esc-bencher commited on
Commit
3a31e45
1 Parent(s): 182e12e

Update esc-datasets.py

Browse files
Files changed (1) hide show
  1. esc-datasets.py +47 -46
esc-datasets.py CHANGED
@@ -32,8 +32,6 @@ from pathlib import Path
32
  from huggingface_hub import HfApi, HfFolder
33
  import datasets
34
 
35
- from .cv_release_stats import STATS as _COMMON_VOICE_STATS
36
-
37
 
38
  _DESCRIPTIONS = {
39
  "ami": """
@@ -391,7 +389,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
391
 
392
  transcriptions[audio_filename] = {
393
  "id": _id,
394
- "text": text,
395
  }
396
 
397
  features = ["id", "text"]
@@ -438,6 +436,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
438
  "local_extracted_archive_paths": local_extracted_archive_paths[subconfig],
439
  "archives": [dl_manager.iter_archive(path) for path in archive_paths[subconfig]],
440
  "meta_path": meta_path[subconfig],
 
441
  },
442
  ),
443
  datasets.SplitGenerator(
@@ -446,6 +445,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
446
  "local_extracted_archive_paths": local_extracted_archive_paths["dev"],
447
  "archives": [dl_manager.iter_archive(path) for path in archive_paths["dev"]],
448
  "meta_path": meta_path["dev"],
 
449
  },
450
  ),
451
  datasets.SplitGenerator(
@@ -454,11 +454,12 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
454
  "local_extracted_archive_paths": local_extracted_archive_paths["test"],
455
  "archives": [dl_manager.iter_archive(path) for path in archive_paths["test"]],
456
  "meta_path": meta_path["test"],
 
457
  },
458
  ),
459
  ]
460
 
461
- def _spgispeech_generate_examples(self, local_extracted_archive_paths, archives, meta_path):
462
  # define the expected metadata dict keys,
463
  # some files have metadata with erroneous entries that we have to filter out
464
  dict_keys = {"id": "wav_filename", "text": "transcript"}
@@ -482,6 +483,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
482
  # get the .wav filename by removing the directory path from the audio filename
483
  wav_filename = "/".join(audio_filename.split("/")[-2:])
484
  example = dict(metadata[wav_filename])
 
485
  example["audio"] = {"path": path, "bytes": audio_file.read()}
486
  example["dataset"] = "spgispeech"
487
  yield audio_filename, example
@@ -502,7 +504,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
502
  split: _VOXPOPULI_METADATA_PATH.format(split=split) for split in splits
503
  }
504
 
505
- dl_manager.download_config.num_proc = len(audio_urls["train"])
506
  meta_paths = dl_manager.download_and_extract(meta_urls)
507
  audio_paths = dl_manager.download(audio_urls)
508
 
@@ -519,6 +521,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
519
  "audio_archives": [dl_manager.iter_archive(archive) for archive in audio_paths["train"]],
520
  "local_extracted_archives_paths": local_extracted_audio_paths["train"],
521
  "meta_path": meta_paths["train"],
 
522
  }
523
  ),
524
  datasets.SplitGenerator(
@@ -527,6 +530,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
527
  "audio_archives": [dl_manager.iter_archive(archive) for archive in audio_paths["dev"]],
528
  "local_extracted_archives_paths": local_extracted_audio_paths["dev"],
529
  "meta_path": meta_paths["dev"],
 
530
  }
531
  ),
532
  datasets.SplitGenerator(
@@ -535,11 +539,12 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
535
  "audio_archives": [dl_manager.iter_archive(archive) for archive in audio_paths["test"]],
536
  "local_extracted_archives_paths": local_extracted_audio_paths["test"],
537
  "meta_path": meta_paths["test"],
 
538
  }
539
  ),
540
  ]
541
 
542
- def _voxpopuli_generate_examples(self, audio_archives, local_extracted_archives_paths, meta_path):
543
  assert len(audio_archives) == len(local_extracted_archives_paths)
544
 
545
  logging.info("Reading voxpopuli metadata.")
@@ -553,7 +558,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
553
 
