File size: 69,389 Bytes
6fa4bc9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
{
    "paper_id": "P01-1002",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T09:29:32.860189Z"
    },
    "title": "Processing Broadcast Audio for Information Access",
    "authors": [
        {
            "first": "Jean-Luc",
            "middle": [],
            "last": "Gauvain",
            "suffix": "",
            "affiliation": {
                "laboratory": "Spoken Language Processing Group LIMSI-CNRS",
                "institution": "",
                "location": {
                    "addrLine": "B.P. 133",
                    "postCode": "91403",
                    "settlement": "Orsay cedex",
                    "country": "France"
                }
            },
            "email": "gauvain@limsi.fr"
        },
        {
            "first": "Lori",
            "middle": [],
            "last": "Lamel",
            "suffix": "",
            "affiliation": {
                "laboratory": "Spoken Language Processing Group LIMSI-CNRS",
                "institution": "",
                "location": {
                    "addrLine": "B.P. 133",
                    "postCode": "91403",
                    "settlement": "Orsay cedex",
                    "country": "France"
                }
            },
            "email": ""
        },
        {
            "first": "Gilles",
            "middle": [],
            "last": "Adda",
            "suffix": "",
            "affiliation": {
                "laboratory": "Spoken Language Processing Group LIMSI-CNRS",
                "institution": "",
                "location": {
                    "addrLine": "B.P. 133",
                    "postCode": "91403",
                    "settlement": "Orsay cedex",
                    "country": "France"
                }
            },
            "email": ""
        },
        {
            "first": "Martine",
            "middle": [],
            "last": "Adda-Decker",
            "suffix": "",
            "affiliation": {
                "laboratory": "Spoken Language Processing Group LIMSI-CNRS",
                "institution": "",
                "location": {
                    "addrLine": "B.P. 133",
                    "postCode": "91403",
                    "settlement": "Orsay cedex",
                    "country": "France"
                }
            },
            "email": ""
        },
        {
            "first": "Claude",
            "middle": [],
            "last": "Barras",
            "suffix": "",
            "affiliation": {
                "laboratory": "Spoken Language Processing Group LIMSI-CNRS",
                "institution": "",
                "location": {
                    "addrLine": "B.P. 133",
                    "postCode": "91403",
                    "settlement": "Orsay cedex",
                    "country": "France"
                }
            },
            "email": ""
        },
        {
            "first": "Langzhou",
            "middle": [],
            "last": "Chen",
            "suffix": "",
            "affiliation": {
                "laboratory": "Spoken Language Processing Group LIMSI-CNRS",
                "institution": "",
                "location": {
                    "addrLine": "B.P. 133",
                    "postCode": "91403",
                    "settlement": "Orsay cedex",
                    "country": "France"
                }
            },
            "email": ""
        },
        {
            "first": "Yannick",
            "middle": [],
            "last": "De Kercadio",
            "suffix": "",
            "affiliation": {
                "laboratory": "Spoken Language Processing Group LIMSI-CNRS",
                "institution": "",
                "location": {
                    "addrLine": "B.P. 133",
                    "postCode": "91403",
                    "settlement": "Orsay cedex",
                    "country": "France"
                }
            },
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "This paper addresses recent progress in speaker-independent, large vocabulary, continuous speech recognition, which has opened up a wide range of near and mid-term applications. One rapidly expanding application area is the processing of broadcast audio for information access. At LIMSI, broadcast news transcription systems have been developed for English, French, German, Mandarin and Portuguese, and systems for other languages are under development. Audio indexation must take into account the specificities of audio data, such as needing to deal with the continuous data stream and an imperfect word transcription. Some near-term applications areas are audio data mining, selective dissemination of information and media monitoring.",
    "pdf_parse": {
        "paper_id": "P01-1002",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "This paper addresses recent progress in speaker-independent, large vocabulary, continuous speech recognition, which has opened up a wide range of near and mid-term applications. One rapidly expanding application area is the processing of broadcast audio for information access. At LIMSI, broadcast news transcription systems have been developed for English, French, German, Mandarin and Portuguese, and systems for other languages are under development. Audio indexation must take into account the specificities of audio data, such as needing to deal with the continuous data stream and an imperfect word transcription. Some near-term applications areas are audio data mining, selective dissemination of information and media monitoring.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "A major advance in speech processing technology is the ability of todays systems to deal with nonhomogeneous data as is exemplified by broadcast data. With the rapid expansion of different media sources, there is a pressing need for automatic processing of such audio streams. Broadcast audio is challenging as it contains segments of various acoustic and linguistic natures, which require appropriate modeling. A special section in the Communications of the ACM devoted to \"News on Demand\" (Maybury, 2000) includes contributions from many of the sites carrying out active research in this area.",
                "cite_spans": [
                    {
                        "start": 491,
                        "end": 506,
                        "text": "(Maybury, 2000)",
                        "ref_id": "BIBREF11"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Via speech recognition, spoken document retrieval (SDR) can support random access to relevant portions of audio documents, reducing the time needed to identify recordings in large multimedia databases. The TREC (Text REtrieval Conference) SDR evaluation showed that only small differences in information retrieval performance are observed for automatic and manual transcriptions (Garofolo et al., 2000) .",
                "cite_spans": [
                    {
                        "start": 379,
                        "end": 402,
                        "text": "(Garofolo et al., 2000)",
