Datasets:
Commit
•
58b647c
1
Parent(s):
3476ae7
add dataset metadate
Browse files- voxpopuli.py +61 -14
voxpopuli.py
CHANGED
@@ -7,6 +7,39 @@ from tqdm.auto import tqdm
|
|
7 |
import datasets
|
8 |
|
9 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
_LANGUAGES = sorted(
|
11 |
[
|
12 |
"en", "de", "fr", "es", "pl", "it", "ro", "hu", "cs", "nl", "fi", "hr",
|
@@ -15,6 +48,14 @@ _LANGUAGES = sorted(
|
|
15 |
)
|
16 |
_LANGUAGES_V2 = [f"{x}_v2" for x in _LANGUAGES]
|
17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
_YEARS = list(range(2009, 2020 + 1))
|
19 |
|
20 |
# unnecessary
|
@@ -22,7 +63,7 @@ _CONFIG_TO_LANGS = {
|
|
22 |
"400k": _LANGUAGES,
|
23 |
"100k": _LANGUAGES,
|
24 |
"10k": _LANGUAGES,
|
25 |
-
"asr":
|
26 |
}
|
27 |
|
28 |
_CONFIG_TO_YEARS = {
|
@@ -43,7 +84,11 @@ _BASE_URL = "https://dl.fbaipublicfiles.com/voxpopuli/"
|
|
43 |
|
44 |
_DATA_URL = _BASE_URL + "audios/{lang}_{year}.tar"
|
45 |
|
46 |
-
|
|
|
|
|
|
|
|
|
47 |
|
48 |
|
49 |
class VoxpopuliConfig(datasets.BuilderConfig):
|
@@ -57,24 +102,23 @@ class VoxpopuliConfig(datasets.BuilderConfig):
|
|
57 |
"""
|
58 |
super().__init__(name=name, **kwargs)
|
59 |
name = name.split("_")[0]
|
60 |
-
self.languages = [name] if name in _LANGUAGES else
|
61 |
self.years = _CONFIG_TO_YEARS[name]
|
62 |
|
63 |
|
64 |
class Voxpopuli(datasets.GeneratorBasedBuilder):
|
65 |
-
"""The
|
66 |
|
67 |
-
VERSION = datasets.Version("1.
|
68 |
BUILDER_CONFIGS = [
|
69 |
VoxpopuliConfig(
|
70 |
name=name,
|
71 |
-
|
72 |
-
description="", # TODO
|
73 |
)
|
74 |
for name in _LANGUAGES + _LANGUAGES_V2 + ["10k", "100k", "400k"]
|
75 |
]
|
76 |
# DEFAULT_CONFIG_NAME = "400k"
|
77 |
-
DEFAULT_WRITER_BATCH_SIZE = 256
|
78 |
|
79 |
def _info(self):
|
80 |
try:
|
@@ -98,14 +142,14 @@ class Voxpopuli(datasets.GeneratorBasedBuilder):
|
|
98 |
}
|
99 |
)
|
100 |
return datasets.DatasetInfo(
|
101 |
-
|
102 |
features=features,
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
)
|
107 |
|
108 |
-
def
|
109 |
# from https://github.com/facebookresearch/voxpopuli/blob/main/voxpopuli/get_unlabelled_data.py#L34
|
110 |
def predicate(id_):
|
111 |
is_plenary = id_.find("PLENARY") > -1
|
@@ -133,8 +177,11 @@ class Voxpopuli(datasets.GeneratorBasedBuilder):
|
|
133 |
|
134 |
return metadata
|
135 |
|
|
|
|
|
|
|
136 |
def _split_generators(self, dl_manager):
|
137 |
-
metadata_path = dl_manager.download_and_extract(
|
138 |
|
139 |
urls = [_DATA_URL.format(lang=language, year=year) for language in self.config.languages for year in self.config.years]
|
140 |
dl_manager.download_config.num_proc = len(urls)
|
|
|
7 |
import datasets
|
8 |
|
9 |
|
10 |
+
_DESCRIPTION = """
|
11 |
+
A large-scale multilingual speech corpus for representation learning, semi-supervised learning and interpretation.
