holylovenia commited on
Commit
cf15793
1 Parent(s): ac273f7

Upload commonvoice_120.py with huggingface_hub

Browse files
Files changed (1) hide show
  1. commonvoice_120.py +208 -0
commonvoice_120.py ADDED
@@ -0,0 +1,208 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ import csv
3
+ import json
4
+ import os
5
+ from pathlib import Path
6
+ from typing import Dict, List, Tuple
7
+
8
+ import datasets
9
+
10
+ from seacrowd.utils import schemas
11
+ from seacrowd.utils.configs import SEACrowdConfig
12
+ from seacrowd.utils.constants import Licenses, Tasks
13
+
14
+ _CITATION = """\
15
+ @inproceedings{commonvoice:2020,
16
+ 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.},
17
+ title = {Common Voice: A Massively-Multilingual Speech Corpus},
18
+ booktitle = {Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020)},
19
+ pages = {4211--4215},
20
+ year = 2020
21
+ }
22
+ """
23
+
24
+ _DATASETNAME = "commonvoice_120"
25
+
26
+ _DESCRIPTION = """\
27
+ The Common Mozilla Voice dataset consists of a unique MP3 and corresponding text file.
28
+ Many of the 26119 recorded hours in the dataset also include demographic metadata like age, sex, and accent that can help improve the accuracy of speech recognition engines.
29
+ The dataset currently consists of 17127 validated hours in 104 languages, but more voices and languages are always added.
30
+
31
+ Before using this dataloader, please accept the acknowledgement at https://huggingface.co/datasets/mozilla-foundation/common_voice_12_0 and use huggingface-cli login for authentication
32
+ """
33
+
34
+ _HOMEPAGE = "https://commonvoice.mozilla.org/en/datasets"
35
+
36
+ _LANGUAGES = ["cnh", "ind", "tha", "vie"]
37
+ _LANG_TO_CVLANG = {"cnh": "cnh", "ind": "id", "tha": "th", "vie": "vi"}
38
+
39
+ _AGE_TO_INT = {"": None, "teens": 10, "twenties": 20, "thirties": 30, "fourties": 40, "fifties": 50, "sixties": 60, "seventies": 70, "eighties": 80}
40
+
41
+ _LICENSE = Licenses.CC0_1_0.value
42
+
43
+ # Note: the dataset is gated in HuggingFace. It's public after providing access token
44
+ _LOCAL = False
45
+
46
+ _COMMONVOICE_URL_TEMPLATE = "https://huggingface.co/datasets/mozilla-foundation/common_voice_12_0/resolve/main/"
47
+ _URLS = {"audio": _COMMONVOICE_URL_TEMPLATE + "audio/{lang}/{split}/{lang}_{split}_{shard_idx}.tar", "transcript": _COMMONVOICE_URL_TEMPLATE + "transcript/{lang}/{split}.tsv", "n_shards": _COMMONVOICE_URL_TEMPLATE + "n_shards.json"}
48
+
49
+ _SUPPORTED_TASKS = [Tasks.SPEECH_RECOGNITION, Tasks.TEXT_TO_SPEECH]
50
+
51
+ _SOURCE_VERSION = "1.0.0"
52
+
53
+ _SEACROWD_VERSION = "2024.06.20"
54
+
55
+
56
+ class Commonvoice120(datasets.GeneratorBasedBuilder):
57
+ """This is the dataloader for CommonVoice 12.0 Mozilla"""
58
+
59
+ SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
60
+ SEACROWD_VERSION = datasets.Version(_SEACROWD_VERSION)
61
+
62
+ BUILDER_CONFIGS = (
63
+ *[
64
+ SEACrowdConfig(
65
+ name=f"{_DATASETNAME}_{lang}{'_' if lang else ''}source",
66
+ version=datasets.Version(_SOURCE_VERSION),
67
+ description=f"{_DATASETNAME} source schema for {lang}",
68
+ schema="source",
69
+ subset_id=f"{_DATASETNAME}{'_' if lang else ''}{lang}",
70
+ )
71
+ for lang in ["", *_LANGUAGES]
72
+ ],
73
+ *[
74
+ SEACrowdConfig(
75
+ name=f"{_DATASETNAME}_{lang}{'_' if lang else ''}seacrowd_sptext",
76
+ version=datasets.Version(_SEACROWD_VERSION),
77
+ description=f"{_DATASETNAME} SEACrowd schema for {lang}",
78
+ schema="seacrowd_sptext",
79
+ subset_id=f"{_DATASETNAME}{'_' if lang else ''}{lang}",
80
+ )
81
+ for lang in ["", *_LANGUAGES]
82
+ ],
83
+ )
84
+
85
+ DEFAULT_CONFIG_NAME = f"{_DATASETNAME}_source"
86
+
87
+ def _info(self) -> datasets.DatasetInfo:
88
+ if self.config.schema == "source":
89
+ features = datasets.Features(
90
+ {
91
+ "client_id": datasets.Value("string"),
92
+ "path": datasets.Value("string"),
93
+ "audio": datasets.features.Audio(sampling_rate=48_000),
94
+ "sentence": datasets.Value("string"),
95
+ "up_votes": datasets.Value("int64"),
96
+ "down_votes": datasets.Value("int64"),
97
+ "age": datasets.Value("string"),
98
+ "gender": datasets.Value("string"),
99
+ "accent": datasets.Value("string"),