                        "ref_id": "BIBREF4"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Large vocabulary continuous speech recognition (LVCSR) is a key technology that can be used to enable content-based information access in audio and video documents. Since most of the linguistic information is encoded in the audio channel of video data, which once transcribed can be accessed using text-based tools. This research has been carried out in a multilingual environment in the context of several recent and ongoing European projects. We highlight recent progress in LVCSR and describe some of our work in developing a system for processing broadcast audio for information access. The system has two main components, the speech transcription component and the information retrieval component. Versions of the LIMSI broadcast news transcription system have been developed in American English, French, German, Mandarin and Portuguese.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Substantial advances in speech recognition technology have been achieved during the last decade. Only a few years ago speech recognition was pri-marily associated with small vocabulary isolated word recognition and with speaker-dependent (often also domain-specific) dictation systems. The same core technology serves as the basis for a range of applications such as voice-interactive database access or limited-domain dictation, as well as more demanding tasks such as the transcription of broadcast data. With the exception of the inherent variability of telephone channels, for most applications it is reasonable to assume that the speech is produced in relatively stable environmental and in some cases is spoken with the purpose of being recognized by the machine.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Progress in LVCSR",
                "sec_num": "2"
            },
            {
                "text": "The ability of systems to deal with nonhomogeneous data as is found in broadcast audio (changing speakers, languages, backgrounds, topics) has been enabled by advances in a variety of areas including techniques for robust signal processing and normalization; improved training techniques which can take advantage of very large audio and textual corpora; algorithms for audio segmentation; unsupervised acoustic model adaptation; efficient decoding with long span language models; ability to use much larger vocabularies than in the past -64 k words or more is common to reduce errors due to out-of-vocabulary words.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Progress in LVCSR",
                "sec_num": "2"
            },
            {
                "text": "With the rapid expansion of different media sources for information dissemination including via the internet, there is a pressing need for automatic processing of the audio data stream. The vast majority of audio and video documents that are produced and broadcast do not have associated annotations for indexation and retrieval purposes, and since most of today's annotation methods require substantial manual intervention, and the cost is too large to treat the ever increasing volume of documents. Broadcast audio is challenging to process as it contains segments of various acoustic and linguistic natures, which require appropriate modeling. Transcribing such data requires significantly higher processing power than what is needed to transcribe read speech data in a controlled environment, such as for speaker adapted dictation. Although it is usually assumed that processing time is not a major issue since computer power has been increasing continuously, it is also known that the amount of data appearing on information channels is increasing at a close rate. Therefore processing time is an important factor in making a speech transcription system viable for audio data mining and other related applications. Transcription word error rates of about 20% have been reported for unrestricted broadcast news data in several languages.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Progress in LVCSR",
                "sec_num": "2"
            },
            {
                "text": "As shown in Figure 1 the LIMSI broadcast news transcription system for automatic indexation consists of an audio partitioner and a speech recognizer.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 12,
                        "end": 20,
                        "text": "Figure 1",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Progress in LVCSR",
                "sec_num": "2"
            },
            {
                "text": "The goal of audio partitioning is to divide the acoustic signal into homogeneous segments, labeling and structuring the acoustic content of the data, and identifying and removing non-speech segments. The LIMSI BN audio partitioner relies on an audio stream mixture model (Gauvain et al., 1998) . While it is possible to transcribe the continuous stream of audio data without any prior segmentation, partitioning offers several advantages over this straight-forward solution. First, in addition to the transcription of what was said, other interesting information can be extracted such as the division into speaker turns and the speaker identities, and background acoustic conditions. This information can be used both directly and indirectly for indexation and retrieval purposes. Second, by clustering segments from the same speaker, acoustic model adaptation can be carried out on a per cluster basis, as opposed to on a single segment basis, thus providing more adaptation data. Third, prior segmentation can avoid problems caused by linguistic discontinuity at speaker changes. Fourth, by using acoustic models trained on particular acoustic conditions (such as wide-band or telephone band), overall performance can be significantly improved. Finally, eliminating non-speech segments substantially reduces the computation time. The result of the partitioning process is a set of speech segments usually corresponding to speaker turns with speaker, gender and telephone/wide-band labels (see Figure 2 ).",
                "cite_spans": [
                    {
                        "start": 271,
                        "end": 293,
                        "text": "(Gauvain et al., 1998)",
                        "ref_id": "BIBREF6"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 1495,
                        "end": 1503,
                        "text": "Figure 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Audio partitioning",
                "sec_num": "3"
            },
            {
                "text": "For each speech segment, the word recognizer determines the sequence of words in the segment, associating start and end times and an optional confidence measure with each word. The LIMSI system, in common with most of today's state-ofthe-art systems, makes use of statistical models of speech generation. From this point of view, message generation is represented by a language model which provides an estimate of the probability of any given word string, and the encoding of the message in the acoustic signal is represented by a probability density function. The speakerindependent 65k word, continuous speech recognizer makes use of 4-gram statistics for language modeling and of continuous density hidden Markov models (HMMs) with Gaussian mixtures for acoustic modeling. Each word is represented by one or more sequences of context-dependent phone models as determined by its pronunciation. The acoustic and language models are trained on large, representative corpora for each task and language.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Transcription of Broadcast News",
                "sec_num": "4"
            },
            {
