holylovenia
commited on
Commit
•
37c56be
1
Parent(s):
70d50b2
Upload voxlingua.py with huggingface_hub
Browse files- voxlingua.py +204 -0
voxlingua.py
ADDED
@@ -0,0 +1,204 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
This new update refers to the this HF dataloader script
|
3 |
+
https://huggingface.co/datasets/csebuetnlp/xlsum/blob/main/xlsum.py
|
4 |
+
while conforming to SEACrowd schema
|
5 |
+
"""
|
6 |
+
|
7 |
+
import os
|
8 |
+
from pathlib import Path
|
9 |
+
from typing import Dict, List, Tuple
|
10 |
+
|
11 |
+
import datasets
|
12 |
+
|
13 |
+
from seacrowd.utils import schemas
|
14 |
+
from seacrowd.utils.configs import SEACrowdConfig
|
15 |
+
from seacrowd.utils.constants import TASK_TO_SCHEMA, Licenses, Tasks
|
16 |
+
|
17 |
+
_CITATION = """\
|
18 |
+
@inproceedings{valk2021slt,
|
19 |
+
title={{VoxLingua107}: a Dataset for Spoken Language Recognition},
|
20 |
+
author={J{\"o}rgen Valk and Tanel Alum{\"a}e},
|
21 |
+
booktitle={Proc. IEEE SLT Workshop},
|
22 |
+
year={2021},
|
23 |
+
}
|
24 |
+
"""
|
25 |
+
|
26 |
+
_LOCAL = False
|
27 |
+
_LANGUAGES = ["ceb", "ind", "jav", "khm", "lao", "zlm", "mya", "sun", "tha", "tgl", "vie", "war"] # We follow ISO639-3 language code (https://iso639-3.sil.org/code_tables/639/data)
|
28 |
+
|
29 |
+
_LANG_TO_DATASOURCE_LANG = {
|
30 |
+
"ceb": "ceb",
|
31 |
+
"ind": "id",
|
32 |
+
"jav": "jw",
|
33 |
+
"khm": "km",
|
34 |
+
"lao": "lo",
|
35 |
+
"zlm": "ms",
|
36 |
+
"mya": "my",
|
37 |
+
"sun": "su",
|
38 |
+
"tha": "th",
|
39 |
+
"tgl": "tl",
|
40 |
+
"vie": "vi",
|
41 |
+
"war": "war"}
|
42 |
+
|
43 |
+
_DATASETNAME = "voxlingua"
|
44 |
+
|
45 |
+
_DESCRIPTION = """\
|
46 |
+
VoxLingua107 is a comprehensive speech dataset designed for training spoken language identification models.
|
47 |
+
It comprises short speech segments sourced from YouTube videos, labeled based on the language indicated in the video
|
48 |
+
title and description. The dataset covers 107 languages and contains a total of 6628 hours of speech data,
|
49 |
+
averaging 62 hours per language. However, the actual amount of data per language varies significantly.
|
50 |
+
Additionally, there is a separate development set consisting of 1609 speech segments from 33 languages,
|
51 |
+
validated by at least two volunteers to ensure the accuracy of language representation.
|
52 |
+
"""
|
53 |
+
|
54 |
+
_HOMEPAGE = "https://bark.phon.ioc.ee/voxlingua107/"
|
55 |
+
|
56 |
+
_LICENSE = Licenses.CC_BY_4_0.value
|
57 |
+
|
58 |
+
_URLS = "https://bark.phon.ioc.ee/voxlingua107/{identifier}.zip"
|
59 |
+
|
60 |
+
_SUPPORTED_TASKS = [Tasks.SPEECH_LANGUAGE_IDENTIFICATION]
|
61 |
+
|
62 |
+
_SOURCE_VERSION = "1.0.0"
|
63 |
+
|
64 |
+
_SEACROWD_VERSION = "2024.06.20"
|
65 |
+
|
66 |
+
|
67 |
+
def construct_configs() -> List[SEACrowdConfig]:
|
68 |
+
"""
|
69 |
+
The function `construct_configs` constructs a list of SEACrowdConfig objects, and returns the list.
