kaushal98b commited on
Commit
2bdfded
1 Parent(s): 6d9e83d

Add code files

Browse files
Files changed (3) hide show
  1. kathbath.py +153 -0
  2. languages.py +14 -0
  3. release_stats.py +22 -0
kathbath.py ADDED
@@ -0,0 +1,153 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """ Kathbath Dataset"""
2
+
3
+ import csv
4
+ import os
5
+
6
+ import datasets
7
+ from datasets.utils.py_utils import size_str
8
+
9
+ from .languages import LANGUAGES
10
+ from .release_stats import STATS
11
+
12
+ _CITATION = """\
13
+ @misc{https://doi.org/10.48550/arxiv.2208.11761,
14
+ doi = {10.48550/ARXIV.2208.11761},
15
+ url = {https://arxiv.org/abs/2208.11761},
16
+ author = {Javed, Tahir and Bhogale, Kaushal Santosh and Raman, Abhigyan and Kunchukuttan, Anoop and Kumar, Pratyush and Khapra, Mitesh M.},
17
+ title = {IndicSUPERB: A Speech Processing Universal Performance Benchmark for Indian languages},
18
+ publisher = {arXiv},
19
+ year = {2022},
20
+ copyright = {arXiv.org perpetual, non-exclusive license}
21
+ }
22
+ """
23
+
24
+ _HOMEPAGE = "https://ai4bharat.iitm.ac.in/indic-superb/"
25
+
26
+ _LICENSE = "https://creativecommons.org/publicdomain/zero/1.0/"
27
+
28
+ _DATA_URL = "https://huggingface.co/datasets/ai4bharat/kathbath/resolve/main/data"
29
+
30
+
31
+ class KathbathConfig(datasets.BuilderConfig):
32
+ """BuilderConfig for Kathbath."""
33
+
34
+ def __init__(self, name, version, **kwargs):
35
+ self.language = kwargs.pop("language", None)
36
+ self.release_date = kwargs.pop("release_date", None)
37
+ self.num_clips = kwargs.pop("num_clips", None)
38
+ self.num_speakers = kwargs.pop("num_speakers", None)
39
+ self.total_hr = kwargs.pop("total_hr", None)
40
+ self.size_bytes = kwargs.pop("size_bytes", None)
41
+ self.size_human = size_str(self.size_bytes)
42
+ description = (
43
+ f"Kathbath speech to text dataset in {self.language} released on {self.release_date}. "
44
+ f"The dataset comprises {self.total_hr} hours of transcribed speech data"
45
+ )
46
+ super(KathbathConfig, self).__init__(
47
+ name=name,
48
+ version=datasets.Version(version),
49
+ description=description,
50
+ **kwargs,
51
+ )
52
+
53
+
54
+ class Kathbath(datasets.GeneratorBasedBuilder):
55
+ DEFAULT_CONFIG_NAME = "_all_"
56
+
57
+ BUILDER_CONFIGS = [
58
+ KathbathConfig(
59
+ name=lang,
60
+ version=STATS["version"],
61
+ language=LANGUAGES[lang],
62
+ release_date=STATS["date"],
63
+ # num_clips=lang_stats["clips"],
64
+ # num_speakers=lang_stats["users"],
65
+ total_hr=float(lang_stats["totalHrs"]) if lang_stats["totalHrs"] else None,
66
+ size_bytes=int(lang_stats["size"]) if lang_stats["size"] else None,
67
+ )
68
+ for lang, lang_stats in STATS["locales"].items()
69
+ ]
70
+
71
+ def _info(self):
72
+ total_languages = len(STATS["locales"])
73
+ total_hours = self.config.total_hr
74
+ description = (
75
+ "LibriVox-Indonesia is a speech dataset generated from LibriVox with only languages from Indonesia."
76
+ f"The dataset currently consists of {total_hours} hours of speech "
77
+ f"in {total_languages} languages, but more voices and languages are always added."
78
+ )
79
+ features = datasets.Features(
80
+ {
81
+ "path": datasets.Value("string"),
82
+ "language": datasets.Value("string"),
83
+ "speaker": datasets.Value("string"),
84
+ "sentence": datasets.Value("string"),
85
+ "audio": datasets.features.Audio(sampling_rate=16000)
86
+ }
87
+ )
88
+
89
+ return datasets.DatasetInfo(
90
+ description=description,
91
+ features=features,
92
+ supervised_keys=None,
93
+ homepage=_HOMEPAGE,
94
+ license=_LICENSE,
95
+ citation=_CITATION,
96
+ version=self.config.version,
97
+ )
98
+
99
+ def _split_generators(self, dl_manager):
100
+ """Returns SplitGenerators."""
