ylacombe commited on
Commit
faf6604
1 Parent(s): 912e8f4

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +245 -2
README.md CHANGED
@@ -1,4 +1,25 @@
1
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2
  configs:
3
  - config_name: default
4
  data_files:
@@ -31,11 +52,233 @@ dataset_info:
31
  num_bytes: 249688889.909
32
  num_examples: 3807
33
  - name: test
34
- num_bytes: 245938961.0
35
  num_examples: 3769
36
  - name: train
37
- num_bytes: 707578913096.0
38
  num_examples: 10808037
39
  download_size: 705179367357
40
  dataset_size: 708074540946.909
41
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ pretty_name: English MLS
3
+ annotations_creators:
4
+ - expert-generated
5
+ language_creators:
6
+ - crowdsourced
7
+ - expert-generated
8
+ language:
9
+ - en
10
+ license:
11
+ - cc-by-4.0
12
+ multilinguality:
13
+ - multilingual
14
+ paperswithcode_id: multilingual-librispeech
15
+ size_categories:
16
+ - 1M<n<10M
17
+ source_datasets:
18
+ - original
19
+ task_categories:
20
+ - automatic-speech-recognition
21
+ - text-to-speech
22
+ - text-to-audio
23
  configs:
24
  - config_name: default
25
  data_files:
 
52
  num_bytes: 249688889.909
53
  num_examples: 3807
54
  - name: test
55
+ num_bytes: 245938961
56
  num_examples: 3769
57
  - name: train
58
+ num_bytes: 707578913096
59
  num_examples: 10808037
60
  download_size: 705179367357
61
  dataset_size: 708074540946.909
62
  ---
63
+
64
+
65
+
66
+ # Dataset Card for English MLS
67
+
68
+ ## Table of Contents
69
+ - [Dataset Description](#dataset-description)
70
+ - [Dataset Summary](#dataset-summary)
71
+ - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
72
+ - [Languages](#languages)
73
+ - [How to use](#how-to-use)
74
+ - [Dataset Structure](#dataset-structure)
75
+ - [Data Instances](#data-instances)
76
+ - [Data Fields](#data-fields)
77
+ - [Data Splits](#data-splits)
78
+ - [Dataset Creation](#dataset-creation)
79
+ - [Curation Rationale](#curation-rationale)
80
+ - [Source Data](#source-data)
81
+ - [Annotations](#annotations)
82
+ - [Personal and Sensitive Information](#personal-and-sensitive-information)
83
+ - [Considerations for Using the Data](#considerations-for-using-the-data)
84
+ - [Social Impact of Dataset](#social-impact-of-dataset)
85
+ - [Discussion of Biases](#discussion-of-biases)
86
+ - [Other Known Limitations](#other-known-limitations)
87
+ - [Additional Information](#additional-information)
88
+ - [Dataset Curators](#dataset-curators)
89
+ - [Licensing Information](#licensing-information)
90
+ - [Citation Information](#citation-information)
91
+ - [Contributions](#contributions)
92
+
93
+ ## Dataset Description
94
+
95
+ - **Homepage:** [MultiLingual LibriSpeech ASR corpus](http://www.openslr.org/94)
96
+ - **Repository:** [Needs More Information]
97
+ - **Paper:** [MLS: A Large-Scale Multilingual Dataset for Speech Research](https://arxiv.org/abs/2012.03411)
98
+ - **Leaderboard:** [🤗 Autoevaluate Leaderboard](https://huggingface.co/spaces/autoevaluate/leaderboards?dataset=facebook%2Fmultilingual_librispeech&only_verified=0&task=automatic-speech-recognition&config=-unspecified-&split=-unspecified-&metric=wer)
99
+
100
+ ### Dataset Summary
101
+
102
+ This is a streamable version of the **English version of the Multilingual LibriSpeech (MLS) dataset**.
103
+ The data archives were restructured from the original ones from [OpenSLR](http://www.openslr.org/94) to make it easier to stream.
104
+
105
+ MLS dataset is a large multilingual corpus suitable for speech research. The dataset is derived from read audiobooks from LibriVox and consists of
106
+ 8 languages - English, German, Dutch, Spanish, French, Italian, Portuguese, Polish. It includes about 44.5K hours of English and a total of about 6K hours for other languages.
107
+
108
+ This dataset card includes the 44.5K hours of English. Refers to this [dataset card](https://huggingface.co/datasets/facebook/multilingual_librispeech) for the other languages.
109
+
110
+ ### Supported Tasks and Leaderboards
111
+
112
+ - `automatic-speech-recognition`, `speaker-identification`: The dataset can be used to train a model for Automatic Speech Recognition (ASR). The model is presented with an audio file and asked to transcribe the audio file to written text. The most common evaluation metric is the word error rate (WER). The task has an active leaderboard which can be found at https://paperswithcode.com/dataset/multilingual-librispeech and ranks models based on their WER.
