bjelkenhed commited on
Commit
212b70c
1 Parent(s): e6f8c5a

initial commit

Browse files
Files changed (2) hide show
  1. README.md +19 -0
  2. babelbox_voice.py +127 -0
README.md ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ annotations_creators:
2
+ - crowdsourced
3
+ language:
4
+ - sv-SE
5
+ language_creators:
6
+ - crowdsourced
7
+ license:
8
+ - cc0-1.0
9
+ multilinguality:
10
+ - monolingual
11
+ pretty_name: Babelbox Voice
12
+ size_categories:
13
+ - 100K<n<1M
14
+ source_datasets: []
15
+ tags:
16
+ - NST
17
+ task_categories:
18
+ - automatic-speech-recognition
19
+ task_ids: []
babelbox_voice.py ADDED
@@ -0,0 +1,127 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """ Babelbox Voice Dataset"""
2
+
3
+ import csv
4
+ import os
5
+ import urllib
6
+
7
+ import datasets
8
+ import requests
9
+ import glob
10
+ import gzip
11
+ from typing import List
12
+ from datasets.utils.py_utils import size_str
13
+ logger = datasets.logging.get_logger(__name__)
14
+ import torchaudio
15
+ import torch
16
+ from tqdm import tqdm
17
+
18
+ _CITATION = """\
19
+ @inproceedings{babelboxvoice:2022,
20
+ author = {Andersson, O. and Bjelkenhed, M. and Bielsa, M. et al},
21
+ title = {Babelbox Voice: A Speech Corpus for training Whisper},
22
+ year = 2022
23
+ }
24
+ """
25
+
26
+ class BabelboxVoiceConfig(datasets.BuilderConfig):
27
+ """BuilderConfig for BabelboxVoice."""
28
+
29
+ def __init__(self, name, version, **kwargs):
30
+ self.name = name
31
+ self.version = version
32
+ self.features = kwargs.pop("features", None)
33
+ self.description = kwargs.pop("description", None)
34
+ self.archive_url = kwargs.pop("archive_url", None)
35
+ self.meta_url = kwargs.pop("meta_url", None)
36
+
37
+ description = (
38
+ f"Babelbox Voice speech to text dataset."
39
+ )
40
+ super(BabelboxVoiceConfig, self).__init__(
41
+ name=name,
42
+ version=version,
43
+ **kwargs,
44
+ )
45
+
46
+
47
+ class BabelboxVoice(datasets.GeneratorBasedBuilder):
48
+
49
+ VERSION = datasets.Version("1.0.0")
50
+
51
+ BUILDER_CONFIGS = [
52
+ BabelboxVoiceConfig(
53
+ name="nst",
54
+ version=VERSION,
55
+ description="This part of Pandora Voice includes data from National Library of Norway",
56
+ features=["path", "audio", "sentence"],
57
+ archive_url="/home/jovyan/shared-data/data/nst/archive",
58
+ meta_url="/home/jovyan/shared-data/data/nst/NST_se.csv"
59
+ )
60
+ ]
61
+
62
+ DEFAULT_CONFIG_NAME = "nst"
63
+
64
+ def _info(self):
65
+ description = (
66
+ "Babelbox Voice is an initiative to help teach machines how real people speak. "
67
+ )
68
+ if self.config.name == "nst":
69
+ features = datasets.Features(
70
+ {
71
+ "path": datasets.Value("string"),
72
+ "audio": datasets.features.Audio(sampling_rate=16_000),
73
+ "sentence": datasets.Value("string")
74
+ }
75
+ )
76
+
77
+ return datasets.DatasetInfo(
78
+ description=description,
79
+ features=features,
80
+ supervised_keys=None,
81
+ version=self.config.version
82
+ )
83
+
84
+ def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
85
+
86
+ archive_dir="/home/jovyan/shared-data/data/nst/archive"
87
+ archive_files = sorted(glob.glob(archive_dir + '/**.tar.gz'), reverse=False)
88
+
89
+ archive_paths = dl_manager.download(archive_files)
90
+
91
+ local_extracted_archive_paths = dl_manager.extract(archive_paths) if not dl_manager.is_streaming else {}
92
+
93
+ meta_url = self.config.meta_url
94
+
95
+ meta_path = dl_manager.download_and_extract(meta_url)
96
+
97
+ metadata = {}
98
+ with open(meta_path, encoding="utf-8") as f:
99
+ reader = csv.DictReader(f)
100
+ for row in tqdm(reader, desc="Reading metadata..."):
101
+ filename = row['filename_channel_1']
102
+ sentence = row['text']
103
+ metadata[filename] = sentence
104
+
105
+ return [
106
+ datasets.SplitGenerator(name=datasets.Split.TRAIN,
107
+ gen_kwargs={
108
+ "local_extracted_archive_paths": local_extracted_archive_paths,
109
+ "archives": [dl_manager.iter_archive(path) for path in archive_paths],
110
+ "metadata": metadata
111
+ })
112
+ ]
113
+
114
+ def _generate_examples(self, local_extracted_archive_paths, archives, metadata):
115
+
116
+ sampling_rate = 16000
117
+
118
+ for i, audio_archive in enumerate(archives):
119
+ for path, file in audio_archive:
120
+ if local_extracted_archive_paths == False:
121
+ path = os.path.join(local_extracted_archive_paths[i], path)
122
+ result = dict()
123
+ result["path"] = path
124
+ result["audio"] = {"path": path, "bytes": file.read()}
125
+ result["sentence"] = metadata[path]
126
+ yield path, result
127
+