darkproger commited on
Commit
5ab0cae
·
1 Parent(s): 6e1fb21

Create new file

Browse files
Files changed (1) hide show
  1. librispeech_asr.py +160 -0
librispeech_asr.py ADDED
@@ -0,0 +1,160 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2021 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+
16
+ # Lint as: python3
17
+ """Librispeech automatic speech recognition dataset."""
18
+
19
+
20
+ import os
21
+
22
+ import datasets
23
+ from datasets.tasks import AutomaticSpeechRecognition
24
+
25
+
26
+ _CITATION = """\
27
+ @inproceedings{panayotov2015librispeech,
28
+ title={Librispeech: an ASR corpus based on public domain audio books},
29
+ author={Panayotov, Vassil and Chen, Guoguo and Povey, Daniel and Khudanpur, Sanjeev},
30
+ booktitle={Acoustics, Speech and Signal Processing (ICASSP), 2015 IEEE International Conference on},
31
+ pages={5206--5210},
32
+ year={2015},
33
+ organization={IEEE}
34
+ }
35
+ """
36
+
37
+ _DESCRIPTION = """\
38
+ LibriSpeech is a corpus of approximately 1000 hours of read English speech with sampling rate of 16 kHz,
39
+ prepared by Vassil Panayotov with the assistance of Daniel Povey. The data is derived from read
40
+ audiobooks from the LibriVox project, and has been carefully segmented and aligned.87
41
+ """
42
+
43
+ _URL = "http://www.openslr.org/12"
44
+ _DL_URL = "http://www.openslr.org/resources/12/"
45
+
46
+
47
+ _DL_URLS = {
48
+ "dev-clean": _DL_URL + "dev-clean.tar.gz",
49
+ "test-clean": _DL_URL + "test-clean.tar.gz",
50
+ "train-clean-100": _DL_URL + "train-clean-100.tar.gz",
51
+ "train-clean-360": _DL_URL + "train-clean-360.tar.gz",
52
+ "test-other": _DL_URL + "test-other.tar.gz",
53
+ "dev-other": _DL_URL + "dev-other.tar.gz",
54
+ "train-other-500": _DL_URL + "train-other-500.tar.gz",
55
+ }
56
+
57
+
58
+ class LibrispeechASRConfig(datasets.BuilderConfig):
59
+ """BuilderConfig for LibriSpeechASR."""
60
+
61
+ def __init__(self, **kwargs):
62
+ """
63
+ Args:
64
+ data_dir: `string`, the path to the folder containing the files in the
65
+ downloaded .tar
66
+ citation: `string`, citation for the data set
67
+ url: `string`, url for information about the data set
68
+ **kwargs: keyword arguments forwarded to super.
69
+ """
70
+ super(LibrispeechASRConfig, self).__init__(version=datasets.Version("2.1.0", ""), **kwargs)
71
+
72
+
73
+ class LibrispeechASR(datasets.GeneratorBasedBuilder):
74
+ """Librispeech dataset."""
75
+
76
+ DEFAULT_WRITER_BATCH_SIZE = 256
77
+ DEFAULT_CONFIG_NAME = "all"
78
+ BUILDER_CONFIGS = [
79
+ LibrispeechASRConfig(name="dev-clean", description="'Clean' speech."),
80
+ LibrispeechASRConfig(name="test-clean", description="'Clean' speech."),
81
+ LibrispeechASRConfig(name="train-clean-100", description="'Clean' speech."),
82
+ LibrispeechASRConfig(name="train-clean-360", description="'Clean' speech."),
83
+ LibrispeechASRConfig(name="dev-other", description="'Other', more challenging, speech."),
84
+ LibrispeechASRConfig(name="test-other", description="'Other', more challenging, speech."),
85
+ LibrispeechASRConfig(name="train-other-500", description="'Other', more challenging, speech."),
86
+ ]
87
+
88
+ def _info(self):
89
+ return datasets.DatasetInfo(
90
+ description=_DESCRIPTION,
91
+ features=datasets.Features(
92
+ {
93
+ "file": datasets.Value("string"),
94
+ "audio": datasets.Audio(sampling_rate=16_000),
95
+ "text": datasets.Value("string"),
96
+ "speaker_id": datasets.Value("int64"),
97
+ "chapter_id": datasets.Value("int64"),
98
+ "id": datasets.Value("string"),
99
+ }
100
+ ),
101
+ supervised_keys=("file", "text"),
102
+ homepage=_URL,
103
+ citation=_CITATION,
104
+ task_templates=[AutomaticSpeechRecognition(audio_column="audio", transcription_column="text")],
105
+ )
106
+
107
+ def _split_generators(self, dl_manager):
108
+ archive_path = dl_manager.download(_DL_URLS[self.config.name])
109
+ # (Optional) In non-streaming mode, we can extract the archive locally to have actual local audio files:
110
+ local_extracted_archive = dl_manager.extract(archive_path) if not dl_manager.is_streaming else {}
111
+
112
+ splits = [
113
+ datasets.SplitGenerator(
114
+ name=self.config.name,
115
+ gen_kwargs={
116
+ "local_extracted_archive": local_extracted_archive.get(self.config.name),
117
+ "files": dl_manager.iter_archive(self.config.name),
118
+ },
119
+ ),
120
+ ]
121
+
122
+ return splits
123
+
124
+ def _generate_examples(self, files, local_extracted_archive):
125
+ """Generate examples from a LibriSpeech archive_path."""
126
+ key = 0
127
+ audio_data = {}
128
+ transcripts = []
129
+ for path, f in files:
130
+ if path.endswith(".flac"):
131
+ id_ = path.split("/")[-1][: -len(".flac")]
132
+ audio_data[id_] = f.read()
133
+ elif path.endswith(".trans.txt"):
134
+ for line in f:
135
+ if line:
136
+ line = line.decode("utf-8").strip()
137
+ id_, transcript = line.split(" ", 1)
138
+ audio_file = f"{id_}.flac"
139
+ speaker_id, chapter_id = [int(el) for el in id_.split("-")[:2]]
140
+ audio_file = (
141
+ os.path.join(local_extracted_archive, audio_file)
142
+ if local_extracted_archive
143
+ else audio_file
144
+ )
145
+ transcripts.append(
146
+ {
147
+ "id": id_,
148
+ "speaker_id": speaker_id,
149
+ "chapter_id": chapter_id,
150
+ "file": audio_file,
151
+ "text": transcript,
152
+ }
153
+ )
154
+ if audio_data and len(audio_data) == len(transcripts):
155
+ for transcript in transcripts:
156
+ audio = {"path": transcript["file"], "bytes": audio_data[transcript["id"]]}
157
+ yield key, {"audio": audio, **transcript}
158
+ key += 1
159
+ audio_data = {}
160
+ transcripts = []