All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Modules Pages
gmm-acc-stats-twofeats.cc File Reference
Include dependency graph for gmm-acc-stats-twofeats.cc:

Go to the source code of this file.

Functions

int main (int argc, char *argv[])
 

Function Documentation

int main ( int  argc,
char *  argv[] 
)

Definition at line 31 of file gmm-acc-stats-twofeats.cc.

References TransitionModel::Accumulate(), AccumAmDiagGmm::AccumulateForGmmTwofeats(), kaldi::ConvertPosteriorToPdfs(), SequentialTableReader< Holder >::Done(), ParseOptions::GetArg(), RandomAccessTableReader< Holder >::HasKey(), rnnlm::i, AccumAmDiagGmm::Init(), TransitionModel::InitStats(), rnnlm::j, KALDI_ASSERT, KALDI_LOG, KALDI_WARN, SequentialTableReader< Holder >::Key(), kaldi::kGmmAll, SequentialTableReader< Holder >::Next(), ParseOptions::NumArgs(), MatrixBase< Real >::NumCols(), MatrixBase< Real >::NumRows(), ParseOptions::PrintUsage(), AmDiagGmm::Read(), ParseOptions::Read(), TransitionModel::Read(), ParseOptions::Register(), MatrixBase< Real >::Row(), Output::Stream(), Input::Stream(), RandomAccessTableReader< Holder >::Value(), SequentialTableReader< Holder >::Value(), AccumAmDiagGmm::Write(), and VectorBase< Real >::Write().

