Datasets:

Modalities:
Audio
Text
Formats:
parquet
Languages:
English
ArXiv:
Libraries:
Datasets
Dask
License:
brthor commited on
Commit
1191b34
1 Parent(s): c99d4bd

add readme

Browse files
Files changed (1) hide show
  1. README.md +177 -0
README.md CHANGED
@@ -1,3 +1,180 @@
1
  ---
2
  license: cc-by-4.0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: cc-by-4.0
3
+ task_categories:
4
+ - text-to-speech
5
+ language:
6
+ - en
7
+ size_categories:
8
+ - 10K<n<100K
9
+
10
+ configs:
11
+ - config_name: dev
12
+ data_files:
13
+ - split: dev.clean
14
+ path: "data/dev.clean/dev.clean*.parquet"
15
+
16
+ - config_name: clean
17
+ data_files:
18
+ - split: dev.clean
19
+ path: "data/dev.clean/dev.clean*.parquet"
20
+ - split: test.clean
21
+ path: "data/test.clean/test.clean*.parquet"
22
+ - split: train.clean.100
23
+ path: "data/train.clean.100/train.clean.100*.parquet"
24
+ - split: train.clean.360'
25
+ path: "data/train.clean.360/train.clean.360*.parquet"
26
+
27
+ - config_name: other
28
+ data_files:
29
+ - split: dev.other
30
+ path: "data/dev.other/dev.other*.parquet"
31
+ - split: test.other
32
+ path: "data/test.other/test.other*.parquet"
33
+ - split: train.other.500
34
+ path: "data/train.other.500/train.other.500*.parquet"
35
+
36
+ - config_name: all
37
+ data_files:
38
+ - split: dev.clean
39
+ path: "data/dev.clean/dev.clean*.parquet"
40
+ - split: dev.other
41
+ path: "data/dev.other/dev.other*.parquet"
42
+ - split: test.clean
43
+ path: "data/test.clean/test.clean*.parquet"
44
+ - split: test.other
45
+ path: "data/test.other/test.other*.parquet"
46
+ - split: train.clean.100
47
+ path: "data/train.clean.100/train.clean.100*.parquet"
48
+ - split: train.clean.360'
49
+ path: "data/train.clean.360/train.clean.360*.parquet"
50
+ - split: train.other.500
51
+ path: "data/train.other.500/train.other.500*.parquet"
52
  ---
53
+ # Dataset Card for LibriTTS-R
54
+
55
+ <!-- Provide a quick summary of the dataset. -->
56
+
57
+ LibriTTS-R [1] is a sound quality improved version of the LibriTTS corpus
58
+ (http://www.openslr.org/60/) which is a multi-speaker English corpus of approximately
59
+ 585 hours of read English speech at 24kHz sampling rate, published in 2019.
60
+
61
+ ## Overview
62
+
63
+ This is the LibriTTS-R dataset, adapted for the `datasets` library.
64
+
65
+ The dataset viewer does not seem to be functional, although most of the code here was adapted from the `librispeech_asr` dataset.
66
+
67
+ ## Usage
68
+
69
+ ### Splits
70
+
71
+ There are 7 splits (dots replace dashes from the original dataset, to comply with hf naming requirements):
72
+
73
+ - dev.clean
74
+ - dev.other
75
+ - test.clean
76
+ - test.other
77
+ - train.clean.100
78
+ - train.clean.360
79
+ - train.other.500
80
+
81
+ ### Configurations
82
+
83
+ There are 3 configurations, each which limits the splits the `load_dataset()` function will download.
84
+
85
+ The default configuration is "all".
86
+
87
+ - "dev": only the "dev.clean" split (good for testing the dataset quickly)
88
+ - "clean": contains only "clean" splits
89
+ - "other": contains only "other" splits
90
+ - "all": contains only "all" splits
91
+
92
+ ### Example
93
+
94