                "text": "Processing time is an important factor in making a speech transcription system viable for automatic indexation of radio and television broadcasts. For many applications there are limitations on the response time and the available computational resources, which in turn can significantly affect the design of the acoustic and language models. Word recognition is carried out in one or more decoding passes with more accurate acoustic and language models used in successive passes. A 4-gram single pass dynamic network decoder has been developed which can achieve faster than real-time decoding with a word error under 30%, running in less than 100 Mb of memory on widely available platforms such Pentium III or Alpha machines.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Transcription of Broadcast News",
                "sec_num": "4"
            },
            {
                "text": "A characteristic of the broadcast news domain is that, at least for what concerns major news events, similar topics are simultaneously covered in different emissions and in different countries and languages. Automatic processing carried out on contemporaneous data sources in different languages can serve for multi-lingual indexation and retrieval. Multilinguality is thus of particular interest for media watch applications, where news may first break in another country or language.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Multilinguality",
                "sec_num": "5"
            },
            {
                "text": "At LIMSI broadcast news transcription systems have been developed for the American English, French, German, Mandarin and Portuguese languages. The Mandarin language was chosen because it is quite different from the other languages (tone and syllable-based), and Mandarin resources are available via the LDC as well as reference performance results.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Multilinguality",
                "sec_num": "5"
            },
            {
                "text": "Our system and other state-of-the-art systems can transcribe unrestricted American English broadcast news data with word error rates under 20%. Our transcription systems for French and German have comparable error rates for news broadcasts (Adda-Decker et al., 2000) . The character error rate for Mandarin is also about 20% (Chen et al., 2000) . Based on our experience, it appears that with appropriately trained models, recognizer performance is more dependent upon the type and source of data, than on the language. For example, documentaries are particularly challenging to transcribe, as the audio quality is often not very high, and there is a large proportion of voice over.",
                "cite_spans": [
                    {
                        "start": 240,
                        "end": 266,
                        "text": "(Adda-Decker et al., 2000)",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 325,
                        "end": 344,
                        "text": "(Chen et al., 2000)",
                        "ref_id": "BIBREF3"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Multilinguality",
                "sec_num": "5"
            },
            {
                "text": "The automatically generated partition and word transcription can be used for indexation and information retrieval purposes. Techniques commonly applied to automatic text indexation can be applied to the automatic transcriptions of the broadcast news radio and TV documents. These techniques are based on document term frequencies, where the terms are obtained after standard text processing, such as text normalization, tokenization, stopping and stemming. Most of these preprocessing steps are the same as those used to prepare the texts for training the speech recognizer language models. While this offers advantages for speech recognition, it can lead to IR errors. For better IR results, some words sequences corresponding to acronymns, multiword namedentities (e.g. Los Angeles), and words preceded by some particular prefixes (anti, co, bi, counter) are rewritten as a single word. Stemming is used to reduce the number of lexical items for a given word sense. The stemming lexicon contains about 32000 entries and was constructed using Porter's algorithm (Porter80, 1980) on the most frequent words in the collection, and then manually corrected.",
                "cite_spans": [
                    {
                        "start": 833,
                        "end": 856,
                        "text": "(anti, co, bi, counter)",
                        "ref_id": null
                    },
                    {
                        "start": 1044,
                        "end": 1079,
                        "text": "Porter's algorithm (Porter80, 1980)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Spoken Document Retrieval",
                "sec_num": "6"
            },
            {
                "text": "The information retrieval system relies on a un- Figure 1 : Overview of an audio transcription system. The audio partitioner divides the data stream into homogeneous acoustic segments, removing non-speech portions. The word recognizer identifies the words in each speech segment, associating time-markers with each word. audiofile filename=19980411 1600 1630 CNN HDL language=english \u00a1 segment type=wideband gender=female spkr=1 stime=50.25 etime=86.83\u00a1 wtime stime=50.38 etime=50.77\u00a1 c.n.n. wtime stime=50.77 etime=51.10\u00a1 headline wtime stime=51.10 etime=51.44\u00a1 news wtime stime=51.44 etime=51.63\u00a1 i'm wtime stime=51.63 etime=51.92\u00a1 robert wtime stime=51.92 etime=52.46\u00a1 johnson it is a day of final farewells in alabama the first funerals for victims of this week's tornadoes are being held today along with causing massive property damage the twisters killed thirty three people in alabama five in georgia and one each in mississippi and north carolina the national weather service says the tornado that hit jefferson county in alabama had winds of more than two hundred sixty miles per hour authorities speculated was the most powerful tornado ever to hit the southeast twisters destroyed two churches to fire stations and a school parishioners were in one church when the tornado struck /segment\u00a1 segment type=wideband gender=female spkr=2 stime=88.37 etime=104.86\u00a1 at one point when the table came onto my back i thought yes this is it i'm ready ready protects protect the children because the children screaming the children were screaming they were screaming in prayer that were screaming god help us /segment\u00a1 segment type=wideband gender=female spkr=1 stime=104.86 etime=132.37\u00a1 vice president al gore toured the area yesterday he called it the worst tornado devastation he's ever seen we will have a complete look at the weather across the u. s. in our extended weather forecast in six minutes /segment\u00a1 . . . segment type=wideband gender=male spkr=19 stime=1635.60 etime=1645.71\u00a1 so if their computing systems don't tackle this problem well we have a potential business disruption and either erroneous deliveries or misdeliveries or whatever savvy businesses are preparing now so the january first two thousand would just be another day on the town not a day when fast food and everything else slows down rick lockridge c.n.n. /segment\u00a1 /audiofile\u00a1 Figure 2 : Example system output obtained by automatic processing of the audio stream of a CNN show broadcasted on April 11, 1998 at 4pm. The output includes the partitioning and transcription results. To improve readability, word time stamps are given only for the first 6 words. Non speech segments have been removed and the following information is provided for each speech segment: signal bandwidth (telephone or wideband), speaker gender, and speaker identity (within the show). igram model per story. The score of a story is obtained by summing the query term weights which are simply the log probabilities of the terms given the story model once interpolated with a general English model. This term weighting has been shown to perform as well as the popular TF\u00a2 IDF weighting scheme (Hiemstra and Wessel, 1998; Miller et al., 1998; Ng, 1999; Sp\u00e4rk Jones et al., 1998) .",
