Blame view
Scripts/utils/write_kwslist.pl
9.89 KB
ec85f8892 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 |
#!/usr/bin/perl # Copyright 2012 Johns Hopkins University (Author: Guoguo Chen) # Apache 2.0. # use strict; use warnings; use Getopt::Long; my $Usage = <<EOU; This script reads the raw keyword search results [result.*] and writes them as the kwslist.xml file. It can also do things like score normalization, decision making, duplicates removal, etc. Usage: utils/write_kwslist.pl [options] <raw_result_in|-> <kwslist_out|-> e.g.: utils/write_kwslist.pl --flen=0.01 --duration=1000 --segments=data/eval/segments --normalize=true --map-utter=data/kws/utter_map raw_results kwslist.xml Allowed options: --beta : Beta value when computing ATWV (float, default = 999.9) --digits : How many digits should the score use (int, default = "infinite") --duptime : Tolerance for duplicates (float, default = 0.5) --duration : Duration of all audio, you must set this (float, default = 999.9) --ecf-filename : ECF file name (string, default = "") --flen : Frame length (float, default = 0.01) --index-size : Size of index (float, default = 0) --kwlist-filename : Kwlist.xml file name (string, default = "") --language : Language type (string, default = "cantonese") --map-utter : Map utterance for evaluation (string, default = "") --normalize : Normalize scores or not (boolean, default = false) --Ntrue-scale : Keyword independent scale factor for Ntrue (float, default = 1.0) --remove-dup : Remove duplicates (boolean, default = false) --remove-NO : Remove the "NO" decision instances (boolean, default = false) --segments : Segments file from Kaldi (string, default = "") --system-id : System ID (string, default = "") --verbose : Verbose level (higher --> more kws section) (integer, default 0) --YES-cutoff : Only keep "\$YES-cutoff" yeses for each kw (int, default = -1) EOU my $segment = ""; my $flen = 0.01; my $beta = 999.9; my $duration = 999.9; my $language = "cantonese"; my $ecf_filename = ""; my $index_size = 0; my $system_id = ""; my $normalize = "false"; my $map_utter = ""; my $Ntrue_scale = 1.0; my $digits = 0; my $kwlist_filename = ""; my $verbose = 0; my $duptime = 0.5; my $remove_dup = "false"; my $remove_NO = "false"; my $YES_cutoff = -1; GetOptions('segments=s' => \$segment, 'flen=f' => \$flen, 'beta=f' => \$beta, 'duration=f' => \$duration, 'language=s' => \$language, 'ecf-filename=s' => \$ecf_filename, 'index-size=f' => \$index_size, 'system-id=s' => \$system_id, 'normalize=s' => \$normalize, 'map-utter=s' => \$map_utter, 'Ntrue-scale=f' => \$Ntrue_scale, 'digits=i' => \$digits, 'kwlist-filename=s' => \$kwlist_filename, 'verbose=i' => \$verbose, 'duptime=f' => \$duptime, 'remove-dup=s' => \$remove_dup, 'YES-cutoff=i' => \$YES_cutoff, 'remove-NO=s' => \$remove_NO); ($normalize eq "true" || $normalize eq "false") || die "$0: Bad value for option --normalize "; ($remove_dup eq "true" || $remove_dup eq "false") || die "$0: Bad value for option --remove-dup "; ($remove_NO eq "true" || $remove_NO eq "false") || die "$0: Bad value for option --remove-NO "; if ($segment) { open(SEG, "<$segment") || die "$0: Fail to open segment file $segment "; } if ($map_utter) { open(UTT, "<$map_utter") || die "$0: Fail to open utterance table $map_utter "; } if (@ARGV != 2) { die $Usage; } # Get parameters my $filein = shift @ARGV; my $fileout = shift @ARGV; # Get input source my $source = ""; if ($filein eq "-") { $source = "STDIN"; } else { open(I, "<$filein") || die "$0: Fail to open input file $filein "; $source = "I"; } # Get symbol table and start time my %tbeg; if ($segment) { while (<SEG>) { chomp; my @col = split(" ", $_); @col == 4 || die "$0: Bad number of columns in $segment \"$_\" "; $tbeg{$col[0]} = $col[2]; } } # Get utterance mapper my %utter_mapper; if ($map_utter) { while (<UTT>) { chomp; my @col = split(" ", $_); @col == 2 || die "$0: Bad number of columns in $map_utter \"$_\" "; $utter_mapper{$col[0]} = $col[1]; } } # Function for printing Kwslist.xml sub PrintKwslist { my ($info, $KWS) = @_; my $kwslist = ""; # Start printing $kwslist .= "<kwslist kwlist_filename=\"$info->[0]\" language=\"$info->[1]\" system_id=\"$info->[2]\"> "; my $prev_kw = ""; foreach my $kwentry (@{$KWS}) { if ($prev_kw ne $kwentry->[0]) { if ($prev_kw ne "") {$kwslist .= " </detected_kwlist> ";} $kwslist .