|
70 |
+
|
71 |
+
output:
|
72 |
+
a list of `SEACrowdConfig` objects.
|
73 |
+
"""
|
74 |
+
|
75 |
+
# set output var
|
76 |
+
config_list = []
|
77 |
+
|
78 |
+
# construct zipped arg for config instantiation
|
79 |
+
CONFIG_SUFFIXES_FOR_TASK = [TASK_TO_SCHEMA.get(task).lower() for task in _SUPPORTED_TASKS]
|
80 |
+
TASKS_AND_CONFIG_SUFFIX_PAIRS = list(zip(_SUPPORTED_TASKS, CONFIG_SUFFIXES_FOR_TASK))
|
81 |
+
|
82 |
+
# implement source schema
|
83 |
+
version, config_name_prefix = _SOURCE_VERSION, "source"
|
84 |
+
config_list += [
|
85 |
+
SEACrowdConfig(
|
86 |
+
name=f"{_DATASETNAME}_{config_name_prefix}",
|
87 |
+
version=datasets.Version(version),
|
88 |
+
description=f"{_DATASETNAME} {config_name_prefix} schema",
|
89 |
+
schema=f"{config_name_prefix}",
|
90 |
+
subset_id=f"{config_name_prefix}",
|
91 |
+
)
|
92 |
+
]
|
93 |
+
|
94 |
+
# implement SEACrowd schema
|
95 |
+
version, config_name_prefix = _SEACROWD_VERSION, "seacrowd"
|
96 |
+
for task_obj, config_name_suffix in TASKS_AND_CONFIG_SUFFIX_PAIRS:
|
97 |
+
config_list += [
|
98 |
+
SEACrowdConfig(
|
99 |
+
name=f"{_DATASETNAME}_{config_name_prefix}_{config_name_suffix}",
|
100 |
+
version=datasets.Version(version),
|
101 |
+
description=f"{_DATASETNAME} {config_name_prefix} schema for {task_obj.name}",
|
102 |
+
schema=f"{config_name_prefix}_{config_name_suffix}",
|
103 |
+
subset_id=f"{config_name_prefix}_{config_name_suffix}",
|
104 |
+
)
|
105 |
+
]
|
106 |
+
return config_list
|
107 |
+
|
108 |
+
|
109 |
+
class VoxLinguaDataset(datasets.GeneratorBasedBuilder):
|
110 |
+
"""Speech Lang ID on dataset VoxLingua."""
|
111 |
+
|
112 |
+
SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
|
113 |
+
SEACROWD_VERSION = datasets.Version(_SEACROWD_VERSION)
|
114 |
+
|
115 |
+
BUILDER_CONFIGS = construct_configs()
|
116 |
+
|
117 |
+
def _info(self) -> datasets.DatasetInfo:
|
118 |
+
if self.config.schema == "source":
|
119 |
+
# since the source only contains audio folder structure,
|
120 |
+
# we will define it using simplified ver of SEACrowd speech_features schema
|
121 |
+
features = datasets.Features({
|
122 |
+
"id": datasets.Value("string"),
|
123 |
+
"path": datasets.Value("string"),
|
124 |
+
"audio": datasets.Audio(sampling_rate=16_000),
|
125 |
+
"labels": datasets.ClassLabel(names=_LANGUAGES)})
|
126 |
+
|
127 |
+
elif self.config.schema == "seacrowd_speech":
|
128 |
+
features = schemas.speech_features(label_names=_LANGUAGES)
|
129 |
+
|
130 |
+
else:
|
131 |
+
raise ValueError(f"Unexpected self.config.schema of {self.config.schema}!")
