101arrowz commited on
Commit
72e63cb
1 Parent(s): 2bb1a0a
Files changed (2) hide show
  1. README.md +193 -1
  2. vox_celeb.py +383 -0
README.md CHANGED
@@ -1,3 +1,195 @@
1
  ---
2
- license: cc-by-4.0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ annotations_creators:
3
+ - crowdsourced
4
+ language: []
5
+ language_creators:
6
+ - crowdsourced
7
+ license:
8
+ - cc-by-4.0
9
+ multilinguality:
10
+ - multilingual
11
+ pretty_name: VoxCeleb
12
+ size_categories:
13
+ - 1K<n<10K
14
+ - 10K<n<100K
15
+ - 100K<n<1M
16
+ source_datasets: []
17
+ tags: []
18
+ task_categories:
19
+ - automatic-speech-recognition
20
+ - audio-classification
21
+ - image-classification
22
+ task_ids:
23
+ - speaker-identification
24
  ---
25
+
26
+ # Dataset Card for VoxCeleb
27
+
28
+ ## Table of Contents
29
+ - [Table of Contents](#table-of-contents)
30
+ - [Dataset Description](#dataset-description)
31
+ - [Dataset Summary](#dataset-summary)
32
+ - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
33
+ - [Languages](#languages)
34
+ - [Dataset Structure](#dataset-structure)
35
+ - [Data Instances](#data-instances)
36
+ - [Data Fields](#data-fields)
37
+ - [Data Splits](#data-splits)
38
+ - [Dataset Creation](#dataset-creation)
39
+ - [Curation Rationale](#curation-rationale)
40
+ - [Source Data](#source-data)
41
+ - [Annotations](#annotations)
42
+ - [Personal and Sensitive Information](#personal-and-sensitive-information)
43
+ - [Considerations for Using the Data](#considerations-for-using-the-data)
44
+ - [Social Impact of Dataset](#social-impact-of-dataset)
45
+ - [Discussion of Biases](#discussion-of-biases)
46
+ - [Other Known Limitations](#other-known-limitations)
47
+ - [Additional Information](#additional-information)
48
+ - [Dataset Curators](#dataset-curators)
49
+ - [Licensing Information](#licensing-information)
50
+ - [Citation Information](#citation-information)
51
+ - [Contributions](#contributions)
52
+
53
+ ## Dataset Description
54
+
55
+ ### Dataset Summary
56
+
57
+ VoxCeleb is an audio-visual dataset consisting of short clips of human speech, extracted from interview videos uploaded to YouTube.
58
+
59
+ NOTE: Although this dataset can be automatically downloaded, you must manually request credentials to access it from the creators' website.
60
+
61
+ ### Supported Tasks and Leaderboards
62
+
63
+ [More Information Needed]
64
+
65
+ ### Languages
66
+
67
+ [More Information Needed]
68
+
69
+ ## Dataset Structure
70
+
71
+ ### Data Instances
72
+
73
+ Each datapoint has a path to the audio/video clip along with metadata about the speaker.
74
+
75
+ ```
76
+ {
77
+ 'file': '/datasets/downloads/extracted/[hash]/wav/id10271/_YimahVgI1A/00003.wav',
78
+ 'file_format': 'wav',
79
+ 'dataset_id': 'vox1',
80
+ 'speaker_id': 'id10271',
81
+ 'speaker_gender': 'm',
82
+ 'speaker_name': 'Ed_Westwick',
83
+ 'speaker_nationality': 'UK',
84
+ 'video_id': '_YimahVgI1A',
85
+ 'clip_id': '00003',
86
+ 'audio': {
87
+ 'path': '/datasets/downloads/extracted/[hash]/wav/id10271/_YimahVgI1A/00003.wav',
88
+ 'array': array([...], dtype=float32),
89
+ 'sampling_rate': 16000
90
+ }
91
+ }
92
+ ```
93
+
94
+ ### Data Fields
95
+
96
+ Each row includes the following fields:
97
+ - `file`: The path to the audio/video clip
98
+ - `file_format`: The file format in which the clip is stored (e.g. `wav`, `aac`, `mp4`)
99
+ - `dataset_id`: The ID of the dataset this clip is from (`vox1`, `vox2`)
100
+ - `speaker_id`: The ID of the speaker in this clip
101
+ - `speaker_gender`: The gender of the speaker (`m`/`f`)
102
+ - `speaker_name` (VoxCeleb1 only): The full name of the speaker in the clip
103
+ - `speaker_nationality` (VoxCeleb1 only): The speaker's country of origin
104
+ - `video_id`: The ID of the video from which this clip was taken
105
+ - `clip_index`: The index of the clip for this specific video
106
+ - `audio` (Audio dataset only): The audio signal data
107
+
108
+ ### Data Splits
109
+
110
+ The dataset has a predefined dev set and test set, but no training set. For training purposes, the dev set may be split into training and validation sets.
