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
|
#!/usr/bin/perl
## ====================================================================
##
## Copyright (c) 1996-2000 Carnegie Mellon University. All rights
## reserved.
##
## Redistribution and use in source and binary forms, with or without
## modification, are permitted provided that the following conditions
## are met:
##
## 1. Redistributions of source code must retain the above copyright
## notice, this list of conditions and the following disclaimer.
##
## 2. Redistributions in binary form must reproduce the above copyright
## notice, this list of conditions and the following disclaimer in
## the documentation and/or other materials provided with the
## distribution.
##
## This work was supported in part by funding from the Defense Advanced
## Research Projects Agency and the National Science Foundation of the
## United States of America, and the CMU Sphinx Speech Consortium.
##
## THIS SOFTWARE IS PROVIDED BY CARNEGIE MELLON UNIVERSITY ``AS IS'' AND
## ANY EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
## THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
## PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL CARNEGIE MELLON UNIVERSITY
## NOR ITS EMPLOYEES BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
## SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
## LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
## DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
## THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
## (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
## OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
##
## ====================================================================
##
## Author: Ricky Houghton
## Author: David Huggins-Daines
##
use strict;
use File::Copy;
use File::Basename;
use File::Spec::Functions;
use File::Path;
use lib catdir(dirname($0), updir(), 'lib');
use SphinxTrain::Config;
use SphinxTrain::Util;
# LDA/MLLT doesn't really have sense with multistream
if ($ST::CFG_NUM_STREAMS != 1) {
Log("MODULE: 01 Train LDA transformation\n");
Log("Skipped for multistream setup, see CFG_NUM_STREAMS configuration\n");
Log("LDA/MLLT only has sense for single stream features");
Log("Skipping LDA training");
exit 0;
}
if ($ST::CFG_LDA_MLLT ne 'yes') {
Log("MODULE: 01 Train LDA transformation\n");
Log("Skipped (set \$CFG_LDA_MLLT = 'yes' to enable)\n");
exit 0;
}
my $iter = 1;
if (@ARGV) {
$iter = shift;
}
my $n_parts = ($ST::CFG_NPART) ? $ST::CFG_NPART : 1;
# If the number of parts is given as command line argument, overwrite
# the number coming from the config file
if (@ARGV) {
$n_parts = shift;
}
$| = 1; # Turn on autoflushing
my $logdir = "$ST::CFG_LOG_DIR/01.lda_train";
if ($iter == 1) {
Log("MODULE: 01 Train LDA transformation\n");
Log("Phase 1: Cleaning up directories:");
# Don't do this on a queue, because of NFS bugs
unless ($ST::CFG_QUEUE_TYPE eq 'Queue::PBS') {
LogProgress("\taccumulator...");
rmtree ($ST::CFG_BWACCUM_DIR, 0, 1);
mkdir ($ST::CFG_BWACCUM_DIR,0777);
}
LogProgress("logs...");
rmtree($logdir, 0, 1);
mkdir ($logdir,0777);
LogProgress("qmanager...\n");
rmtree ($ST::CFG_QMGR_DIR, 0, 1);
mkdir ($ST::CFG_QMGR_DIR,0777);
LogStatus('completed');
# For the first iteration Flat initialize models.
my $return_value = FlatInitialize();
exit ($return_value) if ($return_value);
Log("Phase 3: Forward-Backward");
}
if ($iter eq 'N') {
Log("Phase 4: LDA transform estimation");
my @deps;
for (my $i=1; $i<=$n_parts; $i++) {
push @deps, LaunchScript("bw.lda.$iter.$i", ['baum_welch.pl', $iter, $i, $n_parts, 'yes']);
}
push @deps, LaunchScript("lda", "lda_train.pl", \@deps);
# Explicitly wait for everything
WaitForScript(@deps);
}
else {
my @deps;
for (my $i=1; $i<=$n_parts; $i++) {
push @deps, LaunchScript("bw.lda.$iter.$i", ['baum_welch.pl', $iter, $i, $n_parts, 'no']);
}
push @deps, LaunchScript("norm.$iter", ['norm_and_launchbw.pl', $iter, $n_parts], \@deps);
# Explicitly wait for the BW scripts to avoid zombies
WaitForScript(@deps);
# Report success or failure
my $lda_log = File::Spec->catfile($logdir, "$ST::CFG_EXPTNAME.lda_train.log");
if ($iter == 1) {
# Look for an error
for ($iter = 1; $iter <= $ST::CFG_MAX_ITERATIONS; ++$iter) {
