Datasets:

Modalities:
Audio
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
Dask
felixgwu commited on
Commit
a7bf486
1 Parent(s): 469bc51

fix vp_nel

Browse files
Files changed (2) hide show
  1. data/slue-vp_nel_blind.zip +2 -2
  2. slue-phase-2.py +9 -8
data/slue-vp_nel_blind.zip CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f889fae454a5de7ee71f430172057e8ddfb5f4c2c3d34623cf6e0e7ec75c6679
3
- size 168701758
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:62a1fd6d01a19a08042bd7e1ae8549c08db67001cecf0fd43c96d4da3154e1d7
3
+ size 168568066
slue-phase-2.py CHANGED
@@ -58,9 +58,9 @@ For questions from the other 4 datasets, their question texts, answer strings, a
58
 
59
  SLUE-SQA-5 also contains a subset of Spoken Wikipedia, including the audios placed in “document” directories and their transcripts (document_text and normalized_document_text column in .tsv files). Additionally, we provide the text-to-speech alignments (.txt files in “word2time” directories).These contents are licensed with the same Creative Commons (CC BY-SA 4.0) license as Spoken Wikipedia.
60
  =======================================================
61
- SLUE-VP-NEL Dataset
62
 
63
- SLUE-VP-NEL includes word-level time stamps for dev and test splits of the SLUE-voxpopuli corpus.
64
  For the dev split, the dataset also contains named entity annotations and corresponding time-stamps in a tsv format.
65
  =======================================================
66
 
@@ -106,6 +106,8 @@ def load_word2time(word2time_file):
106
 
107
  def parse_nel_time_spans(nel_timestamps):
108
  nel_timestamps = ast.literal_eval(nel_timestamps)
 
 
109
  return [
110
  {
111
  "ne_label": ne,
@@ -157,7 +159,7 @@ class SLUE2(datasets.GeneratorBasedBuilder):
157
  ),
158
  SLUE2Config(
159
  name="vp_nel",
160
- description="SLUE-VP-NEL set with named entity labels and time-stamps.",
161
  ),
162
  ]
163
 
@@ -209,10 +211,9 @@ class SLUE2(datasets.GeneratorBasedBuilder):
209
  elif self.config.name == "vp_nel":
210
  features = {
211
  "id": datasets.Value("string"),
212
- "split": datasets.Value("string"),
213
  "audio": datasets.Audio(sampling_rate=16_000),
214
  "speaker_id": datasets.Value("string"),
215
- "normalized_text": datasets.Value("string"),
216
  "word_timestamps": datasets.Sequence(
217
  {
218
  "word": datasets.Value("string"),
@@ -220,7 +221,7 @@ class SLUE2(datasets.GeneratorBasedBuilder):
220
  "end_sec": datasets.Value("float64"),
221
  }
222
  ),
223
- "normalized_nel": datasets.Sequence(
224
  {
225
  "ne_label": datasets.Value("string"),
226
  "start_char_idx": datasets.Value("int32"),
@@ -247,7 +248,6 @@ class SLUE2(datasets.GeneratorBasedBuilder):
247
 
248
  dl_dir = dl_manager.download_and_extract(_DL_URLS[config_name])
249
  data_dir = os.path.join(dl_dir, config_name)
250
- print(data_dir)
251
 
252
  splits = []
253
  if self.config.name in ["hvb", "sqa5"]:
@@ -352,7 +352,7 @@ class SLUE2(datasets.GeneratorBasedBuilder):
352
  "word2time": load_word2time(word2time_file),
353
  "answer_spans": parse_qa_answer_spans(row.get("answer_spans", "[]")),
354
  }
355
- elif self.config.name == "slue_nel":
356
  split = "test" if "test" in filepath else "dev"
357
  utt_id = row["id"]
358
  word_alignments_fn = os.path.join(
@@ -360,6 +360,7 @@ class SLUE2(datasets.GeneratorBasedBuilder):
360
  )
361
  audio_file = os.path.join(
362
  data_dir,
 
363
  split,
364
  f"{utt_id}.ogg",
365
  )
 
58
 
59
  SLUE-SQA-5 also contains a subset of Spoken Wikipedia, including the audios placed in “document” directories and their transcripts (document_text and normalized_document_text column in .tsv files). Additionally, we provide the text-to-speech alignments (.txt files in “word2time” directories).These contents are licensed with the same Creative Commons (CC BY-SA 4.0) license as Spoken Wikipedia.
60
  =======================================================
61
+ SLUE-vp_nel Dataset
62
 
63
+ SLUE-vp_nel includes word-level time stamps for dev and test splits of the SLUE-voxpopuli corpus.
64
  For the dev split, the dataset also contains named entity annotations and corresponding time-stamps in a tsv format.
65
  =======================================================
66
 
 
106
 
107
  def parse_nel_time_spans(nel_timestamps):
108
  nel_timestamps = ast.literal_eval(nel_timestamps)
109
+ if nel_timestamps is None:
110
+ return []
111
  return [
112
  {
113
  "ne_label": ne,
 
159
  ),
160
  SLUE2Config(
161
  name="vp_nel",
162
+ description="SLUE-vp_nel set with named entity labels and time-stamps.",
163
  ),
164
  ]
165
 
 
211
  elif self.config.name == "vp_nel":
212
  features = {
213
  "id": datasets.Value("string"),
 
214
  "audio": datasets.Audio(sampling_rate=16_000),
215
  "speaker_id": datasets.Value("string"),
216
+ "text": datasets.Value("string"),
217
  "word_timestamps": datasets.Sequence(
218
  {
219
  "word": datasets.Value("string"),
 
221
  "end_sec": datasets.Value("float64"),
222
  }
223
  ),
224
+ "ne_timestamps": datasets.Sequence(
225
  {
226
  "ne_label": datasets.Value("string"),
227
  "start_char_idx": datasets.Value("int32"),
 
248
 
249
  dl_dir = dl_manager.download_and_extract(_DL_URLS[config_name])
250
  data_dir = os.path.join(dl_dir, config_name)
 
251
 
252
  splits = []
253
  if self.config.name in ["hvb", "sqa5"]:
 
352
  "word2time": load_word2time(word2time_file),
353
  "answer_spans": parse_qa_answer_spans(row.get("answer_spans", "[]")),
354
  }
355
+ elif self.config.name == "vp_nel":
356
  split = "test" if "test" in filepath else "dev"
357
  utt_id = row["id"]
358
  word_alignments_fn = os.path.join(
 
360
  )
361
  audio_file = os.path.join(
362
  data_dir,
363
+ 'audio',
364
  split,
365
  f"{utt_id}.ogg",
366
  )