|
12 |
+
"""
|
13 |
+
|
14 |
+
_CITATION = """
|
15 |
+
@inproceedings{wang-etal-2021-voxpopuli,
|
16 |
+
title = "{V}ox{P}opuli: A Large-Scale Multilingual Speech Corpus for Representation Learning,
|
17 |
+
Semi-Supervised Learning and Interpretation",
|
18 |
+
author = "Wang, Changhan and
|
19 |
+
Riviere, Morgane and
|
20 |
+
Lee, Ann and
|
21 |
+
Wu, Anne and
|
22 |
+
Talnikar, Chaitanya and
|
23 |
+
Haziza, Daniel and
|
24 |
+
Williamson, Mary and
|
25 |
+
Pino, Juan and
|
26 |
+
Dupoux, Emmanuel",
|
27 |
+
booktitle = "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics
|
28 |
+
and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)",
|
29 |
+
month = aug,
|
30 |
+
year = "2021",
|
31 |
+
publisher = "Association for Computational Linguistics",
|
32 |
+
url = "https://aclanthology.org/2021.acl-long.80",
|
33 |
+
doi = "10.18653/v1/2021.acl-long.80",
|
34 |
+
pages = "993--1003",
|
35 |
+
}
|
36 |
+
"""
|
37 |
+
|
38 |
+
_HOMEPAGE = "https://github.com/facebookresearch/voxpopuli"
|
39 |
+
|
40 |
+
_LICENSE = "CC0, also see https://www.europarl.europa.eu/legal-notice/en/"
|
41 |
+
|
42 |
+
|
43 |
_LANGUAGES = sorted(
|
44 |
[
|
45 |
"en", "de", "fr", "es", "pl", "it", "ro", "hu", "cs", "nl", "fi", "hr",
|
|
|
48 |
)
|
49 |
_LANGUAGES_V2 = [f"{x}_v2" for x in _LANGUAGES]
|
50 |
|
51 |
+
_ASR_LANGUAGES = [
|
52 |
+
"en", "de", "fr", "es", "pl", "it", "ro", "hu", "cs", "nl", "fi", "hr",
|
53 |
+
"sk", "sl", "et", "lt"
|
54 |
+
]
|
55 |
+
_ASR_ACCENTED_LANGUAGES = [
|
56 |
+
"en_accented"
|
57 |
+
]
|
58 |
+
|
59 |
_YEARS = list(range(2009, 2020 + 1))
|
60 |
|
61 |
# unnecessary
|
|
|
63 |
"400k": _LANGUAGES,
|
64 |
"100k": _LANGUAGES,
|
65 |
"10k": _LANGUAGES,
|
66 |
+
"asr": _ASR_LANGUAGES, # + _ASR_ACCENTED_LANGUAGES
|
67 |
}
|
68 |
|
69 |
_CONFIG_TO_YEARS = {
|
|
|
84 |
|
85 |
_DATA_URL = _BASE_URL + "audios/{lang}_{year}.tar"
|
86 |
|
87 |
+
_ASR_DATA_URL = _BASE_URL + "audios/original_{year}.tar"
|
88 |
+
|
89 |
+
_UNLABELLED_META_URL = _BASE_URL + "annotations/unlabelled_v2.tsv.gz"
|
90 |
+
|
91 |
+
_ASR_META_URL = _BASE_URL + "annotations/asr/asr_{lang}.tsv.gz"
|
92 |
|
93 |
|
94 |
class VoxpopuliConfig(datasets.BuilderConfig):
|
|
|
102 |
"""
|
103 |
super().__init__(name=name, **kwargs)
|
104 |
name = name.split("_")[0]
|
105 |
+
self.languages = [name] if name in _LANGUAGES else _CONFIG_TO_LANGS[name]
|
106 |
self.years = _CONFIG_TO_YEARS[name]
|
107 |
|
108 |
|
109 |
class Voxpopuli(datasets.GeneratorBasedBuilder):
|
110 |
+
"""The VoxPopuli dataset."""
|
111 |
|
112 |
+
VERSION = datasets.Version("1.3.0") # not sure
|
113 |
BUILDER_CONFIGS = [
|
114 |
VoxpopuliConfig(
|
115 |
name=name,
|
116 |
+
version=datasets.Version("1.3.0"),
|
|
|
117 |
)
|
118 |
for name in _LANGUAGES + _LANGUAGES_V2 + ["10k", "100k", "400k"]
|
119 |
]
|
120 |
# DEFAULT_CONFIG_NAME = "400k"
|
121 |
+
DEFAULT_WRITER_BATCH_SIZE = 256 # SET THIS TO A LOWER VALUE IF IT USES TOO MUCH RAM SPACE
|
122 |
|
123 |
def _info(self):
|
124 |
try:
|
|
|
142 |
}
|
143 |
)
|
144 |
return datasets.DatasetInfo(
|
145 |
+
description=_DESCRIPTION,
|
146 |
features=features,
|
147 |
+
homepage=_HOMEPAGE,
|
148 |
+
license=_LICENSE,
|
149 |
+
citation=_CITATION,
|
150 |
)
|
151 |
|
152 |
+
def _read_metadata_unlabelled(self, metadata_path):
|
153 |
# from https://github.com/facebookresearch/voxpopuli/blob/main/voxpopuli/get_unlabelled_data.py#L34
|
154 |
def predicate(id_):
|
155 |
is_plenary = id_.find("PLENARY") > -1
|
|
|
177 |
|
178 |
return metadata
|
179 |
|
180 |
+
def _read_metadata_asr(self, metadata_paths):
|
181 |
+
pass
|
182 |
+
|
183 |
def _split_generators(self, dl_manager):
|
184 |
+
metadata_path = dl_manager.download_and_extract(_UNLABELLED_META_URL)
|
185 |
|
186 |
urls = [_DATA_URL.format(lang=language, year=year) for language in self.config.languages for year in self.config.years]
|
187 |
dl_manager.download_config.num_proc = len(urls)
|