my $norm_log = File::Spec->catfile($logdir,
"$ST::CFG_EXPTNAME.$iter.norm.log");
if (open LOG, "<$norm_log") {
if (/failed/i or /Aborting/) {
LogError("Training failed in iteration $iter");
exit 1;
}
}
}
if (open LOG, "<$lda_log") {
while (<LOG>) {
if (/failed/i) {
LogError("LDA Training failed");
exit 1;
}
elsif (/complete/) {
Log("LDA Training completed", 'result');
exit 0;
}
}
}
}
}
sub CopyInitialize {
Log("Phase 2: Copy initialize from falign model\n");
my $hmmdir = catdir($ST::CFG_BASE_DIR, "model_parameters");
my $outhmm = catdir($hmmdir, "${ST::CFG_EXPTNAME}.ci_lda_flatinitial");
my $modarchdir = catdir($ST::CFG_BASE_DIR, "model_architecture");
mkdir ($outhmm,0777);
foreach (qw(means variances mixture_weights transition_matrices)) {
copy(catfile($ST::CFG_FORCE_ALIGN_MODELDIR, $_), catfile($outhmm, $_))
or return -1;
}
my $ci_mdeffile = catfile($modarchdir, "$ST::CFG_EXPTNAME.falign_ci.mdef");
my $out_mdeffile = catfile($modarchdir, "$ST::CFG_EXPTNAME.ci.mdef");
copy($ci_mdeffile, $out_mdeffile);
return 0;
}
sub FlatInitialize
{
Log("Phase 2: Flat initialize\n");
#**************************************************************************
# this script given an mdef file and a codebook (means/vars in S3 format)
# produces flat mixture weights in a semicontinuos setting. From the models
# produced by this script we can restart normal baum-welch training
# Flat init might not be the best choice, specially if we have access to
# segmentation files for the whole training database.
#**************************************************************************
my $logdir = "$ST::CFG_LOG_DIR/01.lda_train";
my $modarchdir = "$ST::CFG_BASE_DIR/model_architecture";
my $hmmdir = "$ST::CFG_BASE_DIR/model_parameters";
mkdir ($logdir,0777);
mkdir ($modarchdir,0777);
mkdir ($hmmdir,0777);
my $cmd ="";
#-------------------------------------------------------------------------
# Take the phone list. Put three hyphens after every phone. That is the
# required format. Then make a ci model definition file from it using the
# following program.
#-------------------------------------------------------------------------
#$rawphonefile obtained from variables.def
my $phonefile = "$modarchdir/$ST::CFG_EXPTNAME.phonelist";
my $ci_mdeffile = "$modarchdir/$ST::CFG_EXPTNAME.ci.mdef";
open PHONELIST, "<".$ST::CFG_RAWPHONEFILE;
open PHONEFILE, ">".$phonefile;
my $NUM_PHONES = 0;
my @phonelist;
while (defined(my $line = <PHONELIST>)) {
$line = Trim($line);
next if $line =~ m/^\s*$/;
$line =~ s/$/ - - - /;
push @phonelist, $line;
$NUM_PHONES++;
}
# Make sure the CI phones are sorted, as PocketSphinx requires them to be
foreach (sort @phonelist) {
print PHONEFILE $_, "\n";
}
# system "sed 's+\$+ - - - +g' $ST::CFG_RAWPHONEFILE > $phonefile";
my $logfile = "$logdir/${ST::CFG_EXPTNAME}.make_ci_mdef_fromphonelist.log";
#-------------------------------------------------------------------------
# Decide on what topology to use for the hmms: 3 state, 5 state, blah state
# or what, give it to the variable "statesperhmm" and use it to create
# the topology matrix in the topology file
#-------------------------------------------------------------------------
#$statesperhmm obtained from variables.def
my $topologyfile = "$modarchdir/$ST::CFG_EXPTNAME.topology";
RunScript("../prepare/maketopology.pl", $topologyfile);
my $return_value;
if ($return_value = RunTool('mk_mdef_gen', $logfile, 0,
-phnlstfn => $phonefile,
-ocimdef => $ci_mdeffile,
-n_state_pm => $ST::CFG_STATESPERHMM)) {
return $return_value;
}
#-------------------------------------------------------------------------
# make the flat models using the above topology file and the mdef file
#------------------------------------------------------------------------
my $outhmm = "$hmmdir/${ST::CFG_EXPTNAME}.ci_lda_flatinitial";
mkdir ($outhmm,0777);
my $FLAT = "$ST::CFG_BIN_DIR/mk_flat";
$logfile = "$logdir/${ST::CFG_EXPTNAME}.makeflat_cihmm.log";
if ($return_value = RunTool('mk_flat', $logfile, 0,
-moddeffn => $ci_mdeffile,
-topo => $topologyfile,
-mixwfn => catfile($outhmm, 'mixture_weights'),
-tmatfn => catfile($outhmm, 'transition_matrices'),
-nstream => 1, # Always one stream!
-ndensity => 1 # Always one density!