101
+ dl_manager.download_config.ignore_url_params = True
102
+ audio_path = {}
103
+ local_extracted_archive = {}
104
+ metadata_path = {}
105
+ split_type = {"train": datasets.Split.TRAIN, "valid": datasets.Split.VALIDATION, "test_unknown": datasets.Split.TEST, "test_known": datasets.Split.TEST}
106
+ for split in split_type:
107
+ audio_path[split] = dl_manager.download(f"{_DATA_URL}/audio_{split}.tar")
108
+ local_extracted_archive[split] = dl_manager.extract(audio_path[split]) if not dl_manager.is_streaming else None
109
+ metadata_path[split] = dl_manager.download(f"{_DATA_URL}/metadata_{split}.tsv")
110
+ path_to_clips = "kb_data_clean_m4a"
111
+
112
+ return [
113
+ datasets.SplitGenerator(
114
+ name=split_type[split],
115
+ gen_kwargs={
116
+ "local_extracted_archive": local_extracted_archive[split],
117
+ "audio_files": dl_manager.iter_archive(audio_path[split]),
118
+ "metadata_path": metadata_path[split],
119
+ "path_to_clips": path_to_clips,
120
+ },
121
+ ) for split in split_type
122
+ ]
123
+
124
+ def _generate_examples(
125
+ self,
126
+ local_extracted_archive,
127
+ audio_files,
128
+ metadata_path,
129
+ path_to_clips,
130
+ ):
131
+ """Yields examples."""
132
+ data_fields = list(self._info().features.keys())
133
+ metadata = {}
134
+ with open(metadata_path, "r", encoding="utf-8") as f:
135
+ reader = csv.DictReader(f, delimiter="\t")
136
+ for row in reader:
137
+ if self.config.name == "_all_" or self.config.name == row["language"]:
138
+ row["path"] = os.path.join(path_to_clips, row["path"])
139
+ # if data is incomplete, fill with empty values
140
+ for field in data_fields:
141
+ if field not in row:
142
+ row[field] = ""
143
+ metadata[row["path"]] = row
144
+ id_ = 0
145
+ for path, f in audio_files:
146
+ if path in metadata:
147
+ result = dict(metadata[path])
148
+ # set the audio feature and the path to the extracted file
149
+ path = os.path.join(local_extracted_archive, path) if local_extracted_archive else path
150
+ result["audio"] = {"path": path, "bytes": f.read()}
151
+ result["path"] = path
152
+ yield id_, result
153
+ id_ += 1
languages.py ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ LANGUAGES = {
2
+ "bn": "Bengali",
3
+ "gu": "Gujarati",
4
+ "hi": "Hindi",
5
+ "kn": "Kannada",
6
+ "ml": "Malayalam",
7
+ "mr": "Marathi",
8
+ "or": "Odia",
9
+ "pa": "Punjabi",
10
+ "sa": "Sanskrit",
11
+ "ta": "Tamil",
12
+ "te": "Telugu",
13
+ "ur": "Urdu"
14
+ }
release_stats.py ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ STATS = {
2
+ "name": "Kathbath",
3
+ "version": "1.0.0",
4
+ "date": "2022-08-24",
5
+ "locales": {
6
+ "bn": {'totalHrs': 115.8},
7
+ "gu": {'totalHrs': 129.3},
8
+ "hi": {'totalHrs': 150.2},
9
+ "kn": {'totalHrs': 65.8},
10
+ "ml": {'totalHrs': 147.3},
11
+ "mr": {'totalHrs': 185.2},
12
+ "or": {'totalHrs': 111.6},
13
+ "pa": {'totalHrs': 136.9},
14
+ "sa": {'totalHrs': 115.5},
15
+ "ta": {'totalHrs': 185.5},
16
+ "te": {'totalHrs': 154.9},
17
+ "ur": {'totalHrs': 86.7},
18
+ "_all_": {'totalHrs': 1},
19
+ },
20
+ 'totalDuration': 1, 'totalHrs': 1
21
+ }
22
+