554
  yield audio_id, {
555
  "id": audio_id,
556
- "text": metadata[audio_id]["normalized_text"].lower(),
557
  "audio": {"path": path, "bytes": audio_file.read()},
558
  "dataset": "voxpopuli",
559
  }
@@ -572,6 +577,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
572
  gen_kwargs={
573
  "local_extracted_archives": [local_extracted_archives.get(train_name) for train_name in train_splits],
574
  "archives": [dl_manager.iter_archive(archive_paths[train_name]) for train_name in train_splits],
 
575
  },
576
  )
577
  ]
@@ -581,6 +587,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
581
  gen_kwargs={
582
  "local_extracted_archives": [local_extracted_archives.get("dev.clean")],
583
  "archives": [dl_manager.iter_archive(archive_paths["dev.clean"])],
 
584
  },
585
  ),
586
  datasets.SplitGenerator(
@@ -588,6 +595,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
588
  gen_kwargs={
589
  "local_extracted_archives": [local_extracted_archives.get("dev.other")],
590
  "archives": [dl_manager.iter_archive(archive_paths["dev.other"])],
 
591
  },
592
  ),
593
  ]
@@ -597,6 +605,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
597
  gen_kwargs={
598
  "local_extracted_archives": [local_extracted_archives.get("test.clean")],
599
  "archives": [dl_manager.iter_archive(archive_paths["test.clean"])],
 
600
  },
601
  ),
602
  datasets.SplitGenerator(
@@ -604,12 +613,13 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
604
  gen_kwargs={
605
  "local_extracted_archives": [local_extracted_archives.get("test.other")],
606
  "archives": [dl_manager.iter_archive(archive_paths["test.other"])],
 
607
  },
608
  ),
609
  ]
610
  return train_split + dev_splits + test_splits
611
 
612
- def _librispeech_generate_examples(self, archives, local_extracted_archives):
613
  key = 0
614
  audio_data = {}
615
  transcripts = []
@@ -637,7 +647,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
637
  {
638
  "id": id_,
639
  "file": audio_file,
640
- "text": transcript,
641
  }
642
  )
643
  if audio_data and len(audio_data) == len(transcripts):
@@ -674,12 +684,11 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
674
  "Please set use_auth_token=True or use_auth_token='<TOKEN>' to download this dataset"
675
  )
676
 
677
- bundle_url_template = _COMMON_VOICE_STATS["bundleURLTemplate"]
678
- bundle_version = bundle_url_template.split("/")[0]
679
  dl_manager.download_config.ignore_url_params = True
680
 
681
  self._common_voice_log_download("en", bundle_version, hf_auth_token)
682
- archive_path = dl_manager.download(self._common_voice_get_bundle_url("en", bundle_url_template))
683
  local_extracted_archive = dl_manager.extract(archive_path) if not dl_manager.is_streaming else None
684
 
685
  path_to_data = "/".join([bundle_version, "en"])
@@ -693,15 +702,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
693
  "archive_iterator": dl_manager.iter_archive(archive_path),
694
  "metadata_filepath": "/".join([path_to_data, "train.tsv"]) if path_to_data else "train.tsv",
695
  "path_to_clips": path_to_clips,
696
- },
697
- ),
698
- datasets.SplitGenerator(
699
- name=datasets.Split.TEST,
700
- gen_kwargs={
701
- "local_extracted_archive": local_extracted_archive,
702
- "archive_iterator": dl_manager.iter_archive(archive_path),
703
- "metadata_filepath": "/".join([path_to_data, "test.tsv"]) if path_to_data else "test.tsv",
704
- "path_to_clips": path_to_clips,
705
  },
706
  ),
707
  datasets.SplitGenerator(
@@ -711,26 +712,17 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
711
  "archive_iterator": dl_manager.iter_archive(archive_path),
712
  "metadata_filepath": "/".join([path_to_data, "dev.tsv"]) if path_to_data else "dev.tsv",
713
  "path_to_clips": path_to_clips,
 
