phongdtd commited on
Commit
a873fa7
1 Parent(s): 1c1db08

update link

Browse files
.idea/custom_common_voice.iml CHANGED
@@ -4,7 +4,7 @@
4
  <content url="file://$MODULE_DIR$">
5
  <excludeFolder url="file://$MODULE_DIR$/venv" />
6
  </content>
7
- <orderEntry type="jdk" jdkName="Python 3.8 (custom_common_voice)" jdkType="Python SDK" />
8
  <orderEntry type="sourceFolder" forTests="false" />
9
  </component>
10
  </module>
 
4
  <content url="file://$MODULE_DIR$">
5
  <excludeFolder url="file://$MODULE_DIR$/venv" />
6
  </content>
7
+ <orderEntry type="inheritedJdk" />
8
  <orderEntry type="sourceFolder" forTests="false" />
9
  </component>
10
  </module>
README.md CHANGED
@@ -1,202 +1,30 @@
1
  ---
2
- pretty_name: Common Voice
3
- annotations_creators:
4
- - crowdsourced
5
- language_creators:
6
  - crowdsourced
7
- languages:
8
- - ab
9
- - ar
10
- - as
11
- - br
12
- - ca
13
- - cnh
14
- - cs
15
- - cv
16
- - cy
17
- - de
18
- - dv
19
- - el
20
- - en
21
- - eo
22
- - es
23
- - et
24
- - eu
25
- - fa
26
- - fi
27
- - fr
28
- - fy-NL
29
- - ga-IE
30
- - hi
31
- - hsb
32
- - hu
33
- - ia
34
- - id
35
- - it
36
- - ja
37
- - ka
38
- - kab
39
- - ky
40
- - lg
41
- - lt
42
- - lv
43
- - mn
44
- - mt
45
- - nl
46
- - or
47
- - pa-IN
48
- - pl
49
- - pt
50
- - rm-sursilv
51
- - rm-vallader
52
- - ro
53
- - ru
54
- - rw
55
- - sah
56
- - sl
57
- - sv-SE
58
- - ta
59
- - th
60
- - tr
61
- - tt
62
- - uk
63
  - vi
64
- - vot
65
- - zh-CN
66
- - zh-HK
67
- - zh-TW
68
- licenses:
69
  - cc0-1.0
 
70
  multilinguality:
71
- - multilingual
72
- size_categories:
73
- ab:
74
- - n<1K
75
- ar:
76
- - 10K<n<100K
77
- as:
78
- - n<1K
79
- br:
80
- - 10K<n<100K
81
- ca:
82
- - 100K<n<1M
83
- cnh:
84
- - 1K<n<10K
85
- cs:
86
- - 10K<n<100K
87
- cv:
88
- - 10K<n<100K
89
- cy:
90
- - 10K<n<100K
91
- de:
92
- - 100K<n<1M
93
- dv:
94
- - 1K<n<10K
95
- el:
96
- - 10K<n<100K
97
- en:
98
- - 100K<n<1M
99
- eo:
100
- - 10K<n<100K
101
- es:
102
- - 100K<n<1M
103
- et:
104
- - 10K<n<100K
105
- eu:
106
- - 10K<n<100K
107
- fa:
108
- - 10K<n<100K
109
- fi:
110
- - 1K<n<10K
111
- fr:
112
- - 100K<n<1M
113
- fy-NL:
114
- - 10K<n<100K
115
- ga-IE:
116
- - 1K<n<10K
117
- hi:
118
- - n<1K
119
- hsb:
120
- - 1K<n<10K
121
- hu:
122
- - 1K<n<10K
123
- ia:
124
- - 1K<n<10K
125
- id:
126
- - 10K<n<100K
127
- it:
128
- - 100K<n<1M
129
- ja:
130
- - 1K<n<10K
131
- ka:
132
- - 1K<n<10K
133
- kab:
134
- - 100K<n<1M
135
- ky:
136
- - 10K<n<100K
137
- lg:
138
- - 1K<n<10K
139
- lt:
140
- - 1K<n<10K
141
- lv:
142
- - 1K<n<10K
143
- mn:
144
- - 1K<n<10K
145
- mt:
146
- - 10K<n<100K
147
- nl:
148
- - 10K<n<100K
149
- or:
150
- - 1K<n<10K
151
- pa-IN:
152
- - 1K<n<10K
153
- pl:
154
- - 10K<n<100K
155
- pt:
156
- - 10K<n<100K
157
- rm-sursilv:
158
- - 1K<n<10K
159
- rm-vallader:
160
- - 1K<n<10K
161
- ro:
162
- - 1K<n<10K
163
- ru:
164
- - 10K<n<100K
165
- rw:
166
- - 100K<n<1M
167
- sah:
168
- - 1K<n<10K
169
- sl:
170
- - 1K<n<10K
171
- sv-SE:
172
- - 1K<n<10K
173
- ta:
174
- - 10K<n<100K
175
- th:
176
- - 10K<n<100K
177
- tr:
178
- - 1K<n<10K
179
- tt:
180
- - 10K<n<100K
181
- uk:
182
- - 10K<n<100K
183
  vi:
184
- - 1K<n<10K
185
- vot:
186
- - n<1K
187
- zh-CN:
188
- - 10K<n<100K
189
- zh-HK:
190
- - 10K<n<100K
191
- zh-TW:
192
- - 10K<n<100K
193
  source_datasets:
194
- - extended|common_voice
 
