Delete dataset.py
Browse files- dataset.py +0 -191
dataset.py
DELETED
@@ -1,191 +0,0 @@
|
|
1 |
-
# coding=utf-8
|
2 |
-
"""
|
3 |
-
new_dataset_stt_audio dataset.
|
4 |
-
|
5 |
-
This dataset consists of audio files stored in tar archives and transcript files in TSV format.
|
6 |
-
The dataset structure is as follows:
|
7 |
-
|
8 |
-
new_dataset_stt_audio/
|
9 |
-
├── audio/
|
10 |
-
│ └── uz/
|
11 |
-
│ ├── train/
|
12 |
-
│ │ └── train.tar
|
13 |
-
│ ├── validation/
|
14 |
-
│ │ └── validation.tar
|
15 |
-
│ └── test/
|
16 |
-
│ └── test.tar
|
17 |
-
└── transcript/
|
18 |
-
└── uz/
|
19 |
-
├── train/
|
20 |
-
│ └── train.tsv
|
21 |
-
├── validation/
|
22 |
-
│ └── validation.tsv
|
23 |
-
└── test/
|
24 |
-
└── test.tsv
|
25 |
-
|
26 |
-
Each transcript TSV file has columns:
|
27 |
-
id, path, sentence, duration, age, gender, accents, locale.
|
28 |
-
|
29 |
-
The audio field is loaded using a tar URI, allowing streaming from the tar archive.
|
30 |
-
"""
|
31 |
-
|
32 |
-
import csv
|
33 |
-
import os
|
34 |
-
import tarfile
|
35 |
-
from typing import Iterator, Tuple
|
36 |
-
|
37 |
-
import datasets
|
38 |
-
|
39 |
-
_CITATION = """\
|
40 |
-
@misc{yourcitation2023,
|
41 |
-
title={Your Dataset Title},
|
42 |
-
author={Your Name},
|
43 |
-
year={2023},
|
44 |
-
url={https://huggingface.co/datasets/Elyordev/new_dataset_stt_audio}
|
45 |
-
}
|
46 |
-
"""
|
47 |
-
|
48 |
-
_DESCRIPTION = """\
|
49 |
-
This dataset consists of audio files and corresponding transcripts for speech-to-text tasks.
|
50 |
-
The audio files are stored in tar archives under the audio/uz folder for each split (train, validation, test),
|
51 |
-
and the transcripts are stored as TSV files under transcript/uz for each split.
|
52 |
-
The transcript TSV files have the following columns:
|
53 |
-
id, path, sentence, duration, age, gender, accents, locale.
|
54 |
-
The audio is loaded using a tar URI to enable streaming.
|
55 |
-
"""
|
56 |
-
|
57 |
-
_HOMEPAGE = "https://huggingface.co/datasets/Elyordev/new_dataset_stt_audio"
|
58 |
-
_LICENSE = "MIT"
|
59 |
-
|
60 |
-
class NewDatasetSTTAudioConfig(datasets.BuilderConfig):
|
61 |
-
"""Builder config for new_dataset_stt_audio."""
|
62 |
-
def __init__(self, language="uz", **kwargs):
|
63 |
-
super(NewDatasetSTTAudioConfig, self).__init__(**kwargs)
|
64 |
-
self.language = language
|
65 |
-
|
66 |
-
class NewDatasetSTTAudio(datasets.GeneratorBasedBuilder):
|
67 |
-
"""New Dataset STT Audio builder."""
|
68 |
-
VERSION = datasets.Version("1.0.0")
|
69 |
-
BUILDER_CONFIGS = [
|
70 |
-
NewDatasetSTTAudioConfig(
|
71 |
-
name="default",
|
72 |
-
version=VERSION,
|
73 |
-
description="STT dataset with audio tar archives and transcript TSV files for Uzbek language",
|
74 |
-
language="uz",
|
75 |
-
),
|
76 |
-
]
|
77 |
-
|
78 |
-
def _info(self):
|
79 |
-
features = datasets.Features({
|
80 |
-
"id": datasets.Value("string"),
|
81 |
-
"path": datasets.Value("string"),
|
82 |
-
"sentence": datasets.Value("string"),
|
83 |
-
"duration": datasets.Value("float"),
|
84 |
-
"age": datasets.Value("string"),
|
85 |
-
"gender": datasets.Value("string"),
|
86 |
-
"accents": datasets.Value("string"),
|
87 |
-
"locale": datasets.Value("string"),
|
88 |
-
"audio": datasets.Audio(sampling_rate=16000),
|
89 |
-
})
|
90 |
-
return datasets.DatasetInfo(
|
91 |
-
description=_DESCRIPTION,
|
92 |
-
features=features,
|
93 |
-
supervised_keys=None,
|
94 |
-
homepage=_HOMEPAGE,
|
95 |
-
license=_LICENSE,
|
96 |
-
citation=_CITATION,
|
97 |
-
)
|
98 |
-
|
99 |
-
def _split_generators(self, dl_manager):
|
100 |
-
"""
|
101 |
-
Returns SplitGenerators.