                "cite_spans": [
                    {
                        "start": 3151,
                        "end": 3178,
                        "text": "(Hiemstra and Wessel, 1998;",
                        "ref_id": "BIBREF9"
                    },
                    {
                        "start": 3179,
                        "end": 3199,
                        "text": "Miller et al., 1998;",
                        "ref_id": "BIBREF12"
                    },
                    {
                        "start": 3200,
                        "end": 3209,
                        "text": "Ng, 1999;",
                        "ref_id": "BIBREF13"
                    },
                    {
                        "start": 3210,
                        "end": 3235,
                        "text": "Sp\u00e4rk Jones et al., 1998)",
                        "ref_id": "BIBREF15"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 49,
                        "end": 57,
                        "text": "Figure 1",
                        "ref_id": null
                    },
                    {
                        "start": 2361,
                        "end": 2369,
                        "text": "Figure 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Spoken Document Retrieval",
                "sec_num": "6"
            },
            {
                "text": "Werr Base BRF Closed",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Transcriptions",
                "sec_num": null
            },
            {
                "text": "The text of the query may or may not include the index terms associated with relevant documents. One way to cope with this problem is to use query expansion (Blind Relevance Feedback, BRF (Walker and de Vere, 1990) ) based on terms present in retrieved contemporary texts.",
                "cite_spans": [
                    {
                        "start": 188,
                        "end": 214,
                        "text": "(Walker and de Vere, 1990)",
                        "ref_id": "BIBREF17"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Transcriptions",
                "sec_num": null
            },
            {
                "text": "The system was evaluated in the TREC SDR track, with known story boundaries. The SDR data collection contains 557 hours of broadcast news from the period of February through June 1998. This data includes 21750 stories and a set of 50 queries with the associated relevance judgments (Garofolo et al., 2000) .",
                "cite_spans": [
                    {
                        "start": 282,
                        "end": 305,
                        "text": "(Garofolo et al., 2000)",
                        "ref_id": "BIBREF4"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Transcriptions",
                "sec_num": null
            },
            {
                "text": "In order to assess the effect of the recognition time on the information retrieval results we transcribed the 557 hours of broadcast news data using two decoder configurations: a single pass 1.4xRT system and a three pass 10xRT system. The word error rates are measured on a 10h test subset (Garofolo et al., 2000) . The information retrieval results are given in terms of mean average precision (MAP), as is done for the TREC benchmarks in Table 1 with and without query expansion. For comparison, results are also given for manually produced closed captions. With query expansion comparable IR results are obtained using the closed captions and the 10xRT transcriptions, and a moderate degradation (4% absolute) is observed using the 1.4xRT transcriptions.",
                "cite_spans": [
                    {
                        "start": 291,
                        "end": 314,
                        "text": "(Garofolo et al., 2000)",
                        "ref_id": "BIBREF4"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 441,
                        "end": 448,
                        "text": "Table 1",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Transcriptions",
                "sec_num": null
            },
            {
                "text": "The broadcast news transcription system also provides non-lexical information along with the word transcription. This information is available in the partition of the audio track, which identifies speaker turns. It is interesting to see whether or not such information can be used to help locate story boundaries, since in the general case these are not known. Statistics were made on 100 hours of radio and television broadcast news with manual transcriptions including the speaker identities. Of the 2096 sections manually marked as reports (considered stories), 40% start without a manually annotated speaker change. This means that using only speaker change information for detecting document boundaries would miss 40% of the boundaries. With automatically detected speaker changes, the number of missed boundaries would certainly increase. At the same time, 11,160 of the 12,439 speaker turns occur in the middle of a document, resulting in a false alarm rate of almost 90%. A more detailed analysis shows that about 50% of the sections involve a single speaker, but that the distribution of the number of speaker turns per section falls off very gradually (see Figure 3) . False alarms are not as harmful as missed detections, since it may be possible to merge adjacent turns into a single document in subsequent processing. These results show that even perfect speaker turn boundaries cannot be used as the primary cue for locating document boundaries. They can, however, be used to refine the placement of a document boundary located near a speaker change.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 1167,
                        "end": 1176,
                        "text": "Figure 3)",
                        "ref_id": "FIGREF1"
                    }
                ],
                "eq_spans": [],
                "section": "Locating Story Boundaries",
                "sec_num": "7"
            },
            {
                "text": "We also investigated using simple statistics on the durations of the documents. A histogram of the 2096 sections is shown in Figure 4 . One third of the sections are shorter than 30 seconds. The histogram has a bimodal distribution with a sharp peak around 20 seconds, and a smaller, flat peak around 2 minutes. Very short documents are typical of headlines which are uttered by single speaker, whereas longer documents are more likely to contain data from multiple talkers. This distribution led us to consider using a multi-scale segmentation of the audio stream into documents. Similar statistics were measured on the larger corpus (Figure 4 bottom) .",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 125,
