gmm-align-compiled.cc
5.57 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
// gmmbin/gmm-align-compiled.cc
// Copyright 2009-2013 Microsoft Corporation
// 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 "base/kaldi-common.h"
#include "util/common-utils.h"
#include "gmm/am-diag-gmm.h"
#include "hmm/transition-model.h"
#include "hmm/hmm-utils.h"
#include "fstext/fstext-lib.h"
#include "decoder/decoder-wrappers.h"
#include "gmm/decodable-am-diag-gmm.h"
#include "lat/kaldi-lattice.h" // for {Compact}LatticeArc
int main(int argc, char *argv[]) {
try {
using namespace kaldi;
typedef kaldi::int32 int32;
using fst::SymbolTable;
using fst::VectorFst;
using fst::StdArc;
const char *usage =
"Align features given [GMM-based] models.\n"
"Usage: gmm-align-compiled [options] <model-in> <graphs-rspecifier> "
"<feature-rspecifier> <alignments-wspecifier> [scores-wspecifier]\n"
"e.g.: \n"
" gmm-align-compiled 1.mdl ark:graphs.fsts scp:train.scp ark:1.ali\n"
"or:\n"
" compile-train-graphs tree 1.mdl lex.fst 'ark:sym2int.pl -f 2- words.txt text|' \\\n"
" ark:- | gmm-align-compiled 1.mdl ark:- scp:train.scp t, ark:1.ali\n";
ParseOptions po(usage);
AlignConfig align_config;
BaseFloat acoustic_scale = 1.0;
BaseFloat transition_scale = 1.0;
BaseFloat self_loop_scale = 1.0;
std::string per_frame_acwt_wspecifier;
align_config.Register(&po);
po.Register("transition-scale", &transition_scale,
"Transition-probability scale [relative to acoustics]");
po.Register("acoustic-scale", &acoustic_scale,
"Scaling factor for acoustic likelihoods");
po.Register("self-loop-scale", &self_loop_scale,
"Scale of self-loop versus non-self-loop log probs [relative to acoustics]");
po.Register("write-per-frame-acoustic-loglikes", &per_frame_acwt_wspecifier,
"Wspecifier for table of vectors containing the acoustic log-likelihoods "
"per frame for each utterance. E.g. ark:foo/per_frame_logprobs.1.ark");
po.Read(argc, argv);
if (po.NumArgs() < 4 || po.NumArgs() > 5) {
po.PrintUsage();
exit(1);
}
std::string model_in_filename = po.GetArg(1),
fst_rspecifier = po.GetArg(2),
feature_rspecifier = po.GetArg(3),
alignment_wspecifier = po.GetArg(4),
scores_wspecifier = po.GetOptArg(5);
TransitionModel trans_model;
AmDiagGmm am_gmm;
{
bool binary;
Input ki(model_in_filename, &binary);
trans_model.Read(ki.Stream(), binary);
am_gmm.Read(ki.Stream(), binary);
}
SequentialTableReader<fst::VectorFstHolder> fst_reader(fst_rspecifier);
RandomAccessBaseFloatMatrixReader feature_reader(feature_rspecifier);
Int32VectorWriter alignment_writer(alignment_wspecifier);
BaseFloatWriter scores_writer(scores_wspecifier);
BaseFloatVectorWriter per_frame_acwt_writer(per_frame_acwt_wspecifier);
int num_done = 0, num_err = 0, num_retry = 0;
double tot_like = 0.0;
kaldi::int64 frame_count = 0;
for (; !fst_reader.Done(); fst_reader.Next()) {
std::string utt = fst_reader.Key();
if (!feature_reader.HasKey(utt)) {
num_err++;
KALDI_WARN << "No features for utterance " << utt;
} else {
const Matrix<BaseFloat> &features = feature_reader.Value(utt);
VectorFst<StdArc> decode_fst(fst_reader.Value());
fst_reader.FreeCurrent(); // this stops copy-on-write of the fst
// by deleting the fst inside the reader, since we're about to mutate
// the fst by adding transition probs.
if (features.NumRows() == 0) {
KALDI_WARN << "Zero-length utterance: " << utt;
num_err++;
continue;
}
{ // Add transition-probs to the FST.
std::vector<int32> disambig_syms; // empty.
AddTransitionProbs(trans_model, disambig_syms,
transition_scale, self_loop_scale,
&decode_fst);
}
DecodableAmDiagGmmScaled gmm_decodable(am_gmm, trans_model, features,
acoustic_scale);
KALDI_LOG << utt;
AlignUtteranceWrapper(align_config, utt,
acoustic_scale, &decode_fst, &gmm_decodable,
&alignment_writer, &scores_writer,
&num_done, &num_err, &num_retry,
&tot_like, &frame_count, &per_frame_acwt_writer);
}
}
KALDI_LOG << "Overall log-likelihood per frame is " << (tot_like/frame_count)
<< " over " << frame_count<< " frames.";
KALDI_LOG << "Retried " << num_retry << " out of "
<< (num_done + num_err) << " utterances.";
KALDI_LOG << "Done " << num_done << ", errors on " << num_err;
return (num_done != 0 ? 0 : 1);
} catch(const std::exception &e) {
std::cerr << e.what();
return -1;
}
}