195
  task_categories:
196
  - speech-processing
 
197
  task_ids:
198
  - automatic-speech-recognition
199
- paperswithcode_id: common-voice
200
  ---
201
 
202
  # Dataset Card for common_voice
@@ -227,17 +55,15 @@ paperswithcode_id: common-voice
227
 
228
  ## Dataset Description
229
 
230
- - **Homepage:** https://commonvoice.mozilla.org/en/datasets
231
- - **Repository:** https://github.com/common-voice/common-voice
232
- - **Paper:** https://commonvoice.mozilla.org/en/datasets
233
  - **Leaderboard:** [Needs More Information]
234
  - **Point of Contact:** [Needs More Information]
235
 
236
  ### Dataset Summary
237
 
238
- The Common Voice dataset consists of a unique MP3 and corresponding text file. Many of the 9,283 recorded hours in the dataset also include demographic metadata like age, sex, and accent that can help train the accuracy of speech recognition engines.
239
-
240
- The dataset currently consists of 7,335 validated hours in 60 languages, but were always adding more voices and languages. Take a look at our Languages page to request a language or start contributing.
241
 
242
  ### Supported Tasks and Leaderboards
243
 
@@ -245,7 +71,7 @@ The dataset currently consists of 7,335 validated hours in 60 languages, but we
245
 
246
  ### Languages
247
 
248
- English
249
 
250
  ## Dataset Structure
251
 
@@ -254,47 +80,27 @@ English
254
  A typical data point comprises the path to the audio file, called path and its sentence. Additional fields include accent, age, client_id, up_votes down_votes, gender, locale and segment.
255
 
256
  `
257
- {'accent': 'netherlands', 'age': 'fourties', 'client_id': 'bbbcb732e0f422150c30ff3654bbab572e2a617da107bca22ff8b89ab2e4f124d03b6a92c48322862f60bd0179ae07baf0f9b4f9c4e11d581e0cec70f703ba54', 'down_votes': 0, 'gender': 'male', 'locale': 'nl', 'path': 'nl/clips/common_voice_nl_23522441.mp3', 'segment': "''", 'sentence': 'Ik vind dat een dubieuze procedure.', 'up_votes': 2, 'audio': {'path': `nl/clips/common_voice_nl_23522441.mp3', 'array': array([-0.00048828, -0.00018311, -0.00137329, ..., 0.00079346, 0.00091553, 0.00085449], dtype=float32), 'sampling_rate': 48000}
 
 
 
 
 
258
  `
259
 
260
  ### Data Fields
261
 
262
- client_id: An id for which client (voice) made the recording
263
-
264
- path: The path to the audio file
265
 
