parseTcofSync_10sec.py 12.2 KB
Newer Older
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
#!/usr/bin/env python
# -*- coding: utf-8 -*-

from xml.etree import ElementTree as ET
from sys import argv
from num2words import num2words
from unidecode import unidecode
import re
import os.path
import sys

def transformation_text(text):
    bool=True
    if "###" in text or len(re.findall(r"\[.+\]", text)) > 0 or \
        len(re.findall(r"\p{L}+-[^\p{L}]|\p{L}+-$",text)) > 0 \
        or len(re.findall("[^\p{L}]-\p{L}+|^-\p{L}+", text)) > 0:
        #print text
        #print "Ligne Supprime"
        bool=False
    else:
        # 4x4
        # Remove noise sound (BIP) over Name of places and person
        #text = re.sub(r"¤[^ ]+|[^ ]+¤|¤", "", text.strip())
        if len(re.findall(r"\dx\d",text))>0:
            text=re.sub(r"x","  ",text)
        if len(re.findall("\d+h\d+",text))>0:
            heures=re.findall("\d+h\d+",text)
            for h in heures:
                split_h=h.split('h')
                text_rep=split_h[0]+' heure '+split_h[1]
                text=text.replace(h, text_rep)
        text=re.sub(r',',' ',text)
    # remove silence character : OK
        #text=re.sub(r"(/.+/","remplacer par la 1er",text)
    # Liaison non standard remarquable
        text=re.sub(r'=','',text)
    # Comment Transcriber
        text=re.sub(r'\{.+\}','',text)
        text=re.sub(r'\(.+\}','',text)
        #print "detecter (///|/|<|>)"
    # Remove undecidable variant heared like on (n') en:
        text=re.sub(r"\(.+\)","",text)
        #text = re.sub(r"(\+|[*]+|///|/|<|>)", "", text.strip())
        #text=re.sub(r"-|_|\."," ",text.strip())
        text=re.sub(r'(O.K.)','ok',text)
        text = re.sub(r'(O.K)', 'ok', text)
        # Replace . with ' '
        text=re.sub(r'\.',' ',text)
#text=re.sub(r"{[^{]+}"," ",text.strip())
        # Remove ? ! < > : OK
        #<[^\p{L}]|[^\p{L}]>|#+|<\p{L}+[ ]|<\p{L}+$
        text=re.sub(r":|\?|/|\!|<|>|#+","",text)
    # replace silence character with <sil> : OK
        #text=re.sub(r"(\+)", "<sil>", text)
        text=re.sub(r"(\+)", "!SIL", text)
        text=re.sub(r"(///)", "!SIL", text)
        #text=re.sub(r"(///)", "<long-sil>", text)
        if len(re.findall(r"/.+/", text)) > 0:
            #print "AVANT***********"+text
            for unchoosen_text in re.findall(r"/.+/", text):
                # choose first undecideble word
                unchoosen_word=unchoosen_text.split(',')
                for choosen_word in unchoosen_word:
                    # isn't incomprehensible word
                    if len(re.findall(r"\*+|\d+", choosen_word))==0:
                        choosen_word = choosen_word.replace('/', '')
                        text = text.replace(unchoosen_text, choosen_word)
            #print "Apres************"+text
    # Remove noise sound (BIP) over Name of places and person
        text=re.sub(r"(¤.+¤)",'<NOISE>',text)
    # replace unkown syllable
        text=re.sub(r"\*+","<SPOKEN_NOISE>",text)
    # cut of recording : OK
        text=re.sub(r"\$+","",text)
    # remove " character: OK
        text = re.sub(r"\"+", "", text)
        # t 'avais
        text = re.sub(r"[ ]\'", " ", text)
        text = re.sub(r"\'", "\' ", text)
    # convert number if exist : OK

        num_list = re.findall(" \d+| \d+$", text)
        if len(num_list) > 0:
            #print text
            #print "********************************* NUM2WORD"
            for num in num_list:
                num_in_word = num2words(int(num), lang='fr')
                #num_in_word=normalize('NFKD', num_in_word).encode('ascii', 'ignore')
                text = text.replace(str(num), " " + str(num_in_word) + " ")
            #print text
    # replace n succesive spaces with one space. : OK
        text=re.sub(r"\s{2,}"," ",text)
        text = re.sub("^ ", '', text)
    # change bounding | to < and > : OK
    #balise=set(re.findall(r"\|\w+_?\w+\|",text))
    balise=set(re.findall(r"\|.+\|",text))
    if len(balise)>0:
        #print(balise)
        for b in balise:
            new_balise='<'+b[1:len(b)-1]+'>'
            text=text.replace(b,new_balise)
        #print(text)