111
+
112
+ ## Dataset Creation
113
+
114
+ ### Curation Rationale
115
+
116
+ [More Information Needed]
117
+
118
+ ### Source Data
119
+
120
+ #### Initial Data Collection and Normalization
121
+
122
+ [More Information Needed]
123
+
124
+ #### Who are the source language producers?
125
+
126
+ [More Information Needed]
127
+
128
+ ### Annotations
129
+
130
+ #### Annotation process
131
+
132
+ [More Information Needed]
133
+
134
+ #### Who are the annotators?
135
+
136
+ [More Information Needed]
137
+
138
+ ### Personal and Sensitive Information
139
+
140
+ The dataset includes recordings of clips (mostly of celebrities and public figures) from public YouTube videos. The names of speakers in VoxCeleb1 are provided.
141
+
142
+ ## Considerations for Using the Data
143
+
144
+ ### Social Impact of Dataset
145
+
146
+ [More Information Needed]
147
+
148
+ ### Discussion of Biases
149
+
150
+ [More Information Needed]
151
+
152
+ ### Other Known Limitations
153
+
154
+ [More Information Needed]
155
+
156
+ ## Additional Information
157
+
158
+ ### Dataset Curators
159
+
160
+ [More Information Needed]
161
+
162
+ ### Licensing Information
163
+
164
+ [More Information Needed]
165
+
166
+ ### Citation Information
167
+
168
+ The VoxCeleb authors request that anyone who uses VoxCeleb1 or VoxCeleb2 includes the following three citations:
169
+ ```
170
+ @Article{Nagrani19,
171
+ author = "Arsha Nagrani and Joon~Son Chung and Weidi Xie and Andrew Zisserman",
172
+ title = "Voxceleb: Large-scale speaker verification in the wild",
173
+ journal = "Computer Science and Language",
174
+ year = "2019",
175
+ publisher = "Elsevier",
176
+ }
177
+
178
+ @InProceedings{Chung18b,
179
+ author = "Chung, J.~S. and Nagrani, A. and Zisserman, A.",
180
+ title = "VoxCeleb2: Deep Speaker Recognition",
181
+ booktitle = "INTERSPEECH",
182
+ year = "2018",
183
+ }
184
+
185
+ @InProceedings{Nagrani17,
186
+ author = "Nagrani, A. and Chung, J.~S. and Zisserman, A.",
187
+ title = "VoxCeleb: a large-scale speaker identification dataset",
188
+ booktitle = "INTERSPEECH",
189
+ year = "2017",
190
+ }
191
+ ```
192
+
193
+ ### Contributions
194
+
195
+ Thanks to [@101arrowz](https://github.com/101arrowz) for adding this dataset.
vox_celeb.py ADDED
@@ -0,0 +1,383 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2022 The HuggingFace Datasets Authors and Arjun Barrett.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+
16
+ # Lint as: python3
17
+ """VoxCeleb audio-visual human speech dataset."""