)) {
return $return_value;
}
#-------------------------------------------------------------------------
# Accumulate the means from the training data
#------------------------------------------------------------------------
$logfile = "$logdir/${ST::CFG_EXPTNAME}.initmean_cihmm.log";
open LOG,">$logfile";
my $output_buffer_dir = "$ST::CFG_BWACCUM_DIR/${ST::CFG_EXPTNAME}_buff_1";
mkdir ($output_buffer_dir,0777);
if ($return_value = RunTool('init_gau', $logfile, 0,
-ctlfn => $ST::CFG_LISTOFFILES,
-part => 1, -npart => 1,
-cepdir => $ST::CFG_FEATFILES_DIR,
-cepext => $ST::CFG_FEATFILE_EXTENSION,
-accumdir => $output_buffer_dir,
-agc => $ST::CFG_AGC,
-cmn => $ST::CFG_CMN,
-varnorm => $ST::CFG_VARNORM,
-feat => $ST::CFG_FEATURE,
-ceplen => $ST::CFG_VECTOR_LENGTH,
)) {
return $return_value;
}
#-------------------------------------------------------------------------
# Normalize the means
#------------------------------------------------------------------------
$logfile = "$logdir/${ST::CFG_EXPTNAME}.normmean_cihmm.log";
if ($return_value = RunTool('norm', $logfile, 0,
-accumdir => $output_buffer_dir,
-meanfn => catfile($outhmm, "globalmean"),
)) {
return $return_value;
}
#-------------------------------------------------------------------------
# Accumulate the variances from the training data
#------------------------------------------------------------------------
$logfile = "$logdir/${ST::CFG_EXPTNAME}.initvar_cihmm.log";
if ($return_value = RunTool('init_gau', $logfile, 0,
-meanfn => catfile($outhmm, "globalmean"),
-ctlfn => $ST::CFG_LISTOFFILES,
-part => 1, -npart => 1,
-cepdir => $ST::CFG_FEATFILES_DIR,
-cepext => $ST::CFG_FEATFILE_EXTENSION,
-accumdir => $output_buffer_dir,
-agc => $ST::CFG_AGC,
-cmn => $ST::CFG_CMN,
-varnorm => $ST::CFG_VARNORM,
-feat => $ST::CFG_FEATURE,
-ceplen => $ST::CFG_VECTOR_LENGTH,
-fullvar => $ST::CFG_FULLVAR,
)) {
return $return_value;
}
#-------------------------------------------------------------------------
# Normalize the variances
#------------------------------------------------------------------------
$logfile = "$logdir/${ST::CFG_EXPTNAME}.normvar_cihmm.log";
if ($return_value = RunTool('norm', $logfile, 0,
-accumdir => $output_buffer_dir,
-fullvar => $ST::CFG_FULLVAR,
-varfn => catfile($outhmm, "globalvar"),
)) {
return $return_value;
}
#-------------------------------------------------------------------------
# Create the copy operation file, simply a map between states
#------------------------------------------------------------------------
my $NUM_CI_STATES = $NUM_PHONES * $ST::CFG_STATESPERHMM;
if (open CPFILE, ">$ST::CFG_CP_OPERATION") {
for (my $CI_STATE = 0; $CI_STATE < $NUM_CI_STATES; $CI_STATE++) {
print CPFILE "$CI_STATE\t0\n";
}
close(CPFILE);
} else {
warn "Can't open $ST::CFG_CP_OPERATION\n";
}
#-------------------------------------------------------------------------
# Copy the means to all other states
#------------------------------------------------------------------------
$logfile = "$logdir/${ST::CFG_EXPTNAME}.cpmean_cihmm.log";
if ($return_value = RunTool('cp_parm', $logfile, 0,
-cpopsfn => $ST::CFG_CP_OPERATION,
-igaufn => catfile($outhmm, 'globalmean'),
-ncbout => $NUM_CI_STATES,
-ogaufn => catfile($outhmm, 'means')
)) {
return $return_value;
}
#-------------------------------------------------------------------------
# Copy the variances to all other states
#------------------------------------------------------------------------
$logfile = "$logdir/${ST::CFG_EXPTNAME}.cpvar_cihmm.log";
my @varcpy;
if ($ST::CFG_FULLVAR eq 'yes') {
@varcpy = (-ifullgaufn => catfile($outhmm, 'globalvar'),
-ofullgaufn => catfile($outhmm, 'variances'));
}
else {
@varcpy = (-igaufn => catfile($outhmm, 'globalvar'),
-ogaufn => catfile($outhmm, 'variances'));
}
if ($return_value = RunTool('cp_parm', $logfile, 0,
-cpopsfn => $ST::CFG_CP_OPERATION,
-ncbout => $NUM_CI_STATES,
@varcpy
)) {
return $return_value;
}
unlink $ST::CFG_CP_OPERATION;
return $return_value;
}
|