Blame view
main_tools/ConfidenceMeasure.sh
11.1 KB
e6be5137b
|
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 |
#!/bin/bash #----------------------------------------------------------------------------------------- # Author : Benjamin Lecouteux & Emmanuel FERREIRA (contact emmanuel.ferreira0194@gmail.com) # Brief: Determine les mesures de confiance d'une transcription (res de speeral) #----------------------------------------------------------------------------------------- # where is ConfidenceMeasure.sh if [ -z $MAIN_SCRIPT_PATH ]; then MAIN_SCRIPT_PATH=$(dirname $(readlink -e $0)); fi # where is ConfidenceMeasure.cfg CONFIDENCEMEASURE_CONFIG_FILE=$OTMEDIA_HOME"/cfg/ConfidenceMeasure.cfg" if [ -e $CONFIDENCEMEASURE_CONFIG_FILE ] then . $CONFIDENCEMEASURE_CONFIG_FILE else echo "ERROR : Can't find configuration file $CONFIDENCEMEASURE_CONFIG_FILE" >&2 exit 1 fi PACKAGE_CONF_MEASURE=$CONFIDENCEMEASURE_CONFIG_FILE #------------------ # Parser les options #------------------- while getopts ":c:s:h" OPTION do case $OPTION in h) #Display help echo -e "$0 :" echo -e "\tAuthor : Benjamin Lecouteux & Emmanuel FERREIRA (contact: emmanuel.ferreira0194@gmail.com)" echo -e "\tVersion : 2.0" echo -e "\tBrief : Determine confidence measure of a transcription" echo -e "\tUsage : $0 [OPTIONS] <(i) REP_IN> <REP_NAME>" echo -e "\tOptions:" echo -e "\t\tc) specify the path of the configuration file (default $PACKAGE_CONF_MEASURE)" echo -e "\t\ts) specify PORT@HOST of a SRILM server" exit 1 ;; c) #Change the configuration file PACKAGE_CONF_MEASURE=$OPTARG ;; s) #use an SRILM server (avoid loading arpa model in memory) SERVER=$OPTARG ;; :) echo "BAD USAGE : OPTION $OPTARG need a value" exit 1 ;; \?) echo "BAD USAGE : unknow option '$OPTARG'" exit 1 ;; esac done #------------------------------------------- # Shift options pour recuperation arguments #------------------------------------------- shift $((OPTIND-1)) if [ -z "$1" ] then echo "BAD USAGE: $0 [OPTIONS] <(i) repertoire (ex:20041006_0800_0900_CULTURE)> <REP_NAME (ex:res_p2)>" exit 1 fi if [ -z "$2" ] then echo "BAD USAGE: $0 [OPTIONS] <(i) repertoire (ex:20041006_0800_0900_CULTURE)> <REP_NAME (ex:res_p2)>" exit 1 fi . $PACKAGE_CONF_MEASURE #------------------------------------ # INIT - Creation du workspace #------------------------------------ NAME=`basename $1` CONF_DIR=$1/conf/$2 FICHIER_RES=$2 REF=$CONF_DIR/ref POS=$CONF_DIR/pos MLCLASS=$CONF_DIR/mlclass GVALIGN=$CONF_DIR/gvalign HTK_POST=$CONF_DIR/htk_post HTK_LM=$CONF_DIR/htk_lm WLAT=$CONF_DIR/wlat LIKELIHOOD=$CONF_DIR/likelihood GVCTM=$CONF_DIR/gvctm SEGCTM=$CONF_DIR/segctm SUPER_CTM=$CONF_DIR/super_ctm SCORED_CTM=$CONF_DIR/scored_ctm mkdir -p $CONF_DIR > /dev/null 2>&1 mkdir -p $REF > /dev/null 2>&1 mkdir -p $POS > /dev/null 2>&1 mkdir -p $MLCLASS > /dev/null 2>&1 mkdir -p $GVALIGN > /dev/null 2>&1 mkdir -p $HTK_POST > /dev/null 2>&1 #mkdir -p $HTK_LM ==> generer auto par SRILM si besoin mkdir -p $WLAT > /dev/null 2>&1 mkdir -p $LIKELIHOOD > /dev/null 2>&1 mkdir -p $GVCTM > /dev/null 2>&1 mkdir -p $SEGCTM > /dev/null 2>&1 mkdir -p $SUPER_CTM > /dev/null 2>&1 mkdir -p $SCORED_CTM > /dev/null 2>&1 |
216856376
|
105 |
if [ -z $BOOST_BIN ] && [ "$ARCH" == ".64" ] ; then |
e6be5137b
|
106 107 |
BOOST_BIN=$ROOT/bin/icsiboost-64bit-static-r160 fi |
fad685c5b
|
108 |
if [ -z $BOOST_BIN ] ; then |
fa8235ed8
|
109 |
BOOST_BIN=$ROOT/bin/icsiboost-32bits-static-r176 |
5a14b8678
|
110 |
fi |
e6be5137b
|
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 |