266
  audio: A dictionary containing the path to the downloaded audio file, the decoded audio array, and the sampling rate. Note that when accessing the audio column: `dataset[0]["audio"]` the audio file is automatically decoded and resampled to `dataset.features["audio"].sampling_rate`. Decoding and resampling of a large number of audio files might take a significant amount of time. Thus it is important to first query the sample index before the `"audio"` column, *i.e.* `dataset[0]["audio"]` should **always** be preferred over `dataset["audio"][0]`.
267
 
268
- sentence: The sentence the user was prompted to speak
269
-
270
- up_votes: How many upvotes the audio file has received from reviewers
271
-
272
- down_votes: How many downvotes the audio file has received from reviewers
273
-
274
- age: The age of the speaker.
275
-
276
- gender: The gender of the speaker
277
-
278
- accent: Accent of the speaker
279
-
280
- locale: The locale of the speaker
281
-
282
- segment: Usually empty field
283
 
284
  ### Data Splits
285
 
286
- The speech material has been subdivided into portions for dev, train, test, validated, invalidated, reported and other.
287
-
288
- The validated data is data that has been validated with reviewers and recieved upvotes that the data is of high quality.
289
-
290
- The invalidated data is data has been invalidated by reviewers
291
- and recieved downvotes that the data is of low quality.
292
 
293
- The reported data is data that has been reported, for different reasons.
294
-
295
- The other data is data that has not yet been reviewed.
296
-
297
- The dev, test, train are all data that has been reviewed, deemed of high quality and split into dev, test and train.
298
 
299
  ## Dataset Creation
300
 
@@ -324,13 +130,13 @@ The dev, test, train are all data that has been reviewed, deemed of high quality
324
 
325
  ### Personal and Sensitive Information
326
 
327
- The dataset consists of people who have donated their voice online. You agree to not attempt to determine the identity of speakers in the Common Voice dataset.
328
 
329
  ## Considerations for Using the Data
330
 
331
  ### Social Impact of Dataset
332
 
333
- The dataset consists of people who have donated their voice online. You agree to not attempt to determine the identity of speakers in the Common Voice dataset.
334
 
335
  ### Discussion of Biases
336
 
@@ -348,20 +154,12 @@ The dataset consists of people who have donated their voice online. You agree t
348
 
349
  ### Licensing Information
350
 
351
- Public Domain, [CC-0](https://creativecommons.org/share-your-work/public-domain/cc0/)
352
 
353
  ### Citation Information
354
 
355
- ```
356
- @inproceedings{commonvoice:2020,
357
- author = {Ardila, R. and Branson, M. and Davis, K. and Henretty, M. and Kohler, M. and Meyer, J. and Morais, R. and Saunders, L. and Tyers, F. M. and Weber, G.},
358
- title = {Common Voice: A Massively-Multilingual Speech Corpus},
359
- booktitle = {Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020)},
360
- pages = {4211--4215},
361
- year = 2020
362
- }
363
- ```
364
 
365
  ### Contributions
366
 
367
- Thanks to [@BirgerMoell](https://github.com/BirgerMoell) for adding this dataset.
 
1
  ---
2
+ Pretty_name: Youtube Casual Audio
3
+
4
+ Annotations_creators:
 
5
  - crowdsourced
6
+
7
+ Language_creators:
8
+ - datlq
9
+
10
+ Languages:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
11
  - vi
12
+
13
+ Licenses:
 
 
 
14
  - cc0-1.0
15
+
16
  multilinguality:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
17
  vi:
18
+ - 190K<n<200K
19
+
 
 
 
 
 
 
 
20
  source_datasets:
21
+ - extended|youtube
22
+
23
  task_categories:
24
  - speech-processing
25
+
26
  task_ids:
27
  - automatic-speech-recognition
 
28
  ---
29
 
30
  # Dataset Card for common_voice
 
55
 
56
  ## Dataset Description
57
 
58
+ - **Homepage:** [Needs More Information]
59
+ - **Repository:** [Needs More Information]
60
+ - **Paper:** [Needs More Information]
61
  - **Leaderboard:** [Needs More Information]
62
  - **Point of Contact:** [Needs More Information]
63
 
64
  ### Dataset Summary
65
 
66
+ [Needs More Information]
 