                        "end": 133,
                        "text": "Figure 4",
                        "ref_id": "FIGREF2"
                    },
                    {
                        "start": 635,
                        "end": 652,
                        "text": "(Figure 4 bottom)",
                        "ref_id": "FIGREF2"
                    }
                ],
                "eq_spans": [],
                "section": "Locating Story Boundaries",
                "sec_num": "7"
            },
            {
                "text": "As proposed in (Abberley et al., 1999; Johnson et al., 1999) , we segment the audio stream into overlapping documents of a fixed duration. As a result of optimization, we chose a 30 second window duration with a 15 second overlap.",
                "cite_spans": [
                    {
                        "start": 15,
                        "end": 38,
                        "text": "(Abberley et al., 1999;",
                        "ref_id": "BIBREF0"
                    },
                    {
                        "start": 39,
                        "end": 60,
                        "text": "Johnson et al., 1999)",
                        "ref_id": "BIBREF10"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Locating Story Boundaries",
                "sec_num": "7"
            },
            {
                "text": "Since there are many stories significantly shorter than 30s in broadcast shows (see Figure 4) we conjunctured that it may be of interest to use a double windowing system in order to better target short stories . The window size of the smaller window was selected to be 10 seconds. So for each query, we independently retrieved two sets of documents, one set for each window size. Then for each document set, document recombination is done by merging overlapping documents until no further merges are possible. The score of a combined document is set to maximum score of any one of the components. For each document derived from the 30s windows, we produce a time stamp located at the center point of the document. However, if any smaller documents are embedded in this document, we take the center of the best scoring document. This way we try to take advantage of both window sizes. The MAP using a single 30s window and the double windowing strategy are shown in Table 2 . For comparison, the IR results using the manual story segmentation and the speaker turns located by the audio partitioner are also given. All conditions use the same word hypotheses obtained with a speech recognizer which had no knowledge about the story boundaries. manual segmentation (NIST) 59.6% audio partitioner 33.3% single window (30s) 50.0% double window 52.3% Table 2 : Mean average precision with manual and automatically determined story boundaries. The document collection contains 557 hours of broadcast news from the period of February through June 1998. (21750 stories, 50 queries with the associated relevance judgments.)",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 84,
                        "end": 93,
                        "text": "Figure 4)",
                        "ref_id": "FIGREF2"
                    },
                    {
                        "start": 965,
                        "end": 972,
                        "text": "Table 2",
                        "ref_id": null
                    },
                    {
                        "start": 1345,
                        "end": 1352,
                        "text": "Table 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Locating Story Boundaries",
                "sec_num": "7"
            },
            {
                "text": "From these results we can clearly see the interest of using a search engine specifically designed to retrieve stories in the audio stream. Using an a priori acoustic segmentation, the mean average precision is significantly reduced compared to a \"perfect\" manual segmentation, whereas the window-based search engine results are much closer. Note that in the manual segmentation all non-story segments such as advertising have been removed. This reduces the risk of having out-oftopic hits and explains part of the difference between this condition and the other conditions.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Locating Story Boundaries",
                "sec_num": "7"
            },
            {
                "text": "The problem of locating story boundaries is being further pursued in the context of the ALERT project, where one of the goals is to identify \"documents\" given topic profiles. This project is investigating the combined use of audio and video segmentation to more accurately locate document boundaries in the continuous data stream.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Locating Story Boundaries",
                "sec_num": "7"
            },
            {
                "text": "The work presented in this paper has benefited from a variety of research projects both at the European and National levels. These collaborative efforts have enabled access to real-world data allowing us to develop algorithms and models wellsuited for near-term applications.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Recent Research Projects",
                "sec_num": "8"
            },
            {
                "text": "The European project LE-4 OLIVE: A Multilingual Indexing Tool for Broadcast Material Based on Speech Recognition (http://twentyone.tpd.tno.nl/ olive/) addressed methods to automate the disclosure of the information content of broadcast data thus allowing content-based indexation. Speech recognition was used to produce a time-linked transcript of the audio channel of a broadcast, which was then used to produce a concept index for retrieval. Broadcast news transcription systems for French and German were developed. The French data come from a variety of television news shows and radio stations. The German data consist of TV news and documentaries from ARTE. OLIVE also developed tools for users to query the database, as well as cross-lingual access based on off-line machine translation of the archived documents, and online query translation.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Recent Research Projects",
                "sec_num": "8"
            },
            {
                "text": "The European project IST ALERT: Alert system for selective dissemination (http://www.fb9ti.uni-duisburg.de/alert) aims to associate stateof-the-art speech recognition with audio and video segmentation and automatic topic indexing to develop an automatic media monitoring demonstrator and evaluate it in the context of real world applications. The targeted languages are French, German and Portuguese. Major mediamonitoring companies in Europe are participating in this project.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Recent Research Projects",
                "sec_num": "8"
            },
            {
                "text": "Two other related FP5 IST projects are: CORE-TEX: Improving Core Speech Recognition Technology and ECHO: European CHronicles Online. CORETEX (http://coretex.itc.it/), aims at improving core speech recognition technologies, which are central to most applications involving voice technology. In particular the project addresses the development of generic speech recognition technology and methods to rapidly port technology to new domains and languages with limited supervision, and to produce enriched symbolic speech transcriptions. The ECHO project (http://pc-erato2.iei.pi.cnr.it/echo) aims to develop an infrastructure for access to historical films belonging to large national audiovisual archives. The project will integrate state-of-theart language technologies for indexing, searching and retrieval, cross-language retrieval capabilities and automatic film summary creation.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Recent Research Projects",