714
  },
715
  ),
716
  datasets.SplitGenerator(
717
- name="other",
718
- gen_kwargs={
719
- "local_extracted_archive": local_extracted_archive,
720
- "archive_iterator": dl_manager.iter_archive(archive_path),
721
- "metadata_filepath": "/".join([path_to_data, "other.tsv"]) if path_to_data else "other.tsv",
722
- "path_to_clips": path_to_clips,
723
- },
724
- ),
725
- datasets.SplitGenerator(
726
- name="invalidated",
727
  gen_kwargs={
728
  "local_extracted_archive": local_extracted_archive,
729
  "archive_iterator": dl_manager.iter_archive(archive_path),
730
- "metadata_filepath": "/".join([path_to_data, "invalidated.tsv"])
731
- if path_to_data
732
- else "invalidated.tsv",
733
  "path_to_clips": path_to_clips,
 
734
  },
735
  ),
736
  ]
@@ -741,6 +733,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
741
  archive_iterator,
742
  metadata_filepath,
743
  path_to_clips,
 
744
  ):
745
  """Yields examples."""
746
  data_fields = list(self._info().features.keys())
@@ -783,7 +776,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
783
  text = text[1:-1]
784
  # replace double quotation marks with single
785
  text = text.replace('""', '"')
786
- result["text"] = text
787
 
788
  yield path, result
789
 
@@ -850,7 +843,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
850
  key = "-".join([speaker, start, end, label])
851
  example = {
852
  "audio": {"path": audio_file, "array": samples, "sampling_rate": sampling_rate},
853
- "text": transcript,
854
  "id": key,
855
  "dataset": "tedlium",
856
  }
@@ -920,7 +913,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
920
  key = transcript["id"]
921
  yield key, {
922
  "audio": audio,
923
- "text": transcript["text"],
924
  "dataset": "tedlium",
925
  "id": transcript["id"],
926
  }
@@ -969,7 +962,8 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
969
  dl_manager.iter_archive(archive_path) for archive_path in audio_archives_paths["train"]
970
  ],
971
  "local_audio_archives_paths": local_audio_archives_paths.get("train"),
972
- "meta_paths": meta_paths["train"]
 
973
  },
974
  ),
975
  datasets.SplitGenerator(
@@ -979,7 +973,8 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
979
  dl_manager.iter_archive(archive_path) for archive_path in audio_archives_paths["dev"]
980
  ],
981
  "local_audio_archives_paths": local_audio_archives_paths.get("dev"),
982
- "meta_paths": meta_paths["dev"]
 
983
  },
984
  ),
985
  datasets.SplitGenerator(
@@ -989,12 +984,13 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
989
  dl_manager.iter_archive(archive_path) for archive_path in audio_archives_paths["test"]
990
  ],
991
  "local_audio_archives_paths": local_audio_archives_paths.get("test"),
992
- "meta_paths": meta_paths["test"]
 
993
  },
994
  ),
995
  ]
996
 
997
- def _gigaspeech_generate_examples(self, audio_archives_iterators, local_audio_archives_paths, meta_paths):
998
  assert len(audio_archives_iterators) == len(meta_paths)
999
  if local_audio_archives_paths:
1000
  assert len(audio_archives_iterators) == len(local_audio_archives_paths)
@@ -1031,7 +1027,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
1031
  if len(text) == 0:
1032
  continue
1033
 
1034
- audio_meta["text"] = text
1035
 
1036
  path = os.path.join(local_audio_archives_paths[i], audio_path_in_archive) if local_audio_archives_paths \
1037
  else audio_path_in_archive
@@ -1074,6 +1070,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
1074
  "local_extracted_archive_paths": local_extracted_archive_paths["train"],
1075
  "archives": [dl_manager.iter_archive(path) for path in archive_paths["train"]],
1076
  "metadata": metadata,
 
1077
  },
1078
  ),
1079
  datasets.SplitGenerator(
@@ -1082,6 +1079,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
1082
  "local_extracted_archive_paths": local_extracted_archive_paths["dev"],
1083
  "archives": [dl_manager.iter_archive(path) for path in archive_paths["dev"]],
1084
  "metadata": metadata,
 