18
+
19
+ import json
20
+ import os
21
+ from getpass import getpass
22
+ from hashlib import sha256
23
+ from itertools import repeat
24
+ from multiprocessing import Manager, Pool, Process
25
+ from pathlib import Path
26
+ from queue import Queue
27
+ from shutil import copyfileobj
28
+ from typing import Optional, cast
29
+
30
+ import pandas as pd
31
+ import requests
32
+
33
+ import datasets
34
+
35
+ _CITATION = """\
36
+ @Article{Nagrani19,
37
+ author = "Arsha Nagrani and Joon~Son Chung and Weidi Xie and Andrew Zisserman",
38
+ title = "Voxceleb: Large-scale speaker verification in the wild",
39
+ journal = "Computer Science and Language",
40
+ year = "2019",
41
+ publisher = "Elsevier",
42
+ }
43
+
44
+ @InProceedings{Chung18b,
45
+ author = "Chung, J.~S. and Nagrani, A. and Zisserman, A.",
46
+ title = "VoxCeleb2: Deep Speaker Recognition",
47
+ booktitle = "INTERSPEECH",
48
+ year = "2018",
49
+ }
50
+
51
+ @InProceedings{Nagrani17,
52
+ author = "Nagrani, A. and Chung, J.~S. and Zisserman, A.",
53
+ title = "VoxCeleb: a large-scale speaker identification dataset",
54
+ booktitle = "INTERSPEECH",
55
+ year = "2017",
56
+ }
57
+ """
58
+
59
+ _DESCRIPTION = """\
60
+ VoxCeleb is an audio-visual dataset consisting of short clips of human speech, extracted from interview videos uploaded to YouTube
61
+ """
62
+
63
+ _URL = "https://mm.kaist.ac.kr/datasets/voxceleb"
64
+
65
+ _URLS = {
66
+ "video": {
67
+ "placeholder": "http://143.248.230.30/voxceleb/vox1a/vox2_dev_mp4_parta",
68
+ "dev": (
69
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_mp4_partaa",
70
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_mp4_partab",
71
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_mp4_partac",
72
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_mp4_partad",
73
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_mp4_partae",
74
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_mp4_partaf",
75
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_mp4_partag",
76
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_mp4_partah",
77
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_mp4_partai",
78
+ ),
79
+ "test": "http://143.248.230.30/voxceleb/vox1a/vox2_test_mp4.zip",
80
+ },
81
+ "audio1": {
82
+ "placeholder": "http://143.248.230.30/voxceleb/vox1a/vox1_dev_wav_parta",
83
+ "dev": (
84
+ "http://143.248.230.30/voxceleb/vox1a/vox1_dev_wav_partaa",
85
+ "http://143.248.230.30/voxceleb/vox1a/vox1_dev_wav_partab",
86
+ "http://143.248.230.30/voxceleb/vox1a/vox1_dev_wav_partac",
87
+ "http://143.248.230.30/voxceleb/vox1a/vox1_dev_wav_partad",
88
+ ),
89
+ "test": "http://143.248.230.30/voxceleb/vox1a/vox1_test_wav.zip",
90
+ },
91
+ "audio2": {
92
+ "placeholder": "http://143.248.230.30/voxceleb/vox1a/vox2_dev_aac_parta",
93
+ "dev": (
94
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_aac_partaa",
95
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_aac_partab",
96
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_aac_partac",
97
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_aac_partad",
98
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_aac_partae",
99
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_aac_partaf",
100
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_aac_partag",
101
+ "http://143.248.230.30/voxceleb/vox1a/vox2_dev_aac_partah",
102
+ ),
103
+ "test": "http://143.248.230.30/voxceleb/vox1a/vox2_test_aac.zip",
104
+ },
105
+ }
106
+
107
+ _DATASET_IDS = {"video": "vox2", "audio1": "vox1", "audio2": "vox2"}
108
+
109
+ _PLACEHOLDER_MAPS = dict(
110
+ value
111
+ for urls in _URLS.values()
112
+ for value in ((urls["placeholder"], urls["dev"]), (urls["test"], (urls["test"],)))
113
+ )
114
+
115
+
116
+ def _mp_download(
117
+ url: str,
118
+ tmp_path: Path,
119
+ cred_user: str,
120
+ cred_pass: str,
121
+ resume_pos: int,
122
+ length: int,
123
+ queue: Queue,
124
+ ):
125
+ if length == resume_pos:
126
+ return
127
+ with open(tmp_path, "ab" if resume_pos else "wb") as tmp:
128
+ headers: dict[str, str] = {}
129
+ if resume_pos != 0:
130
+ headers["Range"] = f"bytes={resume_pos}-"
131
+ response = requests.get(
132
+ url, auth=(cred_user, cred_pass), headers=headers, stream=True
133
+ )
134
+ if response.status_code >= 200 and response.status_code < 300:
135
+ for chunk in response.iter_content(chunk_size=65536):
136
+ queue.put(len(chunk))
137
+ tmp.write(chunk)
138
+ else:
139
+ raise ConnectionError("failed to fetch dataset")
140
+
141
+
142
+ # TODO: Name of the dataset usually matches the script name with CamelCase instead of snake_case
143
+ class VoxCeleb(datasets.GeneratorBasedBuilder):
144
+ """TODO: Short description of my dataset."""