= " <detected_kwlist search_time=\"1\" kwid=\"$kwentry->[0]\" oov_count=\"0\"> "; $prev_kw = $kwentry->[0]; } $kwslist .= " <kw file=\"$kwentry->[1]\" channel=\"$kwentry->[2]\" tbeg=\"$kwentry->[3]\" dur=\"$kwentry->[4]\" score=\"$kwentry->[5]\" decision=\"$kwentry->[6]\""; if (defined($kwentry->[7])) {$kwslist .= " threshold=\"$kwentry->[7]\"";} if (defined($kwentry->[8])) {$kwslist .= " raw_score=\"$kwentry->[8]\"";} $kwslist .= "/> "; } $kwslist .= " </detected_kwlist> "; $kwslist .= "</kwslist> "; return $kwslist; } # Function for sorting sub KwslistOutputSort { if ($a->[0] ne $b->[0]) { if ($a->[0] =~ m/[0-9]+$/ && $b->[0] =~ m/[0-9]+$/) { ($a->[0] =~ /([0-9]*)$/)[0] <=> ($b->[0] =~ /([0-9]*)$/)[0] } else { $a->[0] cmp $b->[0]; } } elsif ($a->[5] ne $b->[5]) { $b->[5] <=> $a->[5]; } else { $a->[1] cmp $b->[1]; } } sub KwslistDupSort { my ($a, $b, $duptime) = @_; if ($a->[0] ne $b->[0]) { $a->[0] cmp $b->[0]; } elsif ($a->[1] ne $b->[1]) { $a->[1] cmp $b->[1]; } elsif ($a->[2] ne $b->[2]) { $a->[2] cmp $b->[2]; } elsif (abs($a->[3]-$b->[3]) >= $duptime){ $a->[3] <=> $b->[3]; } elsif ($a->[5] ne $b->[5]) { $b->[5] <=> $a->[5]; } else { $b->[4] <=> $a->[4]; } } # Processing my @KWS; while (<$source>) { chomp; my @col = split(" ", $_); @col == 5 || die "$0: Bad number of columns in raw results \"$_\" "; my $kwid = shift @col; my $utter = $col[0]; my $start = sprintf("%.2f", $col[1]*$flen); my $dur = sprintf("%.2f", $col[2]*$flen-$start); my $score = exp(-$col[3]); if ($segment) { $start = sprintf("%.2f", $start+$tbeg{$utter}); } if ($map_utter) { $utter = $utter_mapper{$utter}; } push(@KWS, [$kwid, $utter, 1, $start, $dur, $score, ""]); } my %Ntrue = (); foreach my $kwentry (@KWS) { if (!defined($Ntrue{$kwentry->[0]})) { $Ntrue{$kwentry->[0]} = 0.0; } $Ntrue{$kwentry->[0]} += $kwentry->[5]; } # Scale the Ntrue my %threshold; foreach my $key (keys %Ntrue) { $Ntrue{$key} *= $Ntrue_scale; $threshold{$key} = $Ntrue{$key}/($duration/$beta+($beta-1)/$beta*$Ntrue{$key}); } # Removing duplicates if ($remove_dup eq "true") { my @tmp = sort {KwslistDupSort($a, $b, $duptime)} @KWS; @KWS = (); push(@KWS, $tmp[0]); for (my $i = 1; $i < scalar(@tmp); $i ++) { my $prev = $KWS[-1]; my $curr = $tmp[$i]; if ((abs($prev->[3]-$curr->[3]) < $duptime ) && ($prev->[2] eq $curr->[2]) && ($prev->[1] eq $curr->[1]) && ($prev->[0] eq $curr->[0])) { next; } else { push(@KWS, $curr); } } } my $format_string = "%g"; if ($digits gt 0 ) { $format_string = "%." . $digits ."f"; } my @info = ($kwlist_filename, $language, $system_id); my %YES_count; foreach my $kwentry (@KWS) { my $threshold = $threshold{$kwentry->[0]}; if ($kwentry->[5] > $threshold) { $kwentry->[6] = "YES"; if (defined($YES_count{$kwentry->[0]})) { $YES_count{$kwentry->[0]} ++; } else { $YES_count{$kwentry->[0]} = 1; } } else { $kwentry->[6] = "NO"; if (!defined($YES_count{$kwentry->[0]})) { $YES_count{$kwentry->[0]} = 0; } } if ($verbose > 0) { push(@{$kwentry}, sprintf("%g", $threshold)); } if ($normalize eq "true") { if ($verbose > 0) { push(@{$kwentry}, $kwentry->[5]); } my $numerator = (1-$threshold)*$kwentry->[5]; my $denominator = (1-$threshold)*$kwentry->[5]+(1-$kwentry->[5])*$threshold; if ($denominator != 0) { $kwentry->[5] = sprintf($format_string, $numerator/$denominator); } else { $kwentry->[5] = sprintf($format_string, $kwentry->[5]); } } else { $kwentry->[5] = sprintf($format_string, $kwentry->[5]); } } # Output sorting my @tmp = sort KwslistOutputSort @KWS; # Process the YES-cutoff. Note that you don't need this for the normal cases where # hits and false alarms are balanced if ($YES_cutoff != -1) { my $count = 1; for (my $i = 1; $i < scalar(@tmp); $i ++) { if ($tmp[$i]->[0] ne $tmp[$i-1]->[0]) { $count = 1; next; } if ($YES_count{$tmp[$i]->[0]} > $YES_cutoff*2) { $tmp[$i]->[6] = "NO"; $tmp[$i]->[5] = 0; next; } if (($count == $YES_cutoff) && ($tmp[$i]->[6] eq "YES")) { $tmp[$i]->[6] = "NO"; $tmp[$i]->[5] = 0; next; } if ($tmp[$i]->[6] eq "YES") { $count ++; } } } # Process the remove-NO decision if ($remove_NO eq "true") { my @KWS = @tmp; @tmp = (); for (my $i = 0; $i < scalar(@KWS); $i ++) { if ($KWS[$i]->[6] eq "YES") { push(@tmp, $KWS[$i]); } } } # Printing my $kwslist = PrintKwslist(\@info, \@tmp); if ($segment) {close(SEG);} if ($map_utter) {close(UTT);} if ($filein ne "-") {close(I);} if ($fileout eq "-") { print $kwslist; } else { open(O, ">$fileout") || die "$0: Fail to open output file $fileout "; print O $kwslist; close(O); } |