1085
  },
1086
  ),
1087
  datasets.SplitGenerator(
@@ -1090,11 +1088,12 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
1090
  "local_extracted_archive_paths": local_extracted_archive_paths["test"],
1091
  "archives": [dl_manager.iter_archive(path) for path in archive_paths["test"]],
1092
  "metadata": metadata,
 
1093
  },
1094
  ),
1095
  ]
1096
 
1097
- def _earnings_generate_examples(self, local_extracted_archive_paths, archives, metadata):
1098
  for local_extracted_archive_path, archive in zip(local_extracted_archive_paths, archives):
1099
  # Here we iterate over all the files within the TAR archive:
1100
  for audio_filename, audio_file in archive:
@@ -1124,7 +1123,7 @@ class ESCDatasets(datasets.GeneratorBasedBuilder):
1124
 
1125
  yield audio_filename, {
1126
  "id": audio_filename,
1127
- "text": text,
1128
  "dataset": "earnings22",
1129
  "audio": {"path": path, "bytes": audio_file.read()}
1130
  }
@@ -1366,7 +1365,7 @@ _AMI_ANNOTATIONS_ARCHIVE_URL = _AMI_BASE_DATA_URL + "annotations/{split}/text"
1366
 
1367
  _SPGISPEECH_BASE_URL = "https://huggingface.co/datasets/kensho/spgispeech/resolve/main/data/"
1368
 
1369
- _SPGISPEECH_AUDIO_BASE_URL = _SPGISPEECH_BASE_URL + "/audio"
1370
 
1371
  _SPGISPEECH_SUBSET_TO_DIR = {
1372
  "s": ["s"],
@@ -1385,7 +1384,7 @@ _SPGISPEECH_AUDIO_ARCHIVES_NAMES = {
1385
  "test": [f"test_part_{i}.tar.gz" for i in range(0, 3)],
1386
  }
1387
 
1388
- _SPGISPEECH_META_BASE_URL = _SPGISPEECH_BASE_URL + "/meta"
1389
 
1390
  _SPGISPEECH_META_FILENAMES = {
1391
  "s": "train_small.csv",
@@ -1417,6 +1416,8 @@ _LIBRISPEECH_DL_URLS = {
1417
 
1418
  _COMMON_VOICE_API_URL = "https://commonvoice.mozilla.org/api/v1"
1419
 
 
 
1420
  _TEDLIUM_BASE_URL = "https://huggingface.co/datasets/LIUM/tedlium/resolve/main/TEDLIUM_release3/legacy/"
1421
 
1422
  _TEDLIUM_URLS = {
 
32
  from huggingface_hub import HfApi, HfFolder
33
  import datasets
34
 
 
 
35
 
36
  _DESCRIPTIONS = {
37
  "ami": """
 
389
 
390
  transcriptions[audio_filename] = {
391
  "id": _id,
392
+ "text": text if split != "eval" else "",
393
  }
394
 
395
  features = ["id", "text"]
 
436
  "local_extracted_archive_paths": local_extracted_archive_paths[subconfig],
437
  "archives": [dl_manager.iter_archive(path) for path in archive_paths[subconfig]],
438
  "meta_path": meta_path[subconfig],
439
+ "is_test": False,
440
  },
441
  ),
442
  datasets.SplitGenerator(
 
445
  "local_extracted_archive_paths": local_extracted_archive_paths["dev"],
446
  "archives": [dl_manager.iter_archive(path) for path in archive_paths["dev"]],
447
  "meta_path": meta_path["dev"],
448
+ "is_test": False,
449
  },
450
  ),
451
  datasets.SplitGenerator(
 
454
  "local_extracted_archive_paths": local_extracted_archive_paths["test"],
455
  "archives": [dl_manager.iter_archive(path) for path in archive_paths["test"]],
456
  "meta_path": meta_path["test"],
457
+ "is_test": True,
458
  },
459
  ),
460
  ]
461
 