|
132 |
+
|
133 |
+
return datasets.DatasetInfo(
|
134 |
+
description=_DESCRIPTION,
|
135 |
+
features=features,
|
136 |
+
homepage=_HOMEPAGE,
|
137 |
+
license=_LICENSE,
|
138 |
+
citation=_CITATION,
|
139 |
+
)
|
140 |
+
|
141 |
+
def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
|
142 |
+
# since this is a Speech LID, all languages must be downloaded in a single lists
|
143 |
+
# for train data, the identifier is a lang_code defined in `_LANG_TO_DATASOURCE_LANG`
|
144 |
+
train_url_list = [_URLS.format(identifier=_LANG_TO_DATASOURCE_LANG[lang_val]) for lang_val in _LANGUAGES]
|
145 |
+
train_data_dir = dl_manager.download_and_extract(train_url_list)
|
146 |
+
|
147 |
+
# for val data, the `dev.zip` doesn't contain any data indicated in _LANGUAGES
|
148 |
+
|
149 |
+
return [
|
150 |
+
datasets.SplitGenerator(
|
151 |
+
name=datasets.Split.TRAIN,
|
152 |
+
gen_kwargs={
|
153 |
+
"filepath": train_data_dir,
|
154 |
+
},
|
155 |
+
)
|
156 |
+
]
|
157 |
+
|
158 |
+
def _generate_examples(self, filepath: List[Path]) -> Tuple[int, Dict]:
|
159 |
+
|
160 |
+
# this is defined as -1 so that in the first loop it will have value of 0
|
161 |
+
example_idx = -1
|
162 |
+
|
163 |
+
for idx, child_path in enumerate(filepath):
|
164 |
+
# check for 2 things:
|
165 |
+
|
166 |
+
# 1. for every filepath list element (which contain 1 lang data), it will contain only 1 subdir and named its lang code in source
|
167 |
+
first_level_rel_dir = os.listdir(child_path)
|
168 |
+
expected_lang_label = _LANG_TO_DATASOURCE_LANG[_LANGUAGES[idx]]
|
169 |
+
assert first_level_rel_dir == [expected_lang_label], f"The structure of path is unexpected! Expected {[expected_lang_label]} got: {first_level_rel_dir}"
|
170 |
+
|
171 |
+
# 2. within the first_level_dir, all of them are file (no directory)
|
172 |
+
first_level_dir = os.path.join(child_path, first_level_rel_dir[0])
|
173 |
+
second_level_dir = os.listdir(first_level_dir)
|
174 |
+
assert not all(os.path.isdir(expected_file) for expected_file in second_level_dir), f"Found directory within folder {first_level_dir}!"
|
175 |
+
|
176 |
+
# extract sound data with format ".wav"
|
177 |
+
wav_files = [os.path.join(first_level_dir, file) for file in second_level_dir if file.endswith(".wav")]
|
178 |
+
|
179 |
+
if self.config.schema == "source":
|
180 |
+
for _fp in wav_files:
|
181 |
+
example_idx += 1
|
182 |
+
ex = {"id": example_idx, "path": _fp, "audio": _fp, "labels": _LANGUAGES.index(expected_lang_label)}
|
183 |
+
yield example_idx, ex
|
184 |
+
|
185 |
+
elif self.config.schema == "seacrowd_speech":
|
186 |
+
for _fp in wav_files:
|
187 |
+
example_idx += 1
|
188 |
+
# audio = {"path": file, "bytes": file.read()}
|
189 |
+
ex = {
|
190 |
+
"id": example_idx,
|
191 |
+
"path": _fp,
|
192 |
+
"audio": _fp,
|
193 |
+
"speaker_id": "",
|
194 |
+
"labels": _LANGUAGES.index(expected_lang_label),
|
195 |
+
"metadata": {
|
196 |
+
# unavailable, filled with default val
|
197 |
+
"speaker_age": -1,
|
198 |
+
"speaker_gender": "",
|
199 |
+
},
|
200 |
+
}
|
201 |
+
yield example_idx, ex
|
202 |
+
|
203 |
+
else:
|
204 |
+
raise ValueError(f"Invalid config schema of {self.config.schema}!")
|