145
+
146
+ VERSION = datasets.Version("1.0.0")
147
+
148
+ BUILDER_CONFIGS = [
149
+ datasets.BuilderConfig(
150
+ name="video", version=VERSION, description="Video clips of human speech"
151
+ ),
152
+ datasets.BuilderConfig(
153
+ name="audio", version=VERSION, description="Audio clips of human speech"
154
+ ),
155
+ datasets.BuilderConfig(
156
+ name="audio1",
157
+ version=datasets.Version("1.0.0"),
158
+ description="Audio clips of human speech from VoxCeleb1",
159
+ ),
160
+ datasets.BuilderConfig(
161
+ name="audio2",
162
+ version=datasets.Version("2.0.0"),
163
+ description="Audio clips of human speech from VoxCeleb2",
164
+ ),
165
+ ]
166
+
167
+ def _info(self):
168
+ features = cast(
169
+ dict,
170
+ {
171
+ "file": datasets.Value("string"),
172
+ "file_format": datasets.Value("string"),
173
+ "dataset_id": datasets.Value("string"),
174
+ "speaker_id": datasets.Value("string"),
175
+ "speaker_gender": datasets.Value("string"),
176
+ "video_id": datasets.Value("string"),
177
+ "clip_index": datasets.Value("int32"),
178
+ },
179
+ )
180
+ if self.config.name == "audio1":
181
+ features["speaker_name"] = datasets.Value("string")
182
+ features["speaker_nationality"] = datasets.Value("string")
183
+ if self.config.name.startswith("audio"):
184
+ features["audio"] = datasets.Audio(sampling_rate=16000)
185
+
186
+ return datasets.DatasetInfo(
187
+ description=_DESCRIPTION,
188
+ homepage=_URL,
189
+ supervised_keys=datasets.info.SupervisedKeysData("file", "speaker_id"),
190
+ features=datasets.Features(features),
191
+ citation=_CITATION,
192
+ )
193
+
194
+ def _split_generators(self, dl_manager):
195
+ if dl_manager.is_streaming:
196
+ raise TypeError("Streaming is not supported for VoxCeleb")
197
+ targets = (
198
+ ["audio1", "audio2"] if self.config.name == "audio" else [self.config.name]
199
+ )
200
+ cred_user: Optional[str] = os.environ.get("HUGGING_FACE_VOX_CELEB_USER")
201
+ cred_pass: Optional[str] = os.environ.get("HUGGING_FACE_VOX_CELEB_PASS")
202
+ creds_path = Path(
203
+ f"~/.huggingface/voxceleb_{self.VERSION}_credentials"
204
+ ).expanduser()
205
+
206
+ if cred_user is None and cred_pass is None:
207
+ if creds_path.exists():
208
+ with open(creds_path, "r") as creds:
209
+ cred_user, cred_pass = json.load(creds)
210
+ else:
211
+ print(
212
+ "You need a temporary username and password to access VoxCeleb.",
213
+ f"Go to the project homepage ({_URL}) and fill out the form to request credentials.",
214
+ )
215
+ cred_user = input("VoxCeleb username: ")
216
+ cred_pass = getpass("VoxCeleb password: ")
217
+
218
+ if not cred_user or not cred_pass:
219
+ raise ValueError("could not find username and password to log in")
220
+
221
+ saved_credentials = False
222
+
223
+ def save_credentials():
224
+ nonlocal saved_credentials, cred_user, cred_pass, creds_path
225
+ if not saved_credentials:
226
+ creds_path.parent.mkdir(exist_ok=True)
227
+ with open(creds_path, "w") as creds:
228
+ json.dump((cred_user, cred_pass), creds)
229
+ saved_credentials = True
230
+
231
+ def download_custom(placeholder_url: str, path: str):
232
+ nonlocal dl_manager, cred_user, cred_pass
233
+ sources = _PLACEHOLDER_MAPS[placeholder_url]
234
+ tmp_paths: list[Path] = []
235
+ lengths: list[int] = []
236
+ start_positions: list[int] = []
237
+ for url in sources:
238
+ head = requests.head(url, auth=(cred_user, cred_pass))
239
+ if head.status_code == 401:
240
+ raise ValueError("failed to authenticate with VoxCeleb host")
241
+ if head.status_code < 200 or head.status_code >= 300:
242
+ raise ValueError("failed to fetch dataset")
243
+ save_credentials()
244
+ content_length = head.headers.get("Content-Length")