                "sec_num": "8"
            },
            {
                "text": "This paper has described some of the ongoing research activites at LIMSI in automatic transcription and indexation of broadcast data. Much of this research, which is at the forefront of todays technology, is carried out with partners with real needs for advanced audio processing technologies.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusions",
                "sec_num": "9"
            },
            {
                "text": "Automatic speech recognition is a key technology for audio and video indexing. Most of the linguistic information is encoded in the audio channel of video data, which once transcribed can be accessed using text-based tools. This is in contrast to the image data for which no common description language is widely adpoted. A variety of near-term applications are possible such as audio data mining, selective dissemination of information (News-on-Demand), media monitoring, content-based audio and video retrieval.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusions",
                "sec_num": "9"
            },
            {
                "text": "It appears that with word error rates on the order of 20%, comparable IR results to those obtained on text data can be achieved. Even with higher word error rates obtained by running a faster transcription system or by transcribing compressed audio data (Barras et al., 2000; J.M. Van Thong et al., 2000) (such as that can be loaded over the Internet), the IR performance remains quite good.",
                "cite_spans": [
                    {
                        "start": 254,
                        "end": 275,
                        "text": "(Barras et al., 2000;",
                        "ref_id": "BIBREF7"
                    },
                    {
                        "start": 276,
                        "end": 304,
                        "text": "J.M. Van Thong et al., 2000)",
                        "ref_id": "BIBREF16"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusions",
                "sec_num": "9"
            }
        ],
        "back_matter": [
            {
                "text": "This work has been partially financed by the European Commission and the French Ministry of Defense. The authors thank Jean-Jacques Gangolf, Sylvia Hermier and Patrick Paroubek for their participation in the development of different aspects of the automatic indexation system described here.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Acknowledgments",
                "sec_num": null
            }
        ],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "The THISL SDR System at TREC-8",
                "authors": [
                    {
                        "first": "Dave",
                        "middle": [],
                        "last": "Abberley",
                        "suffix": ""
                    },
                    {
                        "first": "Steve",
                        "middle": [],
                        "last": "Renals",
                        "suffix": ""
                    },
                    {
                        "first": "Dan",
                        "middle": [
                            "Ellis"
                        ],
                        "last": "",
                        "suffix": ""
                    },
                    {
                        "first": "Tony",
                        "middle": [],
                        "last": "Robinson",
                        "suffix": ""
                    }
                ],
                "year": 1999,
                "venue": "Proc. of the 8th Text Retrieval Conference TREC-8",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Dave Abberley, Steve Renals, Dan Ellis and Tony Robinson, \"The THISL SDR System at TREC-8\", Proc. of the 8th Text Retrieval Conference TREC-8, Nov 1999.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "Investigating text normalization and pronunciation variants for German broadcast transcription",
                "authors": [
                    {
                        "first": "Martine",
                        "middle": [],
                        "last": "Adda-Decker",
                        "suffix": ""
                    },
                    {
                        "first": "Gilles",
                        "middle": [],
                        "last": "Adda",
                        "suffix": ""
                    },
                    {
                        "first": "Lori",
                        "middle": [],
                        "last": "Lamel",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "Proc. ICSLP'2000",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Martine Adda-Decker, Gilles Adda, Lori Lamel, \"In- vestigating text normalization and pronunciation variants for German broadcast transcription,\" Proc. ICSLP'2000, Beijing, China, October 2000.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Automatic Transcription of Compressed Broadcast Audio Proc",
                "authors": [
                    {
                        "first": "Claude",
                        "middle": [],
                        "last": "Barras",
                        "suffix": ""
                    },
                    {
                        "first": "Lori",
                        "middle": [],
                        "last": "Lamel",
                        "suffix": ""
                    },
                    {
                        "first": "Jean-Luc",
                        "middle": [],
                        "last": "Gauvain",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "ICASSP",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Claude Barras, Lori Lamel, Jean-Luc Gauvain, \"Auto- matic Transcription of Compressed Broadcast Au- dio Proc. ICASSP'2001, Salt Lake City, May 2001.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Broadcast News Transcription in Mandarin",
                "authors": [
                    {
                        "first": "Langzhou",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "Lori",
                        "middle": [],
                        "last": "Lamel",
                        "suffix": ""
                    },
                    {
                        "first": "Gilles",
                        "middle": [],
                        "last": "Adda",
                        "suffix": ""
                    },
                    {
                        "first": "Jean-Luc",
                        "middle": [],
                        "last": "Gauvain",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "Proc. ICSLP'2000",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Langzhou Chen, Lori Lamel, Gilles Adda and Jean- Luc Gauvain, \"Broadcast News Transcription in Mandarin,\" Proc. ICSLP'2000, Beijing, China, Oc- tober 2000.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Trec-8 Spoken Document Retrieval Track Overview and Results",
