row(i));
(means_.Row(i)).AddSpVec(1.0, vars_[i], mean_times_invcovar, 0.0);
}
}
void FullGmmNormal::CopyToFullGmm(FullGmm *fullgmm, GmmFlagsType flags) {
KALDI_ASSERT(weights_.Dim() == fullgmm->weights_.Dim()
&& means_.NumCols() == fullgmm->Dim());
FullGmmNormal oldg(*fullgmm);
if (flags & kGmmWeights)
fullgmm->weights_.CopyFromVec(weights_);
size_t num_comp = fullgmm->NumGauss(), dim = fullgmm->Dim();
for (size_t i = 0; i < num_comp; i++) {
if (flags & kGmmVariances) {
fullgmm->inv_covars_[i].CopyFromSp(vars_[i]);
fullgmm->inv_covars_[i].InvertDouble();
// update the mean-related natural part with old mean, if necessary
if (!(flags & kGmmMeans)) {
Vector<BaseFloat> mean_times_inv(dim);
Vector<BaseFloat> mhelp(oldg.means_.Row(i));
mean_times_inv.AddSpVec(1.0, fullgmm->inv_covars_[i], mhelp, 0.0f);
fullgmm->means_invcovars_.Row(i).CopyFromVec(mean_times_inv);
}
}
if (flags & kGmmMeans) {
Vector<BaseFloat> mean_times_inv(dim), mean(means_.Row(i));
mean_times_inv.AddSpVec(1.0, fullgmm->inv_covars_[i], mean, 0.0f);
fullgmm->means_invcovars_.Row(i).CopyFromVec(mean_times_inv);
}
}
fullgmm->valid_gconsts_ = false;
}
void FullGmmNormal::Rand(MatrixBase<BaseFloat> *feats) {
int32 dim = means_.NumCols(), num_frames = feats->NumRows(),
num_gauss = means_.NumRows();
KALDI_ASSERT(feats->NumCols() == dim);
std::vector<TpMatrix<BaseFloat> > sqrt_var(num_gauss);
for (int32 i = 0; i < num_gauss; i++) {
sqrt_var[i].Resize(dim);
sqrt_var[i].Cholesky(SpMatrix<BaseFloat>(vars_[i]));
}
Vector<BaseFloat> rand(dim);
for (int32 t = 0; t < num_frames; t++) {
int32 i = weights_.RandCategorical(); // index with prob propto weights_[i].
SubVector<BaseFloat> frame(*feats, t);
frame.CopyFromVec(means_.Row(i));
rand.SetRandn();
frame.AddTpVec(1.0, sqrt_var[i], kNoTrans, rand, 1.0);
}
}
} // End namespace kaldi