100
+ "locale": datasets.Value("string"),
101
+ "segment": datasets.Value("string"),
102
+ }
103
+ )
104
+ elif self.config.schema == "seacrowd_sptext":
105
+ features = schemas.speech_text_features
106
+
107
+ return datasets.DatasetInfo(
108
+ description=_DESCRIPTION,
109
+ features=features,
110
+ homepage=_HOMEPAGE,
111
+ license=_LICENSE,
112
+ citation=_CITATION,
113
+ )
114
+
115
+ def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
116
+ lang_code = self.config.subset_id.split("_")[-1]
117
+ languages = [_LANG_TO_CVLANG.get(lang, lang) for lang in (_LANGUAGES if lang_code == "120" else [lang_code])]
118
+ n_shards_path = dl_manager.download_and_extract(_URLS["n_shards"])
119
+ with open(n_shards_path, encoding="utf-8") as f:
120
+ n_shards = json.load(f)
121
+
122
+ audio_urls = {}
123
+ meta_urls = {}
124
+ splits = ("train", "dev", "test")
125
+ for split in splits:
126
+ audio_urls[split] = [_URLS["audio"].format(lang=lang, split=split, shard_idx=i) for lang in languages for i in range(n_shards[lang][split])]
127
+ meta_urls[split] = [_URLS["transcript"].format(lang=lang, split=split) for lang in languages]
128
+ archive_paths = dl_manager.download(audio_urls)
129
+ local_extracted_archive_paths = dl_manager.extract(archive_paths)
130
+ meta_paths = dl_manager.download_and_extract(meta_urls)
131
+
132
+ split_names = {
133
+ "train": datasets.Split.TRAIN,
134
+ "dev": datasets.Split.VALIDATION,
135
+ "test": datasets.Split.TEST,
136
+ }
137
+ return [
138
+ datasets.SplitGenerator(
139
+ name=split_names.get(split, split),
140
+ gen_kwargs={
141
+ "local_extracted_archive_paths": local_extracted_archive_paths.get(split),
142
+ "audio_archives": [dl_manager.iter_archive(path) for path in archive_paths.get(split)],
143
+ "meta_paths": meta_paths[split],
144
+ "split": "train",
145
+ },
146
+ )
147
+ for split in splits
148
+ ]
149
+
150
+ def _generate_examples(self, local_extracted_archive_paths: [Path], audio_archives: [Path], meta_paths: [Path], split: str) -> Tuple[int, Dict]:
151
+ data_fields = list(self._info().features.keys())
152
+ metadata = {}
153
+ for meta_path in meta_paths:
154
+ with open(meta_path, encoding="utf-8") as f:
155
+ reader = csv.DictReader(f, delimiter="\t", quoting=csv.QUOTE_NONE)
156
+ for row in reader:
157
+ if not row["path"].endswith(".mp3"):
158
+ row["path"] += ".mp3"
159
+ if "accents" in row:
160
+ row["accent"] = row["accents"]
161
+ del row["accents"]
162
+ for field in data_fields:
163
+ if field not in row:
164
+ row[field] = ""
165
+ metadata[row["path"]] = row
166
+
167
+ if self.config.schema == "source":
168
+ for i, audio_archive in enumerate(audio_archives):
169
+ for path, file in audio_archive:
170
+ _, filename = os.path.split(path)
171
+ if filename in metadata:
172
+ src_result = dict(metadata[filename])
173
+ path = os.path.join(local_extracted_archive_paths[i], path)
174
+ result = {
175
+ "client_id": src_result["client_id"],
176
+ "path": path,
177
+ "audio": {"path": path, "bytes": file.read()},
178
+ "sentence": src_result["sentence"],
179
+ "up_votes": src_result["up_votes"],
180
+ "down_votes": src_result["down_votes"],
181
+ "age": src_result["age"],
182
+ "gender": src_result["gender"],
183
+ "accent": src_result["accent"],
184
+ "locale": src_result["locale"],
185
+ "segment": src_result["segment"],
186
+ }
187
+ yield path, result
188
+
189
+ elif self.config.schema == "seacrowd_sptext":
190
+ for i, audio_archive in enumerate(audio_archives):
191
+ for path, file in audio_archive:
192
+ _, filename = os.path.split(path)
193
+ if filename in metadata:
194
+ src_result = dict(metadata[filename])
195
+ # set the audio feature and the path to the extracted file
196
+ path = os.path.join(local_extracted_archive_paths[i], path)
197
+ result = {
198
+ "id": src_result["path"].replace(".mp3", ""),
199
+ "path": path,
200
+ "audio": {"path": path, "bytes": file.read()},
201
+ "text": src_result["sentence"],
202
+ "speaker_id": src_result["client_id"],
203
+ "metadata": {
204
+ "speaker_age": _AGE_TO_INT[src_result["age"]],
205
+ "speaker_gender": src_result["gender"],
206
+ },
207
+ }
208
+ yield path, result