|
102 |
-
Expects the dataset to be provided manually via the repository.
|
103 |
-
The manual_dir should contain the following structure:
|
104 |
-
|
105 |
-
new_dataset_stt_audio/
|
106 |
-
audio/uz/{train, validation, test}/*.tar
|
107 |
-
transcript/uz/{train, validation, test}/*.tsv
|
108 |
-
"""
|
109 |
-
manual_dir = dl_manager.manual_dir if dl_manager.manual_dir is not None else ""
|
110 |
-
language = self.config.language
|
111 |
-
|
112 |
-
splits = {
|
113 |
-
"train": {
|
114 |
-
"transcript": os.path.join(manual_dir, "transcript", language, "train", "train.tsv"),
|
115 |
-
"audio": os.path.join(manual_dir, "audio", language, "train", "train.tar")
|
116 |
-
},
|
117 |
-
"validation": {
|
118 |
-
"transcript": os.path.join(manual_dir, "transcript", language, "validation", "validation.tsv"),
|
119 |
-
"audio": os.path.join(manual_dir, "audio", language, "validation", "validation.tar")
|
120 |
-
},
|
121 |
-
"test": {
|
122 |
-
"transcript": os.path.join(manual_dir, "transcript", language, "test", "test.tsv"),
|
123 |
-
"audio": os.path.join(manual_dir, "audio", language, "test", "test.tar")
|
124 |
-
}
|
125 |
-
}
|
126 |
-
|
127 |
-
return [
|
128 |
-
datasets.SplitGenerator(
|
129 |
-
name=datasets.Split.TRAIN,
|
130 |
-
gen_kwargs={
|
131 |
-
"transcript_path": splits["train"]["transcript"],
|
132 |
-
"audio_tar_path": splits["train"]["audio"],
|
133 |
-
},
|
134 |
-
),
|
135 |
-
datasets.SplitGenerator(
|
136 |
-
name=datasets.Split.VALIDATION,
|
137 |
-
gen_kwargs={
|
138 |
-
"transcript_path": splits["validation"]["transcript"],
|
139 |
-
"audio_tar_path": splits["validation"]["audio"],
|
140 |
-
},
|
141 |
-
),
|
142 |
-
datasets.SplitGenerator(
|
143 |
-
name=datasets.Split.TEST,
|
144 |
-
gen_kwargs={
|
145 |
-
"transcript_path": splits["test"]["transcript"],
|
146 |
-
"audio_tar_path": splits["test"]["audio"],
|
147 |
-
},
|
148 |
-
),
|
149 |
-
]
|
150 |
-
|
151 |
-
def _generate_examples(self, transcript_path: str, audio_tar_path: str) -> Iterator[Tuple[str, dict]]:
|
152 |
-
"""
|
153 |
-
Yields examples as (key, example) tuples.
|
154 |
-
|
155 |
-
Args:
|
156 |
-
transcript_path (str): Path to the transcript TSV file.
|
157 |
-
audio_tar_path (str): Path to the audio tar archive.
|
158 |
-
"""
|
159 |
-
# 1. Read the transcript TSV file into a dictionary mapping file name to metadata.
|
160 |
-
metadata_map = {}
|
161 |
-
with open(transcript_path, encoding="utf-8") as f:
|
162 |
-
reader = csv.DictReader(f, delimiter="\t")
|
163 |
-
for row in reader:
|
164 |
-
file_name = row["path"].strip()
|
165 |
-
if not file_name.endswith(".mp3"):
|
166 |
-
file_name += ".mp3"
|
167 |
-
metadata_map[file_name] = row
|
168 |
-
|
169 |
-
# 2. Create a base audio URI for streaming from the tar archive.
|
170 |
-
base_audio_uri = f"tar://{audio_tar_path}#"
|
171 |
-
|
172 |
-
# 3. Open the tar archive and iterate through its members.
|
173 |
-
id_ = 0
|
174 |
-
with tarfile.open(audio_tar_path, "r") as tar:
|
175 |
-
for member in tar.getmembers():
|
176 |
-
file_name = os.path.basename(member.name)
|
177 |
-
if file_name in metadata_map:
|
178 |
-
row = metadata_map[file_name]
|
179 |
-
audio_uri = base_audio_uri + file_name
|
180 |
-
yield str(id_), {
|
181 |
-
"id": row["id"],
|
182 |
-
"path": row["path"],
|
183 |
-
"sentence": row["sentence"],
|
184 |
-
"duration": float(row.get("duration", 0.0)),
|
185 |
-
"age": row.get("age", ""),
|
186 |
-
"gender": row.get("gender", ""),
|
187 |
-
"accents": row.get("accents", ""),
|
188 |
-
"locale": row.get("locale", ""),
|
189 |
-
"audio": audio_uri,
|
190 |
-
}
|
191 |
-
id_ += 1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|