462
+ def _spgispeech_generate_examples(self, local_extracted_archive_paths, archives, meta_path, is_test):
463
  # define the expected metadata dict keys,
464
  # some files have metadata with erroneous entries that we have to filter out
465
  dict_keys = {"id": "wav_filename", "text": "transcript"}
 
483
  # get the .wav filename by removing the directory path from the audio filename
484
  wav_filename = "/".join(audio_filename.split("/")[-2:])
485
  example = dict(metadata[wav_filename])
486
+ if is_test: example["text"] = ""
487
  example["audio"] = {"path": path, "bytes": audio_file.read()}
488
  example["dataset"] = "spgispeech"
489
  yield audio_filename, example
 
504
  split: _VOXPOPULI_METADATA_PATH.format(split=split) for split in splits
505
  }
506
 
507
+ dl_manager.download_config.num_proc = len(audio_urls["train"]) // 4
508
  meta_paths = dl_manager.download_and_extract(meta_urls)
509
  audio_paths = dl_manager.download(audio_urls)
510
 
 
521
  "audio_archives": [dl_manager.iter_archive(archive) for archive in audio_paths["train"]],
522
  "local_extracted_archives_paths": local_extracted_audio_paths["train"],
523
  "meta_path": meta_paths["train"],
524
+ "is_test": False,
525
  }
526
  ),
527
  datasets.SplitGenerator(
 
530
  "audio_archives": [dl_manager.iter_archive(archive) for archive in audio_paths["dev"]],
531
  "local_extracted_archives_paths": local_extracted_audio_paths["dev"],
532
  "meta_path": meta_paths["dev"],
533
+ "is_test": False,
534
  }
535
  ),
536
  datasets.SplitGenerator(
 
539
  "audio_archives": [dl_manager.iter_archive(archive) for archive in audio_paths["test"]],
540
  "local_extracted_archives_paths": local_extracted_audio_paths["test"],
541
  "meta_path": meta_paths["test"],
542
+ "is_test": True,
543
  }
544
  ),
545
  ]
546
 
547
+ def _voxpopuli_generate_examples(self, audio_archives, local_extracted_archives_paths, meta_path, is_test):
548
  assert len(audio_archives) == len(local_extracted_archives_paths)
549
 
550
  logging.info("Reading voxpopuli metadata.")
 
558
 
559
  yield audio_id, {
560
  "id": audio_id,
561
+ "text": metadata[audio_id]["normalized_text"].lower() if not is_test else "",
562
  "audio": {"path": path, "bytes": audio_file.read()},
563
  "dataset": "voxpopuli",
564
  }
 
577
  gen_kwargs={
578
  "local_extracted_archives": [local_extracted_archives.get(train_name) for train_name in train_splits],
579
  "archives": [dl_manager.iter_archive(archive_paths[train_name]) for train_name in train_splits],
580
+ "is_test": False,
581
  },
582
  )
583
  ]
 
587
  gen_kwargs={
588
  "local_extracted_archives": [local_extracted_archives.get("dev.clean")],
589
  "archives": [dl_manager.iter_archive(archive_paths["dev.clean"])],
590
+ "is_test": False,
591
  },
592
  ),
593
  datasets.SplitGenerator(
 
595
  gen_kwargs={
596
  "local_extracted_archives": [local_extracted_archives.get("dev.other")],
597
  "archives": [dl_manager.iter_archive(archive_paths["dev.other"])],
598
+ "is_test": False,
599
  },
600
  ),
601
  ]
 
605
  gen_kwargs={
606
  "local_extracted_archives": [local_extracted_archives.get("test.clean")],
607
  "archives": [dl_manager.iter_archive(archive_paths["test.clean"])],
608
+ "is_test": True,
609
  },
610
  ),
611
  datasets.SplitGenerator(
 
613
  gen_kwargs={
614
  "local_extracted_archives": [local_extracted_archives.get("test.other")],
615
  "archives": [dl_manager.iter_archive(archive_paths["test.other"])],
616
+ "is_test": True,
617
  },
618
  ),
619
  ]
620
  return train_split + dev_splits + test_splits
621
 