 
67
 
68
  ### Supported Tasks and Leaderboards
69
 
 
71
 
72
  ### Languages
73
 
74
+ Vietnamese
75
 
76
  ## Dataset Structure
77
 
 
80
  A typical data point comprises the path to the audio file, called path and its sentence. Additional fields include accent, age, client_id, up_votes down_votes, gender, locale and segment.
81
 
82
  `
83
+ {
84
+ 'file_path': 'audio/_1OsFqkFI38_34.304_39.424.wav',
85
+ 'script': 'Ik vind dat een dubieuze procedure.',
86
+ 'audio': {'path': 'audio/_1OsFqkFI38_34.304_39.424.wav',
87
+ 'array': array([-0.00048828, -0.00018311, -0.00137329, ..., 0.00079346, 0.00091553, 0.00085449], dtype=float32),
88
+ 'sampling_rate': 16000}
89
  `
90
 
91
  ### Data Fields
92
 
93
+ file_path: The path to the audio file
 
 
94
 
95
  audio: A dictionary containing the path to the downloaded audio file, the decoded audio array, and the sampling rate. Note that when accessing the audio column: `dataset[0]["audio"]` the audio file is automatically decoded and resampled to `dataset.features["audio"].sampling_rate`. Decoding and resampling of a large number of audio files might take a significant amount of time. Thus it is important to first query the sample index before the `"audio"` column, *i.e.* `dataset[0]["audio"]` should **always** be preferred over `dataset["audio"][0]`.
96
 
97
+ script: The sentence the user was prompted to speak
 
 
 
 
 
 
 
 
 
 
 
 
 
 
98
 
99
  ### Data Splits
100
 
101
+ The speech material has been subdivided into portions for train, test, validated.
 
 
 
 
 
102
 
103
+ The val, test, train are all data that has been reviewed, deemed of high quality and split into val, test and train.
 
 
 
 
104
 
105
  ## Dataset Creation
106
 
 
130
 
131
  ### Personal and Sensitive Information
132
 
133
+ [Needs More Information]
134
 
135
  ## Considerations for Using the Data
136
 
137
  ### Social Impact of Dataset
138
 
139
+ [Needs More Information]
140
 
141
  ### Discussion of Biases
142
 
 
154
 
155
  ### Licensing Information
156
 
157
+ [Needs More Information]
158
 
159
  ### Citation Information
160
 
161
+ [Needs More Information]
 
 
 
 
 
 
 
 
162
 
163
  ### Contributions
164
 
165
+ Thanks to [@datlq](https://github.com/datlq98) for adding this dataset.
dataset_infos.json CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6474db188c1ee3b46ac983c0c6febbe0598953626075303f567210caf11cde2c
3
- size 1659
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:04c7ab7fce049fbadb625937bd5594f09763621100cf5ad7601850795f702654
3
+ size 1081
custom_common_voice.py → youtube_casual_audio.py RENAMED
@@ -21,25 +21,22 @@ import pandas as pd
21
  import re
22
 
23
 
24
- _DATA_URL = "https://dutudn-my.sharepoint.com/:u:/g/personal/122180028_sv1_dut_udn_vn/EYsBHQzK4JVFhmN50e5vRFMBizbGJGXe_HlxV9uRlLaTyg?e=s1czWW?download=1"
25
  _PROMPTS_URLS = {
26
- "train": "https://drive.google.com/uc?export=download&id=13sANpjVoF9FIXj_rGGNvVNuK0GscPVsW",
27
- "test": "https://drive.google.com/uc?export=download&id=173oUWFMbeFUBnfoVke4dH2fiHdgOu9xb",
28
- "validation": "https://drive.google.com/uc?export=download&id=1J1zTG0IMPIRWnnw3dyr2UTyiq-KvlcX5"
29
  }
30
 
31
  _DESCRIPTION = """\
32
- Common Voice is Mozilla's initiative to help teach machines how real people speak.
33
- The dataset currently consists of 7,335 validated hours of speech in 60 languages, but we’re always adding more voices
34
- and languages.
35
  """
36
 