    # c'est l'essaim ....
    text=text.lower()
    return bool,text
if __name__=="__main__":
    # Inputs
    duration=5
    file_trs=argv[1]
    #print(file_trs)
    #print file_trs
    outdir=argv[2]
    basename=os.path.basename(file_trs.split('.')[0])
    # MetaData File
    file_meta = file_trs.split('.')[0] + '.xml'
    #print file_trs.split('.')[0]
    # Output File needed for kaldi input
    segments_file = open(outdir + '/segments', 'a')
    utt2spk_file = open(outdir + '/utt2spk', 'a')
    text_file = open(outdir + '/text', 'a')
    wav_scp = open(outdir + '/wav.scp', 'a')
    spk2gender= open(outdir + '/spk2gender', 'a')
    # Read Trans File
    tree_trs = ET.parse(file_trs)
    trsdoc= tree_trs.getroot()
    #Read MetaData Of speaker ( ID and Name)
    speaker_id=[]
    namespk=[]
    for spk in trsdoc.iter('Speaker'):
        id_spk=spk.get('id')
        name_spk=unidecode(spk.get('name'))
        #if isinstance(name_spk,str):
        #print(type(name_spk))
        #name_spk=normalize('NFKD', name_spk).encode('ascii', 'ignore')
        speaker_id.append(id_spk.replace(" ",""))
        namespk.append(name_spk.lower().replace(" ",""))
    #Read MetaData To get Gender of Speaker (Gender and Name)
    tree_meta = ET.parse(file_meta)
    metadoc= tree_meta.getroot()
    speaker_gender=[]
    #print namespk
    #print speaker_id
    for loc in metadoc.iter('locuteur'):
        if loc.attrib!=dict({}):
            name_loc=loc.get('identifiant')
            name_loc = unidecode(name_loc)
            name_loc=name_loc.replace(" ","")
            #print name_loc
            #print name_loc
            #If the gender of speaker doesn't mentioned
            #print loc.findall('sexe')
            if loc.findall('sexe')==[]:
                speaker_gender.append([speaker_id[namespk.index(name_loc.lower())],'m'])
            else:
                # case 1 represent gender informat
                gender_loc=loc.find('sexe').text
                if gender_loc==None:
                    speaker_gender.append([speaker_id[namespk.index(name_loc.lower())], 'm'])
                else:
                    #print namespk
                    #print namespk.index(name_loc.lower())
                    speaker_gender.append([speaker_id[namespk.index(name_loc.lower())],gender_loc.lower()])
    #Turnlist= trsdoc.getElementsByTagName('Turn')
    #Synclist= trsdoc.getElementsByTagName('Sync')
    #print len(Turnlist)
    #print len(Synclist)
    #list_time_sync=[]Q
    #for sync in Synclist:
    #    list_time_sync.append(float(sync.attributes['time'].value))
    #a=""
    #count=1
    #print "#id_utt\tid_Seg\tid_Spkr\tstartTime\tendTime\tText"
    text=""
    Turn_count=0
    count=0
    has_attrib_speaker=False
    # set for uniq add
    Spk_that_contribute_to_meeting=set([])
    start_utt=0
    end_utt=0
    sourceEncoding = "iso-8859-1"
    targetEncoding = "utf-8"
    seg_duration=0
    for Element in trsdoc.iter():
        if Element.tag=="Turn" and Element.get('speaker') is None:
            has_attrib_speaker=False
        elif Element.tag=="Turn":
            # If the latest Utterance of previous Speaker is the latest one of his Turn speech
            if Turn_count>0:
                count = 0
                #print text
                ### Save Files For Kaldi ###
                seg_id = str(basename) + '_spk-%03d_Turn-%03d_seg-%07d' % (int(spkr.split('spk')[1]), int(Turn_count), int(count))
                spkr_id=str(basename)+'_spk-%03d' % int(spkr.split('spk')[1])
                bool, text = transformation_text(text)
                # File wav.scp
                # File utt2spk
                # File text
                # File speaker_gender
                seg_duration=seg_duration+(float(endTime)-float(start_utt))
                if bool and text!="" and seg_duration>=duration:
                    Spk_that_contribute_to_meeting.add(spkr)
                    segments_file.write(seg_id+" "+basename+" "+str(start_utt)+" "+str(endTime)+"\n")
                    start_utt=endTime
                    seg_duration=0
                    utt2spk_file.write(seg_id+" "+spkr_id+"\n")
