Blame view
src/gmmbin/gmm-est-rescale.cc
3.62 KB
8dcb6dfcb
first commit
|
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 |
// gmmbin/gmm-est-rescale.cc
// Copyright 2012 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/indirect-diff-diag-gmm.h"
#include "tree/context-dep.h"
#include "hmm/transition-model.h"
int main(int argc, char *argv[]) {
using namespace kaldi;
typedef kaldi::int32 int32;
try {
const char *usage =
"Do \"re-scaling\" re-estimation of GMM-based model
"
" (this update changes the model as features change, but preserves
"
" the difference between the model and the features, to keep
"
" the effect of any prior discriminative training). Used in fMPE.
"
" Does not update the transitions or weights.
"
"Usage: gmm-est-rescale [options] <model-in> <old-stats-in> <new-stats-in> <model-out>
"
"e.g.: gmm-est-rescale 1.mdl old.acc new.acc 2.mdl
";
bool binary_write = true;
MleDiagGmmOptions opts; // Not passed to command-line-- just a mechanism to
// ensure our options have the same default values as those ones.
BaseFloat min_variance = opts.min_variance;
BaseFloat min_gaussian_occupancy = opts.min_gaussian_occupancy;
ParseOptions po(usage);
po.Register("binary", &binary_write, "Write output in binary mode");
po.Register("min-variance", &min_variance,
"Variance floor (absolute variance).");
po.Register("min-gaussian-occupancy", &min_gaussian_occupancy,
"Minimum occupancy to update a Gaussian.");
po.Read(argc, argv);
if (po.NumArgs() != 4) {
po.PrintUsage();
exit(1);
}
std::string model_rxfilename = po.GetArg(1),
old_stats_rxfilename = po.GetArg(2),
new_stats_rxfilename = po.GetArg(3),
model_wxfilename = po.GetArg(4);
AmDiagGmm am_gmm;
TransitionModel trans_model;
{
bool binary_read;
Input ki(model_rxfilename, &binary_read);
trans_model.Read(ki.Stream(), binary_read);
am_gmm.Read(ki.Stream(), binary_read);
}
AccumAmDiagGmm old_gmm_accs, new_gmm_accs;
{
Vector<double> transition_accs;
bool binary;
Input ki(old_stats_rxfilename, &binary);
transition_accs.Read(ki.Stream(), binary);
old_gmm_accs.Read(ki.Stream(), binary, true);
}
{
Vector<double> transition_accs;
bool binary;
Input ki(new_stats_rxfilename, &binary);
transition_accs.Read(ki.Stream(), binary);
new_gmm_accs.Read(ki.Stream(), binary, true);
}
DoRescalingUpdate(old_gmm_accs, new_gmm_accs,
min_variance, min_gaussian_occupancy,
&am_gmm);
{
Output ko(model_wxfilename, binary_write);
trans_model.Write(ko.Stream(), binary_write);
am_gmm.Write(ko.Stream(), binary_write);
}
KALDI_LOG << "Rescaled model and wrote to " << model_wxfilename;
return 0;
} catch(const std::exception &e) {
std::cerr << e.what() << '
';
return -1;
}
}
|