622
+ def _librispeech_generate_examples(self, archives, local_extracted_archives, is_test):
623
  key = 0
624
  audio_data = {}
625
  transcripts = []
 
647
  {
648
  "id": id_,
649
  "file": audio_file,
650
+ "text": transcript if not is_test else "",
651
  }
652
  )
653
  if audio_data and len(audio_data) == len(transcripts):
 
684
  "Please set use_auth_token=True or use_auth_token='<TOKEN>' to download this dataset"
685
  )
686
 
687
+ bundle_version = _COMMON_VOICE_BUNDLE_URL_TEMPLATE.split("/")[0]
 
688
  dl_manager.download_config.ignore_url_params = True
689
 
690
  self._common_voice_log_download("en", bundle_version, hf_auth_token)
691
+ archive_path = dl_manager.download(self._common_voice_get_bundle_url("en", _COMMON_VOICE_BUNDLE_URL_TEMPLATE))
692
  local_extracted_archive = dl_manager.extract(archive_path) if not dl_manager.is_streaming else None
693
 
694
  path_to_data = "/".join([bundle_version, "en"])
 
702
  "archive_iterator": dl_manager.iter_archive(archive_path),
703
  "metadata_filepath": "/".join([path_to_data, "train.tsv"]) if path_to_data else "train.tsv",
704
  "path_to_clips": path_to_clips,
705
+ "is_test": False,
 
 
 
 
 
 
 
 
706
  },
707
  ),
708
  datasets.SplitGenerator(
 
712
  "archive_iterator": dl_manager.iter_archive(archive_path),
713
  "metadata_filepath": "/".join([path_to_data, "dev.tsv"]) if path_to_data else "dev.tsv",
714
  "path_to_clips": path_to_clips,
715
+ "is_test": False,
716
  },
717
  ),
718
  datasets.SplitGenerator(
719
+ name=datasets.Split.TEST,
 
 
 
 
 
 
 
 
 
720
  gen_kwargs={
721
  "local_extracted_archive": local_extracted_archive,
722
  "archive_iterator": dl_manager.iter_archive(archive_path),
723
+ "metadata_filepath": "/".join([path_to_data, "test.tsv"]) if path_to_data else "test.tsv",
 
 
724
  "path_to_clips": path_to_clips,
725
+ "is_test": True,
726
  },
727
  ),
728
  ]
 
733
  archive_iterator,
734
  metadata_filepath,
735
  path_to_clips,
736
+ is_test,
737
  ):
738
  """Yields examples."""
739
  data_fields = list(self._info().features.keys())
 
776
  text = text[1:-1]
777
  # replace double quotation marks with single
778
  text = text.replace('""', '"')
779
+ result["text"] = text if not is_test else ""
780
 
781
  yield path, result
782
 
 
843
  key = "-".join([speaker, start, end, label])
844
  example = {
845
  "audio": {"path": audio_file, "array": samples, "sampling_rate": sampling_rate},
846
+ "text": transcript if split_path != "test" else "",
847
  "id": key,
848
  "dataset": "tedlium",
849
  }
 
913
  key = transcript["id"]
914
  yield key, {
915
  "audio": audio,
916
+ "text": transcript["text"] if split_path != "test" else "",
917
  "dataset": "tedlium",
918
  "id": transcript["id"],
919
  }
 
962
  dl_manager.iter_archive(archive_path) for archive_path in audio_archives_paths["train"]
963
  ],
964
  "local_audio_archives_paths": local_audio_archives_paths.get("train"),
965
+ "meta_paths": meta_paths["train"],
966
+ "is_test": False,
967
  },
968
  ),
969
  datasets.SplitGenerator(
 
973
  dl_manager.iter_archive(archive_path) for archive_path in audio_archives_paths["dev"]
974
  ],
975
  "local_audio_archives_paths": local_audio_archives_paths.get("dev"),
976
+ "meta_paths": meta_paths["dev"],
977
+ "is_test": False,
978
  },
979
  ),
980
  datasets.SplitGenerator(
 