                "authors": [
                    {
                        "first": "John",
                        "middle": [
                            "S"
                        ],
                        "last": "Garofolo",
                        "suffix": ""
                    },
                    {
                        "first": "Cedric",
                        "middle": [
                            "G P"
                        ],
                        "last": "Auzanne",
                        "suffix": ""
                    },
                    {
                        "first": "Ellen",
                        "middle": [
                            "M"
                        ],
                        "last": "Voorhees ; John",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Garofolo",
                        "suffix": ""
                    }
                ],
                "year": 1999,
                "venue": "Proc. 8th Text Retrieval Conference TREC-8",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "John S. Garofolo, Cedric G.P. Auzanne, and Ellen M. Voorhees, \"The TREC Spoken Document Re- trieval Track: A Success Story,\" Proc. of the 6th RIAO Conference, Paris, April 2000. Also John S. Garofolo et al., \"1999 Trec-8 Spoken Docu- ment Retrieval Track Overview and Results,\" Proc. 8th Text Retrieval Conference TREC-8, Nov 1999. (http://trec.nist.gov).",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Fast Decoding for Indexation of Broadcast Data",
                "authors": [
                    {
                        "first": "Jean-Luc",
                        "middle": [],
                        "last": "Gauvain",
                        "suffix": ""
                    },
                    {
                        "first": "Lori",
                        "middle": [],
                        "last": "Lamel",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "Proc. ICSLP'2000",
                "volume": "3",
                "issue": "",
                "pages": "794--798",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jean-Luc Gauvain, Lori Lamel, \"Fast Decoding for Indexation of Broadcast Data,\" Proc. ICSLP'2000, 3:794-798, Oct 2000.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Partitioning and Transcription of Broadcast News Data",
                "authors": [
                    {
                        "first": "Jean-Luc",
                        "middle": [],
                        "last": "Gauvain",
                        "suffix": ""
                    },
                    {
                        "first": "Lori",
                        "middle": [],
                        "last": "Lamel",
                        "suffix": ""
                    },
                    {
                        "first": "Gilles",
                        "middle": [],
                        "last": "Adda",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "1335--1338",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jean-Luc Gauvain, Lori Lamel, Gilles Adda, \"Parti- tioning and Transcription of Broadcast News Data,\" ICSLP'98, 5, pp. 1335-1338, Dec. 1998.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Yannick de Kercadio \"The LIMSI SDR system for TREC-9",
                "authors": [
                    {
                        "first": "Jean-Luc",
                        "middle": [],
                        "last": "Gauvain",
                        "suffix": ""
                    },
                    {
                        "first": "Lori",
                        "middle": [],
                        "last": "Lamel",
                        "suffix": ""
                    },
                    {
                        "first": "Claude",
                        "middle": [],
                        "last": "Barras",
                        "suffix": ""
                    },
                    {
                        "first": "Gilles",
                        "middle": [],
                        "last": "Adda",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "Proc. of the 9th Text Retrieval Conference TREC-9",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jean-Luc Gauvain, Lori Lamel, Claude Barras, Gilles Adda, Yannick de Kercadio \"The LIMSI SDR sys- tem for TREC-9,\" Proc. of the 9th Text Retrieval Conference TREC-9, Nov 2000.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Informedia: News-on-Demand Multimedia Information Acquisition and Retrieval",
                "authors": [
                    {
                        "first": "Alexander",
                        "middle": [
                            "G"
                        ],
                        "last": "Hauptmann",
                        "suffix": ""
                    },
                    {
                        "first": "Michael",
                        "middle": [
                            "J"
                        ],
                        "last": "Witbrock",
                        "suffix": ""
                    }
                ],
                "year": 1997,
                "venue": "Proc Intelligent Multimedia Information Retrieval, M. Maybury",
                "volume": "",
                "issue": "",
                "pages": "213--239",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Alexander G. Hauptmann and Michael J. Witbrock, \"Informedia: News-on-Demand Multimedia Infor- mation Acquisition and Retrieval,\" Proc Intelli- gent Multimedia Information Retrieval, M. May- bury, ed., AAAI Press, pp. 213-239, 1997.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "Twenty-One at TREC-7: Ad-hoc and Cross-language track",
                "authors": [
                    {
                        "first": "Djoerd",
                        "middle": [],
                        "last": "Hiemstra",
                        "suffix": ""
                    },
                    {
                        "first": "Wessel",
                        "middle": [],
                        "last": "Kraaij",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "Proc. of the 8th Text Retrieval Conference TREC-7",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Djoerd Hiemstra, Wessel Kraaij, \"Twenty-One at TREC-7: Ad-hoc and Cross-language track,\" Proc. of the 8th Text Retrieval Conference TREC-7, Nov 1998.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Spoken Document Retrieval for TREC-8 at Cambridge University",
                "authors": [
                    {
                        "first": "Sue",
                        "middle": [
                            "E"
                        ],
                        "last": "Johnson",
                        "suffix": ""
                    },
                    {
                        "first": "Pierre",
                        "middle": [],
                        "last": "Jourlin",
                        "suffix": ""
                    },
                    {
                        "first": "Karen",
                        "middle": [
                            "Sp\u00e4rck"
                        ],
                        "last": "Jones",
                        "suffix": ""
                    },
                    {
                        "first": "Phil",
                        "middle": [
                            "C"
                        ],
                        "last": "Woodland",
                        "suffix": ""