245
+ if content_length is None:
246
+ raise ValueError("expected non-empty Content-Length")
247
+ content_length = int(content_length)
248
+ tmp_path = Path(path + "." + sha256(url.encode("utf-8")).hexdigest())
249
+ tmp_paths.append(tmp_path)
250
+ lengths.append(content_length)
251
+ start_positions.append(
252
+ tmp_path.stat().st_size
253
+ if tmp_path.exists() and dl_manager.download_config.resume_download
254
+ else 0
255
+ )
256
+
257
+ def progress(q: Queue[int], cur: int, total: int):
258
+ with datasets.utils.logging.tqdm(
259
+ unit="B",
260
+ unit_scale=True,
261
+ total=total,
262
+ initial=cur,
263
+ desc="Downloading",
264
+ disable=not datasets.utils.logging.is_progress_bar_enabled(),
265
+ ) as progress:
266
+ while cur < total:
267
+ try:
268
+ added = q.get(timeout=1)
269
+ progress.update(added)
270
+ cur += added
271
+ except:
272
+ continue
273
+
274
+ manager = Manager()
275
+ q = manager.Queue()
276
+ with Pool(len(sources)) as pool:
277
+ proc = Process(
278
+ target=progress,
279
+ args=(q, sum(start_positions), sum(lengths)),
280
+ daemon=True,
281
+ )
282
+ proc.start()
283
+ pool.starmap(
284
+ _mp_download,
285
+ zip(
286
+ sources,
287
+ tmp_paths,
288
+ repeat(cred_user),
289
+ repeat(cred_pass),
290
+ start_positions,
291
+ lengths,
292
+ repeat(q),
293
+ ),
294
+ )
295
+ pool.close()
296
+ proc.join()
297
+ with open(path, "wb") as out:
298
+ for tmp_path in tmp_paths:
299
+ with open(tmp_path, "rb") as tmp:
300
+ copyfileobj(tmp, out)
301
+ tmp_path.unlink()
302
+
303
+ metadata = dl_manager.download(
304
+ dict(
305
+ (
306
+ target,
307
+ f"https://mm.kaist.ac.kr/datasets/voxceleb/meta/{_DATASET_IDS[target]}_meta.csv",
308
+ )
309
+ for target in targets
310
+ )
311
+ )
312
+
313
+ mapped_paths = cast(
314
+ dict[str, dict[str, str]],
315
+ dl_manager.extract(
316
+ dl_manager.download_custom(
317
+ dict(
318
+ (
319
+ placeholder_key,
320
+ dict(
321
+ (target, _URLS[target][placeholder_key])
322
+ for target in targets
323
+ ),
324
+ )
325
+ for placeholder_key in ("placeholder", "test")
326
+ ),
327
+ download_custom,
328
+ )
329
+ ),
330
+ )
331
+
332
+ return [
333
+ datasets.SplitGenerator(
334
+ name="validation",
335
+ gen_kwargs={
336
+ "paths": mapped_paths["placeholder"],
337
+ "meta_paths": metadata,
338
+ },
339
+ ),
340
+ datasets.SplitGenerator(
341
+ name="test",
342
+ gen_kwargs={
343
+ "paths": mapped_paths["test"],
344
+ "meta_paths": metadata,
345
+ },
346
+ ),
347
+ ]
348
+
349
+ def _generate_examples(self, paths: dict[str, str], meta_paths: dict[str, str]):
350
+ key = 0
351
+ for conf in paths:
352
+ dataset_id = "vox1" if conf == "audio1" else "vox2"
353
+ meta = pd.read_csv(
354
+ meta_paths[conf],
355
+ sep="\t" if conf == "audio1" else " ,",
356
+ index_col=0,
357
+ engine="python",
358
+ )
359
+ dataset_path = next(Path(paths[conf]).iterdir())
360
+ dataset_format = dataset_path.name
361
+ for speaker_path in dataset_path.iterdir():
362
+ speaker = speaker_path.name
363
+ speaker_info = meta.loc[speaker]
364
+ for video in speaker_path.iterdir():
365
+ video_id = video.name
366
+ for clip in video.iterdir():
367
+ clip_index = int(clip.stem)
368
+ info = {
369
+ "file": str(clip),
370
+ "file_format": dataset_format,
371
+ "dataset_id": dataset_id,
372
+ "speaker_id": speaker,
373
+ "speaker_gender": speaker_info["Gender"],
374
+ "video_id": video_id,
375
+ "clip_index": clip_index,
376
+ }
377
+ if dataset_id == "vox1":
378
+ info["speaker_name"] = speaker_info["VGGFace1 ID"]
379
+ info["speaker_nationality"] = speaker_info["Nationality"]
380
+ if conf.startswith("audio"):
381
+ info["audio"] = info["file"]
382
+ yield key, info
383
+ key += 1