31  {
32  using namespace kaldi;
33  try {
34  const char *usage =
35  "Accumulate stats for GMM training, computing posteriors with one set of features\n"
36  "but accumulating statistics with another.\n"
37  "First features are used to get posteriors, second to accumulate stats\n"
38  "Usage: gmm-acc-stats-twofeats [options] <model-in> <feature1-rspecifier> <feature2-rspecifier> <posteriors-rspecifier> <stats-out>\n"
39  "e.g.: \n"
40  " gmm-acc-stats-twofeats 1.mdl 1.ali scp:train.scp scp:train_new.scp ark:1.ali 1.acc\n";
41 
42  ParseOptions po(usage);
43  bool binary = true;
44  po.Register("binary", &binary, "Write output in binary mode");
45  po.Read(argc, argv);
46 
47  if (po.NumArgs() != 5) {
48  po.PrintUsage();
49  exit(1);
50  }
51 
52  std::string model_filename = po.GetArg(1),
53  feature1_rspecifier = po.GetArg(2),
54  feature2_rspecifier = po.GetArg(3),
55  posteriors_rspecifier = po.GetArg(4),
56  accs_wxfilename = po.GetArg(5);
57 
58  using namespace kaldi;
59  typedef kaldi::int32 int32;
60 
61  AmDiagGmm am_gmm;
62  TransitionModel trans_model;
63  {
64  bool binary;
65  Input ki(model_filename, &binary);
66  trans_model.Read(ki.Stream(), binary);
67  am_gmm.Read(ki.Stream(), binary);
68  }
69 
70  Vector<double> transition_accs;
71  trans_model.InitStats(&transition_accs);
72  int32 new_dim = 0;
73  AccumAmDiagGmm gmm_accs;
74  // will initialize once we know new_dim.
75 
76  double tot_like = 0.0;
77  double tot_t = 0.0;
78 
79  SequentialBaseFloatMatrixReader feature1_reader(feature1_rspecifier);
80  RandomAccessBaseFloatMatrixReader feature2_reader(feature2_rspecifier);
81  RandomAccessPosteriorReader posteriors_reader(posteriors_rspecifier);
82 
83  int32 num_done = 0, num_no2ndfeats = 0, num_no_posterior = 0, num_other_error = 0;
84  for (; !feature1_reader.Done(); feature1_reader.Next()) {
85  std::string key = feature1_reader.Key();
86  if (!feature2_reader.HasKey(key)) {
87  KALDI_WARN << "For utterance " << key << ", second features not present.";
88  num_no2ndfeats ++;
89  } else if (!posteriors_reader.HasKey(key)) {
90  num_no_posterior++;
91  } else {
92  const Matrix<BaseFloat> &mat1 = feature1_reader.Value();
93  const Matrix<BaseFloat> &mat2 = feature2_reader.Value(key);
94  KALDI_ASSERT(mat1.NumRows() == mat2.NumRows());
95  if (new_dim == 0) {
96  new_dim = mat2.NumCols();
97  gmm_accs.Init(am_gmm, new_dim, kGmmAll);
98  }
99  const Posterior &posterior = posteriors_reader.Value(key);
100 
101  if (posterior.size() != mat1.NumRows()) {
102  KALDI_WARN << "Posteriors has wrong size "<< (posterior.size()) << " vs. "<< (mat1.NumRows());
103  num_other_error++;
104  continue;
105  }
106  if (mat1.NumRows() != mat2.NumRows()) {
107  KALDI_WARN << "Features have mismatched numbers of frames "
108  << mat1.NumRows() << " vs. " << mat2.NumRows();
109  num_other_error++;
110  continue;
111  }
112 
113  num_done++;
114  BaseFloat tot_like_this_file = 0.0,
115  tot_weight_this_file = 0.0;
116 
117  Posterior pdf_posterior;
118  ConvertPosteriorToPdfs(trans_model, posterior, &pdf_posterior);
119  for (size_t i = 0; i < posterior.size(); i++) {
120  // Accumulates for GMM.
121  for (size_t j = 0; j <pdf_posterior[i].size(); j++) {
122  int32 pdf_id = pdf_posterior[i][j].first;
123  BaseFloat weight = pdf_posterior[i][j].second;
124  tot_like_this_file += weight *
125  gmm_accs.AccumulateForGmmTwofeats(am_gmm,
126  mat1.Row(i),
127  mat2.Row(i),
128  pdf_id,
129  weight);
130  tot_weight_this_file += weight;
131  }
132 
133  // Accumulates for transitions.
134  for (size_t j = 0; j < posterior[i].size(); j++) {
135  int32 tid = posterior[i][j].first;
136  BaseFloat weight = posterior[i][j].second;
137  trans_model.Accumulate(weight, tid, &transition_accs);
138  }
139  }
140  KALDI_LOG << "Average like for this file is "
141  << (tot_like_this_file/tot_weight_this_file) << " over "
142  << tot_weight_this_file <<" frames.";
143  tot_like += tot_like_this_file;
144  tot_t += tot_weight_this_file;
145  if (num_done % 10 == 0)
146  KALDI_LOG << "Avg like per frame so far is " << (tot_like/tot_t);
147  }
148  }
149 
150  KALDI_LOG << "Done " << num_done << " files, " << num_no_posterior
151  << " with no posteriors, " << num_no2ndfeats
152  << " with no second features, " << num_other_error
153  << " with other errors.";
154 
155  KALDI_LOG << "Overall avg like per frame (Gaussian only) = "
156  << (tot_like/tot_t) << " over " << tot_t << " frames.";
157 
158  {
159  Output ko(accs_wxfilename, binary);
160  transition_accs.Write(ko.Stream(), binary);
161  gmm_accs.Write(ko.Stream(), binary);
162  }
163  KALDI_LOG << "Written accs.";
164  if (num_done != 0) return 0;
165  else return 1;
166  } catch(const std::exception &e) {
167  std::cerr << e.what();
168  return -1;
169  }
170 }
Relabels neural network egs with the read pdf-id alignments.
Definition: chain.dox:20
void Write(std::ostream &Out, bool binary) const
Writes to C++ stream (option to write in binary).
void InitStats(Vector< double > *stats) const
BaseFloat AccumulateForGmmTwofeats(const AmDiagGmm &model, const VectorBase< BaseFloat > &data1, const VectorBase< BaseFloat > &data2, int32 gmm_index, BaseFloat weight)
Accumulate stats for a single GMM in the model; uses data1 for getting posteriors and data2 for stats...
const SubVector< Real > Row(MatrixIndexT i) const
Return specific row of matrix [const].
Definition: kaldi-matrix.h:182
Allows random access to a collection of objects in an archive or script file; see The Table concept...
Definition: kaldi-table.h:233
void Accumulate(BaseFloat prob, int32 trans_id, Vector< double > *stats) const
float BaseFloat
Definition: kaldi-types.h:29
std::vector< std::vector< std::pair< int32, BaseFloat > > > Posterior
Posterior is a typedef for storing acoustic-state (actually, transition-id) posteriors over an uttera...
Definition: posterior.h:43
void Write(std::ostream &out_stream, bool binary) const
The class ParseOptions is for parsing command-line options; see Parsing command-line options for more...
Definition: parse-options.h:36
void Read(std::istream &is, bool binary)
A templated class for reading objects sequentially from an archive or script file; see The Table conc...
Definition: kaldi-table.h:287
#define KALDI_WARN
Definition: kaldi-error.h:130
MatrixIndexT NumRows() const
Returns number of rows (or zero for emtpy matrix).
Definition: kaldi-matrix.h:58
MatrixIndexT NumCols() const
Returns number of columns (or zero for emtpy matrix).
Definition: kaldi-matrix.h:61
#define KALDI_ASSERT(cond)
Definition: kaldi-error.h:169
void ConvertPosteriorToPdfs(const TransitionModel &tmodel, const Posterior &post_in, Posterior *post_out)
Converts a posterior over transition-ids to be a posterior over pdf-ids.
Definition: posterior.cc:322
#define KALDI_LOG
Definition: kaldi-error.h:133
void Read(std::istream &in_stream, bool binary)
Definition: am-diag-gmm.cc:147
void Init(const AmDiagGmm &model, GmmFlagsType flags)
Initializes accumulators for each GMM based on the number of components and dimension.