+ Loading the `clean` config with only the `train.clean.360` split.
95
+ ```
96
+ load_dataset("blabble-io/libritts_r", "clean", split="train.clean.100")
97
+ ```
98
+
99
+ Streaming is also supported.
100
+ ```
101
+ load_dataset("blabble-io/libritts_r", streaming=True)
102
+ ```
103
+
104
+ ### Columns
105
+
106
+ ```
107
+ {
108
+ "audio": datasets.Audio(sampling_rate=24_000),
109
+ "text_normalized": datasets.Value("string"),
110
+ "text_original": datasets.Value("string"),
111
+ "speaker_id": datasets.Value("string"),
112
+ "path": datasets.Value("string"),
113
+ "chapter_id": datasets.Value("string"),
114
+ "id": datasets.Value("string"),
115
+ }
116
+ ```
117
+
118
+ ### Example Row
119
+
120
+ ```
121
+ {
122
+ 'audio': {
123
+ 'path': '/home/user/.cache/huggingface/datasets/downloads/extracted/5551a515e85b9e463062524539c2e1cb52ba32affe128dffd866db0205248bdd/LibriTTS_R/dev-clean/3081/166546/3081_166546_000028_000002.wav',
124
+ 'array': ...,
125
+ 'sampling_rate': 24000
126
+ },
127
+ 'text_normalized': 'How quickly he disappeared!"',
128
+ 'text_original': 'How quickly he disappeared!"',
129
+ 'speaker_id': '3081',
130
+ 'path': '/home/user/.cache/huggingface/datasets/downloads/extracted/5551a515e85b9e463062524539c2e1cb52ba32affe128dffd866db0205248bdd/LibriTTS_R/dev-clean/3081/166546/3081_166546_000028_000002.wav',
131
+ 'chapter_id': '166546',
132
+ 'id': '3081_166546_000028_000002'
133
+ }
134
+ ```
135
+
136
+ ## Dataset Details
137
+
138
+ ### Dataset Description
139
+
140
+ - **License:** CC BY 4.0
141
+
142
+ ### Dataset Sources [optional]
143
+
144
+ <!-- Provide the basic links for the dataset. -->
145
+
146
+ - **Homepage:** https://www.openslr.org/141/
147
+ - **Paper:** https://arxiv.org/abs/2305.18802
148
+
149
+ ## Citation
150
+
151
+ <!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. -->
152
+
153
+ ```
154
+ @ARTICLE{Koizumi2023-hs,
155
+ title = "{LibriTTS-R}: A restored multi-speaker text-to-speech corpus",
156
+ author = "Koizumi, Yuma and Zen, Heiga and Karita, Shigeki and Ding,
157
+ Yifan and Yatabe, Kohei and Morioka, Nobuyuki and Bacchiani,
158
+ Michiel and Zhang, Yu and Han, Wei and Bapna, Ankur",
159
+ abstract = "This paper introduces a new speech dataset called
160
+ ``LibriTTS-R'' designed for text-to-speech (TTS) use. It is
161
+ derived by applying speech restoration to the LibriTTS
162
+ corpus, which consists of 585 hours of speech data at 24 kHz
163
+ sampling rate from 2,456 speakers and the corresponding
164
+ texts. The constituent samples of LibriTTS-R are identical
165
+ to those of LibriTTS, with only the sound quality improved.
166
+ Experimental results show that the LibriTTS-R ground-truth
167
+ samples showed significantly improved sound quality compared
168
+ to those in LibriTTS. In addition, neural end-to-end TTS
169
+ trained with LibriTTS-R achieved speech naturalness on par
170
+ with that of the ground-truth samples. The corpus is freely
171
+ available for download from
172
+ \textbackslashurl\{http://www.openslr.org/141/\}.",
173
+ month = may,
174
+ year = 2023,
175
+ copyright = "http://creativecommons.org/licenses/by-nc-nd/4.0/",
176
+ archivePrefix = "arXiv",
177
+ primaryClass = "eess.AS",
178
+ eprint = "2305.18802"
179
+ }
180
+ ```