37
  _LANGUAGES = {
38
  "vi": {
39
  "Language": "Vietnamese",
40
- "Date": "2020-12-11",
41
- "Size": "50 MB",
42
- "Version": "vi_1h_2020-12-11",
43
  "Validated_Hr_Total": 0.74,
44
  "Overall_Hr_Total": 1,
45
  "Number_Of_Voice": 62,
@@ -83,11 +80,6 @@ class CustomCommonVoice(datasets.GeneratorBasedBuilder):
83
  name=lang_id,
84
  language=_LANGUAGES[lang_id]["Language"],
85
  sub_version=_LANGUAGES[lang_id]["Version"],
86
- # date=_LANGUAGES[lang_id]["Date"],
87
- # size=_LANGUAGES[lang_id]["Size"],
88
- # val_hrs=_LANGUAGES[lang_id]["Validated_Hr_Total"],
89
- # total_hrs=_LANGUAGES[lang_id]["Overall_Hr_Total"],
90
- # num_of_voice=_LANGUAGES[lang_id]["Number_Of_Voice"],
91
  )
92
  for lang_id in _LANGUAGES.keys()
93
  ]
@@ -115,8 +107,7 @@ class CustomCommonVoice(datasets.GeneratorBasedBuilder):
115
  """Returns SplitGenerators."""
116
  archive = dl_manager.download(_DATA_URL)
117
  tsv_files = dl_manager.download(_PROMPTS_URLS)
118
- path_to_data = "content/data_2/"
119
- path_to_clips = path_to_data + "audio"
120
 
121
  return [
122
  datasets.SplitGenerator(
@@ -124,7 +115,7 @@ class CustomCommonVoice(datasets.GeneratorBasedBuilder):
124
  gen_kwargs={
125
  "tsv_files": tsv_files["train"],
126
  "audio_files": dl_manager.iter_archive(archive),
127
- "path_to_clips": path_to_clips,
128
  },
129
  ),
130
  datasets.SplitGenerator(
@@ -132,7 +123,7 @@ class CustomCommonVoice(datasets.GeneratorBasedBuilder):
132
  gen_kwargs={
133
  "tsv_files": tsv_files["test"],
134
  "audio_files": dl_manager.iter_archive(archive),
135
- "path_to_clips": path_to_clips,
136
  },
137
  ),
138
  datasets.SplitGenerator(
@@ -140,25 +131,9 @@ class CustomCommonVoice(datasets.GeneratorBasedBuilder):
140
  gen_kwargs={
141
  "tsv_files": tsv_files["validation"],
142
  "audio_files": dl_manager.iter_archive(archive),
143
- "path_to_clips": path_to_clips,
144
  },
145
  ),
146
- # datasets.SplitGenerator(
147
- # name="other",
148
- # gen_kwargs={
149
- # "files": dl_manager.iter_archive(archive),
150
- # "filepath": "/".join([path_to_data, "other.tsv"]),
151
- # "path_to_clips": path_to_clips,
152
- # },
153
- # ),
154
- # datasets.SplitGenerator(
155
- # name="invalidated",
156
- # gen_kwargs={
157
- # "files": dl_manager.iter_archive(archive),
158
- # "filepath": "/".join([path_to_data, "invalidated.tsv"]),
159
- # "path_to_clips": path_to_clips,
160
- # },
161
- # ),
162
  ]
163
 
164
  def _generate_examples(self, tsv_files, audio_files, path_to_clips):
@@ -190,49 +165,8 @@ class CustomCommonVoice(datasets.GeneratorBasedBuilder):
190
  "duration": duration
191
  }
192
 