984
  dl_manager.iter_archive(archive_path) for archive_path in audio_archives_paths["test"]
985
  ],
986
  "local_audio_archives_paths": local_audio_archives_paths.get("test"),
987
+ "meta_paths": meta_paths["test"],
988
+ "is_test": True,
989
  },
990
  ),
991
  ]
992
 
993
+ def _gigaspeech_generate_examples(self, audio_archives_iterators, local_audio_archives_paths, meta_paths, is_test):
994
  assert len(audio_archives_iterators) == len(meta_paths)
995
  if local_audio_archives_paths:
996
  assert len(audio_archives_iterators) == len(local_audio_archives_paths)
 
1027
  if len(text) == 0:
1028
  continue
1029
 
1030
+ audio_meta["text"] = text if not is_test else ""
1031
 
1032
  path = os.path.join(local_audio_archives_paths[i], audio_path_in_archive) if local_audio_archives_paths \
1033
  else audio_path_in_archive
 
1070
  "local_extracted_archive_paths": local_extracted_archive_paths["train"],
1071
  "archives": [dl_manager.iter_archive(path) for path in archive_paths["train"]],
1072
  "metadata": metadata,
1073
+ "is_test": False,
1074
  },
1075
  ),
1076
  datasets.SplitGenerator(
 
1079
  "local_extracted_archive_paths": local_extracted_archive_paths["dev"],
1080
  "archives": [dl_manager.iter_archive(path) for path in archive_paths["dev"]],
1081
  "metadata": metadata,
1082
+ "is_test": False,
1083
  },
1084
  ),
1085
  datasets.SplitGenerator(
 
1088
  "local_extracted_archive_paths": local_extracted_archive_paths["test"],
1089
  "archives": [dl_manager.iter_archive(path) for path in archive_paths["test"]],
1090
  "metadata": metadata,
1091
+ "is_test": True,
1092
  },
1093
  ),
1094
  ]
1095
 
1096
+ def _earnings_generate_examples(self, local_extracted_archive_paths, archives, metadata, is_test):
1097
  for local_extracted_archive_path, archive in zip(local_extracted_archive_paths, archives):
1098
  # Here we iterate over all the files within the TAR archive:
1099
  for audio_filename, audio_file in archive:
 
1123
 
1124
  yield audio_filename, {
1125
  "id": audio_filename,
1126
+ "text": text if not is_test else "",
1127
  "dataset": "earnings22",
1128
  "audio": {"path": path, "bytes": audio_file.read()}
1129
  }
 
1365
 
1366
  _SPGISPEECH_BASE_URL = "https://huggingface.co/datasets/kensho/spgispeech/resolve/main/data/"
1367
 
1368
+ _SPGISPEECH_AUDIO_BASE_URL = _SPGISPEECH_BASE_URL + "audio"
1369
 
1370
  _SPGISPEECH_SUBSET_TO_DIR = {
1371
  "s": ["s"],
 
1384
  "test": [f"test_part_{i}.tar.gz" for i in range(0, 3)],
1385
  }
1386
 
1387
+ _SPGISPEECH_META_BASE_URL = _SPGISPEECH_BASE_URL + "meta"
1388
 
1389
  _SPGISPEECH_META_FILENAMES = {
1390
  "s": "train_small.csv",
 
1416
 
1417
  _COMMON_VOICE_API_URL = "https://commonvoice.mozilla.org/api/v1"
1418
 
1419
+ _COMMON_VOICE_BUNDLE_URL_TEMPLATE = 'cv-corpus-9.0-2022-04-27/cv-corpus-9.0-2022-04-27-{locale}.tar.gz'
1420
+
1421
  _TEDLIUM_BASE_URL = "https://huggingface.co/datasets/LIUM/tedlium/resolve/main/TEDLIUM_release3/legacy/"
1422
 
1423
  _TEDLIUM_URLS = {