                    text_file.write(seg_id+" "+text+"\n")
                    text=""
                    #for spk_tuple in speaker_gender:
                    #    if spk_tuple[0]==spkr:
                    #        print >> spk2gender,'%s %s' % (seg_id, spk_tuple[1])
                    #        break
            has_attrib_speaker=True
            # Get id_spkr
            spkr = Element.get('speaker')
            #print file_trs
            spkr=spkr.split()[0]
            #print spkr
            # Get StartSegment
            startTime = Element.get('startTime')
            # Get EndSegment
            endTime = Element.get('endTime')
            # count sync for computing start and end utterance
            Turn_count = Turn_count+1
        elif Element.tag=="Sync" and has_attrib_speaker:
            Time_start_current_sync=Element.get('time')
            if count>0:
                #print text
                ### Save Files For Kaldi ###
                seg_id = str(basename) + '_spk-%03d_Turn-%03d_seg-%07d' % (int(spkr.split('spk')[1]), int(Turn_count) , int(count))
                spkr_id=str(basename)+'_spk-%03d' % int(spkr.split('spk')[1])
                bool, text = transformation_text(text)
                end_utt=Time_start_current_sync
                seg_duration=seg_duration+(float(end_utt)-float(start_utt))
                if bool and text!="" and seg_duration>=duration:
                    Spk_that_contribute_to_meeting.add(spkr)
                    segments_file.write(seg_id+" "+basename+" "+str(start_utt)+" "+str(end_utt)+"\n")
                    start_utt=Time_start_current_sync
                    utt2spk_file.write(seg_id+" "+spkr_id+"\n")
                    text_file.write(seg_id+" "+text+"\n")
                    text=""
                    seg_duration=0
            text=text+" "+Element.tail.replace('\n', '')
            count=count+1
        elif Element.tag=="Comment" and has_attrib_speaker and not Element.tail is None:
            text=text+" "+Element.tail.replace('\n', '')
        elif Element.tag=="Event" and has_attrib_speaker and not Element.tail is None :
            if Element.get('type')=='noise':
                if Element.get('desc')=='rire':
                    text=text+" |LAUGH| "+Element.tail.replace('\n', '')
                else:
                    text=text+" |NOISE| "+Element.tail.replace('\n', '')
            elif Element.get('type')=='pronounce':
                text=text+" |SPOKEN_NOISE| "+Element.tail.replace('\n', '')
            else:
                text=text+" |NOISE| "+Element.tail.replace('\n', '')
        elif Element.tag=="Who" and has_attrib_speaker and not Element.tail is None:
            text=text+" "+Element.tail.replace('\n', '')
        #else:
        #    print Element.attrib,Element.tag
        #    text=str(Element.tail)
        #    print "*********warning********"+text
            # Les phrases appartenant � un tour de parole
    # The last Turn, check if count >0 and add latest utterance
    #print count
    #print has_attrib_speaker
    #print Element.tail
    if count > 0 and has_attrib_speaker and not Element.tail is None and seg_duration>duration:
        #print text
        ### Save Files For Kaldi ###
        seg_id = str(basename) + '_spk-%03d_Turn-%03d_seg-%07d' % (
        int(spkr.split('spk')[1]), int(Turn_count), int(count))
        spkr_id = str(basename) + '_spk-%03d' % int(spkr.split('spk')[1])
        bool, text = transformation_text(text)
        if bool and text != "":
            Spk_that_contribute_to_meeting.add(spkr)
            segments_file.write(seg_id+" "+basename+" "+str(start_utt)+" "+str(endTime)+"\n")
            utt2spk_file.write(seg_id+" "+spkr_id+"\n")
            text_file.write(seg_id+" "+text+"\n")
    for spk in speaker_gender:
        if spk[0] in Spk_that_contribute_to_meeting:
            spk_id = str(basename)+'_spk-%03d' % int(spk[0].split('spk')[1])
            spk2gender.write(spk_id+" "+spk[1]+"\n")
    wav_scp.write(basename+" sox "+os.path.dirname(file_trs) + '/' + basename + '.wav'+" -t wav -r 16000 -c 1 - |\n")
    segments_file.close()
    utt2spk_file.close()
    text_file.close()
    wav_scp.close()