193
- # inside_clips_dir = False
194
-
195
  for path, f in audio_files:
196
  if path.startswith(path_to_clips):
197
- # inside_clips_dir = True
198
  if path in examples:
199
  audio = {"path": path, "bytes": f.read()}
200
  yield path, {**examples[path], "audio": audio}
201
- # elif "custom_common_voice.tsv" in path:
202
- # continue
203
- # elif ".txt" in path:
204
- # continue
205
- # elif inside_clips_dir:
206
- # break
207
-
208
- # for path, f in tsv_files:
209
- # if path == filepath:
210
- # metadata_found = True
211
- # lines = f.readlines()
212
- # headline = lines[0]
213
- # column_names = headline.strip().split("\t")
214
- # assert (
215
- # column_names == data_fields
216
- # ), f"The file should have {data_fields} as column names, but has {column_names}"
217
- # for line in lines[1:]:
218
- # field_values = line.strip().split("\t")
219
- # # set full path for mp3 audio file
220
- # audio_path = path_to_clips + "/" + field_values[path_idx]
221
- # all_field_values[audio_path] = field_values
222
- # elif path.startswith(path_to_clips):
223
- # assert metadata_found, "Found audio clips before the metadata TSV file."
224
- # if not all_field_values:
225
- # break
226
- # if path in all_field_values:
227
- # field_values = all_field_values[path]
228
- #
229
- # # if data is incomplete, fill with empty values
230
- # if len(field_values) < len(data_fields):
231
- # field_values += (len(data_fields) - len(field_values)) * ["''"]
232
- #
233
- # result = {key: value for key, value in zip(data_fields, field_values)}
234
- #
235
- # # set audio feature
236
- # result["audio"] = {"path": path, "bytes": f.read()}
237
- #
238
- # yield path, result
 
21
  import re
22
 
23
 
24
+ _DATA_URL = "https://dutudn-my.sharepoint.com/:u:/g/personal/122180028_sv1_dut_udn_vn/Ed5mI5CjXIxHgb2qqPOElj0BBgn7FGT75SUgPdIuMS1LDw?download=1"
25
  _PROMPTS_URLS = {
26
+ "train": "https://drive.google.com/uc?export=download&id=1s5d-1ZTzcxsnxUjiBLsv9KCB-yBcXyQ9",
27
+ "test": "https://drive.google.com/uc?export=download&id=1-l1QdNQ98DGZM63-GOKIVnFvxTz2SGeK",
28
+ "validation": "https://drive.google.com/uc?export=download&id=1GM_6s5icko6zRrldx8LcbANyl0geMSl8"
29
  }
30
 
31
  _DESCRIPTION = """\
 
 
 
32
  """
33
 
34
  _LANGUAGES = {
35
  "vi": {
36
  "Language": "Vietnamese",
37
+ "Date": "2021-12-11",
38
+ "Size": "17000 MB",
39
+ "Version": "vi_100h_2020-12-11",
40
  "Validated_Hr_Total": 0.74,
41
  "Overall_Hr_Total": 1,
42
  "Number_Of_Voice": 62,
 
80
  name=lang_id,
81
  language=_LANGUAGES[lang_id]["Language"],
82
  sub_version=_LANGUAGES[lang_id]["Version"],
 
 
 
 
 
83
  )
84
  for lang_id in _LANGUAGES.keys()
85
  ]
 
107
  """Returns SplitGenerators."""
108
  archive = dl_manager.download(_DATA_URL)
109
  tsv_files = dl_manager.download(_PROMPTS_URLS)
110
+ path_to_data = "audio"
 
111
 
112
  return [
113
  datasets.SplitGenerator(
 
115
  gen_kwargs={
116
  "tsv_files": tsv_files["train"],
117
  "audio_files": dl_manager.iter_archive(archive),
118
+ "path_to_clips": path_to_data,
119
  },
120
  ),
121
  datasets.SplitGenerator(
 
123
  gen_kwargs={
124
  "tsv_files": tsv_files["test"],
125
  "audio_files": dl_manager.iter_archive(archive),
126
+ "path_to_clips": path_to_data,
127
  },
128
  ),
129
  datasets.SplitGenerator(
 
131
  gen_kwargs={
132
  "tsv_files": tsv_files["validation"],
133
  "audio_files": dl_manager.iter_archive(archive),
134
+ "path_to_clips": path_to_data,
135
  },
136
  ),
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
137
  ]
138
 
139
  def _generate_examples(self, tsv_files, audio_files, path_to_clips):
 
165
  "duration": duration
166
  }
167
 
 
 
168
  for path, f in audio_files:
169
  if path.startswith(path_to_clips):
 
170
  if path in examples:
171
  audio = {"path": path, "bytes": f.read()}
172
  yield path, {**examples[path], "audio": audio}