                    }
                ],
                "year": 1999,
                "venue": "Proc. of the 8th Text Retrieval Conference TREC-8",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Sue E. Johnson, Pierre Jourlin, Karen Sp\u00e4rck Jones, Phil C. Woodland, \"Spoken Document Retrieval for TREC-8 at Cambridge University\", Proc. of the 8th Text Retrieval Conference TREC-8, Nov 1999.",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Special Section on",
                "authors": [
                    {
                        "first": "",
                        "middle": [],
                        "last": "Mark Maybury",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "Communications of the ACM",
                "volume": "43",
                "issue": "2",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Mark Maybury, ed., Special Section on \"News on De- mand\", Communications of the ACM, 43(2), Feb 2000.",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "Using Hidden Markov Models for Information Retrieval",
                "authors": [
                    {
                        "first": "David",
                        "middle": [],
                        "last": "Miller",
                        "suffix": ""
                    },
                    {
                        "first": "Tim",
                        "middle": [],
                        "last": "Leek",
                        "suffix": ""
                    },
                    {
                        "first": "Richard",
                        "middle": [],
                        "last": "Schwartz",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "Proc. of the 8th Text Retrieval Conference TREC-7",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "David Miller, Tim Leek, Richard Schwartz, \"Using Hidden Markov Models for Information Retrieval\", Proc. of the 8th Text Retrieval Conference TREC-7, Nov 1998.",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "A Maximum Likelihood Ratio Information Retrieval Model",
                "authors": [
                    {
                        "first": "Kenney",
                        "middle": [],
                        "last": "Ng",
                        "suffix": ""
                    }
                ],
                "year": 1999,
                "venue": "Proc. of the 8th Text Retrieval Conference TREC-8",
                "volume": "",
                "issue": "",
                "pages": "413--435",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Kenney Ng, \"A Maximum Likelihood Ratio Informa- tion Retrieval Model,\" Proc. of the 8th Text Re- trieval Conference TREC-8, 413-435, Nov 1999.",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "An algorithm for suffix stripping",
                "authors": [
                    {
                        "first": "M",
                        "middle": [
                            "F"
                        ],
                        "last": "Porter",
                        "suffix": ""
                    }
                ],
                "year": 1980,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "130--137",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "M. F. Porter, \"An algorithm for suffix stripping\", Pro- gram, 14, pp. 130-137, 1980.",
                "links": null
            },
            "BIBREF15": {
                "ref_id": "b15",
                "title": "A probabilistic model of information retrieval: development and status",
                "authors": [
                    {
                        "first": "Karen",
                        "middle": [],
                        "last": "Sp\u00e4rk Jones",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Walker",
                        "suffix": ""
                    },
                    {
                        "first": "Stephen",
                        "middle": [
                            "E"
                        ],
                        "last": "Robertson",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Karen Sp\u00e4rk Jones, S. Walker, Stephen E. Robert- son, \"A probabilistic model of information retrieval: development and status,\" Technical Report of the Computer Laboratory, University of Cambridge, U.K., 1998.",
                "links": null
            },
            "BIBREF16": {
                "ref_id": "b16",
                "title": "SpeechBot: a Speech Recognition based Audio Indexing System for the Web",
                "authors": [
                    {
                        "first": "J",
                        "middle": [
                            "M"
                        ],
                        "last": "Van Thong",
                        "suffix": ""
                    },
                    {
                        "first": "David",
                        "middle": [],
                        "last": "Goddeau",
                        "suffix": ""
                    },
                    {
                        "first": "Anna",
                        "middle": [],
                        "last": "Litvinova",
                        "suffix": ""
                    },
                    {
                        "first": "Beth",
                        "middle": [],
                        "last": "Logan",
                        "suffix": ""
                    },
                    {
                        "first": "Pedro",
                        "middle": [],
                        "last": "Moreno",
                        "suffix": ""
                    },
                    {
                        "first": "Michael",
                        "middle": [],
                        "last": "Swain",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "Proc. of the 6th RIAO Conference",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "J.M. Van Thong, David Goddeau, Anna Litvi- nova, Beth Logan, Pedro Moreno, Michael Swain, \"SpeechBot: a Speech Recognition based Audio In- dexing System for the Web\", Proc. of the 6th RIAO Conference, Paris, April 2000.",
                "links": null
            },
            "BIBREF17": {
                "ref_id": "b17",
                "title": "Improving subject retrieval in online catalogues: 2. Relevance feedback and query expansion",
                "authors": [
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Walker",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "De Vere",
                        "suffix": ""
                    }
                ],
                "year": 1990,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "S. Walker, R. de Vere, \"Improving subject retrieval in online catalogues: 2. Relevance feedback and query expansion\", British Library Research Paper 72, British Library, London, U.K., 1990.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF1": {
                "text": "Histogram of the number of speaker turns per section in 100 hours of audio data from radio and TV sources(NPR, ABC, CNN, CSPAN)from May-June 1996.",
                "num": null,
                "uris": null,
                "type_str": "figure"
            },
            "FIGREF2": {
                "text": "Distribution of document durations for 100 hours of data from May-June 1996 (top) and for 557 hours from February-June 1998 (bottom).",
                "num": null,
                "uris": null,
                "type_str": "figure"
            }
        }
    }
}