#----------------------------------------------------------------- # STEP 1 - Extension des treillis + ajout posteriors (format htk) #----------------------------------------------------------------- if [ $EXTEND == 1 ] then echo "EXTEND step..." rm -r $HTK_LM > /dev/null 2>&1 rm $HTK_POST/* > /dev/null 2>&1 # # --> Ajout des scores linguistiques dans le HTK # ls $1/$FICHIER_RES/*.treil > $CONF_DIR/Liste_treil_${NAME}.lst LM_ACCESS="-lm $ML" if [ ! -z $SERVER ]; then LM_ACCESS="-use-server $SERVER -cache-served-ngrams" fi echo "$SRILM_BIN/lattice-tool -read-htk -in-lattice-list $CONF_DIR/Liste_treil_${NAME}.lst $LM_ACCESS -order $ORDER -htk-logbase 10 -htk-lmscale $FUDGE -htk-wdpenalty $PENALITE -write-htk -out-lattice-dir $HTK_LM"; $SRILM_BIN/lattice-tool -read-htk -in-lattice-list $CONF_DIR/Liste_treil_${NAME}.lst $LM_ACCESS -order $ORDER -htk-logbase 10 -htk-lmscale $FUDGE -htk-wdpenalty $PENALITE -write-htk -out-lattice-dir $HTK_LM # # --> Calcul des posteriors a partir des scores acoustiques et linguistiques present dans le HTK # for file in `ls $HTK_LM/*.treil` do base=`basename $file .treil`; #echo "lattice-tool -read-htk -in-lattice $file -compute-posteriors -write-htk -out-lattice $HTK_POST/${base}.htk" $SRILM_BIN/lattice-tool -read-htk -in-lattice $file -compute-posteriors -write-htk -out-lattice $HTK_POST/${base}.htk done fi #--------------------------------------------------------------------------------------------------------------- # STEP 2 - alignement res et wlat pour creer res avec scores + infos (utilise un fastnc modifie) # Exemple : # ok amendement 0.814885 ( time=36 nodes=3 min=0.0016862 max=0.814885 mean=0.333896 var=0.363849 svar=0.603199 ) #---------------------------------------------------------------------------------------------------------------- if [ $FASTNC == 1 ] then echo "FASTNC step..." rm -f $POS/* $WLAT/* > /dev/null 2>&1 for file in `ls $HTK_LM/*.treil` do base=`basename $file .treil`; #echo "$ROOT/bin/fastnc_v1.4 $HTK_POST/${base}.htk $WLAT/${base}.wlat $1/$FICHIER_RES/${base}.res rien -dtw2 > $POS/$base.pos2&" $ROOT/bin/fastnc_v1.4 $HTK_POST/${base}.htk $WLAT/${base}.wlat $1/$FICHIER_RES/${base}.res rien -dtw2 > $POS/$base.pos2 done fi #------------------------------------------------------------------------------------------------------------ # STEP 3 - recuperation de la probabilite pour chaque mot + info relatives au modele de langue (backoff, ...) #------------------------------------------------------------------------------------------------------------ if [ $PPL == 1 ] then echo "PPL step..." rm -f $REF/* $CONF_DIR/${NAME}_ALLREF.* $MLCLASS/* > /dev/null 2>&1 # # --> Creation des references a partir des .res (uniquement si .treil present) # for file in `ls $1/$FICHIER_RES/*.res` do base=`basename $file .res`; if [ -f $1/$FICHIER_RES/$base.treil ];then cat $file | cut -f5 -d' ' | tr " " " " > $REF/${base}.ref fi done # # --> creation d'un fichier contenant l'ensemble des transcriptions du show # compteur=0 for file in `du -sh $REF/*.ref | grep -v "^0" | cut -f2` do base=`basename $file .ref`; cat $file >> $CONF_DIR/${NAME}_ALLREF.txt echo "" >> $CONF_DIR/${NAME}_ALLREF.txt ListeFichiers[$compteur]=$base.mlclass compteur=$(( $compteur + 1 )) done # # --> recuperation de la probabilite pour chaque mot provenant des resultats de l'ASR + informations linguistiques (backoff used, ngram,...) # $SRILM_BIN/ngram -lm $ML -order $ORDER -ppl $CONF_DIR/${NAME}_ALLREF.txt -debug 2 > $CONF_DIR/${NAME}_ALLREF.mlclass # # --> creation d'un fichier par fichier .ref # compteur=0 cat $CONF_DIR/${NAME}_ALLREF.mlclass | while read line do echo $line | grep "^$" > /dev/null if [ $? == 0 ];then compteur=$(( $compteur + 1 )) else echo "$line" | grep "p(" > /dev/null if [ $? == 0 ];then echo "$line" >> $MLCLASS/${ListeFichiers[${compteur}]}; fi fi done fi #---------------------------------------------------------- # STEP 4 - recuperation du score acoustique de chaque mot #---------------------------------------------------------- if [ $ACOUST == 1 ] then echo "ACOUST step..." rm -f $GVALIGN/* > /dev/null 2>&1 rm -f $GVCTM/* > /dev/null 2>&1 rm -f $SEGCTM/* > /dev/null 2>&1 rm -f $LIKELIHOOD/* > /dev/null 2>&1 for file in `ls $1/$FICHIER_RES/*.res` do base=`basename $file .res` if [ -f $1/$FICHIER_RES/$base.treil ];then #echo "$ROOT/script/MakeListForGVAlign.pl $file $GVALIGN"; $ROOT/script/MakeListForGVAlign.pl $file $GVALIGN fi done for file in `ls $GVALIGN/*.gvalign` do base=`basename $file .gvalign`; type=`echo $base | cut -f2 -d: | cut -f2- -d\# | sed -e "s/[0-9]\+//"` case "$type" in "M#S") HMM=$mod_ms ;; "F#S") HMM=$mod_fs ;; "M#T") HMM=$mod_mt ;; "F#T") HMM=$mod_ft ;; esac #echo "$ROOT/bin/gvalign.old $HMM $PHON $file -e $1/${REP_PLP}/ -f .plp -r $GVALIGN -g .gv -C FAST -W $GVCTM -O CTM -s $SEGCTM > $LIKELIHOOD/${base}.likelihood | sed -e 's/Decoding/\ Decoding/g' > $LIKELIHOOD/${base}.likelihood"; #$ROOT/bin/gvalign.old $HMM $PHON $file -e $1/${REP_PLP}/ -f .plp -r $GVALIGN -g .gv -C FAST -W $GVCTM -O CTM -s $SEGCTM | sed -e 's/Decoding/ Decoding/g' > $LIKELIHOOD/${base}.likelihood touch $LIKELIHOOD/${base}.likelihood done fi #-------------------------------------------------------------------------------------------------------------------------------- # STEP 5 - Merge de tous les scores caclules => res (ctm) avec scores/params utiliser dans la classif # Format : # mot NbNode MinNode MaxNode MeanNode VarNode SVarNode Posterior AcousticLogLikelihood AcousticLogLikelihood/Frame ... # AcousticConfidenceLikelihood AcousticConstraintLikeLihood AcousticNoConstraint Likelihood ClasseRepliLinguistique ... # RepliLinguistique LogLinguistique LogUnigramme NbMotsFenetre NbNulNode NbTrame #--------------------------------------------------------------------------------------------------------------------------------- if [ $EXTRACT == 1 ] then echo "EXTRACT step..." rm -f $SUPER_CTM/* > /dev/null 2>&1 for file in `ls $1/$FICHIER_RES/*.res` do base=`basename $file .res`; like=`echo "$base" | sed -e 's/\..*//'`; if [ -f $1/$FICHIER_RES/$base.treil ]; then echo "$ROOT/scripts/ExtractData.pl $pathML $nameML $POS/${base}.pos2 $file $LIKELIHOOD/${like}.likelihood $MLCLASS/${base}.mlclass $TYPE_ML > $SUPER_CTM/${base}.ctm"; $ROOT/script/ExtractData.pl $pathML $nameML $POS/${base}.pos2 $file $LIKELIHOOD/${like}.likelihood $MLCLASS/${base}.mlclass $TYPE_ML > $SUPER_CTM/${base}.ctm # $ROOT/script/ExtractData.pl $pathML $nameML $POS/${base}.pos2 $file $LIKELIHOOD/${like}.likelihood $MLCLASS/${base}.mlclass $TYPE_ML > $SUPER_CTM/${base}.ctm fi done fi #---------------------------------------------------------------- # STEP 6 - Calcul effectif du score de confiance pour chaque mot #---------------------------------------------------------------- if [ $BOOST == 1 ] then echo "BOOST step..." rm -f $SCORED_CTM/* $CONF_DIR/${NAME}.sctm $CONF_DIR/${NAME}.boost* $CONF_DIR/${NAME}.resboost* $CONF_DIR/${NAME}.corres* > /dev/null 2>&1 # utilise pour le test sans etiquette $ROOT/script/DissociateErroneousFromDecoded.pl $SUPER_CTM 2 equilibre > $CONF_DIR/${NAME}.sctm $ROOT/script/ConvertSuperCTMtoDataSVM.pl $CONF_DIR/${NAME}.sctm boost 2 0 0 > $CONF_DIR/${NAME}.boost $ROOT/script/ConvertSuperCTMtoDataSVM.pl $CONF_DIR/${NAME}.sctm boost 2 0 1 > $CONF_DIR/${NAME}.boost_refs $BOOST_BIN -S $ROOT/TRAIN -C --posteriors < $CONF_DIR/${NAME}.boost > $CONF_DIR/${NAME}.resboost cat $CONF_DIR/${NAME}.resboost | cut -f4 -d" " > $CONF_DIR/${NAME}.resboost2 cat $CONF_DIR/${NAME}.boost_refs | sed -e 's/.*ref=//' > $CONF_DIR/${NAME}.corres paste $CONF_DIR/${NAME}.corres $CONF_DIR/${NAME}.resboost2 | sed -e 's/\.ctm/\.res/' > $CONF_DIR/${NAME}.corres2 $ROOT/script/AssociateScoreToCtm.pl $CONF_DIR/${NAME}.corres2 $1/$FICHIER_RES/ $SCORED_CTM/ fi echo "END" |