113
+ - `text-to-speech`, `text-to-audio`: The dataset can also be used to train a model for Text-To-Speech (TTS).
114
+
115
+ ### How to use
116
+
117
+ The `datasets` library allows you to load and pre-process your dataset in pure Python, at scale. The dataset can be downloaded and prepared in one call to your local drive by using the `load_dataset` function.
118
+
119
+ For example, to download the German config, simply specify the corresponding language config name (i.e., "german" for German):
120
+ ```python
121
+ from datasets import load_dataset
122
+
123
+ mls = load_dataset("parler-tts/mls_eng", split="train")
124
+ ```
125
+
126
+ Using the datasets library, you can also stream the dataset on-the-fly by adding a `streaming=True` argument to the `load_dataset` function call. Loading a dataset in streaming mode loads individual samples of the dataset at a time, rather than downloading the entire dataset to disk.
127
+ ```python
128
+ from datasets import load_dataset
129
+
130
+ mls = load_dataset("parler-tts/mls_eng", split="train", streaming=True)
131
+
132
+ print(next(iter(mls)))
133
+ ```
134
+
135
+ *Bonus*: create a [PyTorch dataloader](https://huggingface.co/docs/datasets/use_with_pytorch) directly with your own datasets (local/streamed).
136
+
137
+ Local:
138
+
139
+ ```python
140
+ from datasets import load_dataset
141
+ from torch.utils.data.sampler import BatchSampler, RandomSampler
142
+
143
+ mls = load_dataset("parler-tts/mls_eng", split="train")
144
+ batch_sampler = BatchSampler(RandomSampler(mls), batch_size=32, drop_last=False)
145
+ dataloader = DataLoader(mls, batch_sampler=batch_sampler)
146
+ ```
147
+
148
+ Streaming:
149
+
150
+ ```python
151
+ from datasets import load_dataset
152
+ from torch.utils.data import DataLoader
153
+
154
+ mls = load_dataset("parler-tts/mls_eng", split="train", streaming=True)
155
+ dataloader = DataLoader(mls, batch_size=32)
156
+ ```
157
+
158
+ To find out more about loading and preparing audio datasets, head over to [hf.co/blog/audio-datasets](https://huggingface.co/blog/audio-datasets).
159
+
160
+ ### Example scripts
161
+
162
+ Train your own CTC or Seq2Seq Automatic Speech Recognition models on MultiLingual Librispeech with `transformers` - [here](https://github.com/huggingface/transformers/tree/main/examples/pytorch/speech-recognition).
163
+
164
+ ## Dataset Structure
165
+
166
+ ### Data Fields
167
+
168
+ - file: A filename .flac format.
169
+
170
+ - audio: A dictionary containing the audio filename, the decoded audio array, and the sampling rate. Note that when accessing the audio column: `dataset[0]["audio"]` the audio file is automatically decoded and resampled to `dataset.features["audio"].sampling_rate`. Decoding and resampling of a large number of audio files might take a significant amount of time. Thus it is important to first query the sample index before the `"audio"` column, *i.e.* `dataset[0]["audio"]` should **always** be preferred over `dataset["audio"][0]`.
171
+
172
+ - text: the transcription of the audio file.
173
+
174
+ - id: unique id of the data sample.
175
+
176
+ - speaker_id: unique id of the speaker. The same speaker id can be found for multiple data samples.
177
+
178
+ - chapter_id: id of the audiobook chapter which includes the transcription.
179
+
180
+ ## Dataset Creation
181
+
182
+ ### Curation Rationale
183
+
184
+ [Needs More Information]
185
+
186
+ ### Source Data
187
+
188
+ #### Initial Data Collection and Normalization
189
+
190
+ [Needs More Information]
191
+
192
+ #### Who are the source language producers?
193
+
194
+ [Needs More Information]
195
+
196
+ ### Annotations
197
+
198
+ #### Annotation process
199
+
200
+ [Needs More Information]
201
+
202
+ #### Who are the annotators?
203
+
204
+ [Needs More Information]
205
+
206
+ ### Personal and Sensitive Information
207
+
208
+ The dataset consists of people who have donated their voice online. You agree to not attempt to determine the identity of speakers in this dataset.
209
+
210
+ ## Considerations for Using the Data
211
+
212
+ ### Social Impact of Dataset
213
+
214
+ [More Information Needed]
215
+
216
+ ### Discussion of Biases
217
+
218
+ [More Information Needed]
219
+
220
+ ### Other Known Limitations
221
+
222
+ [Needs More Information]
223
+
224
+ ## Additional Information
225
+
226
+ ### Dataset Curators
227
+
228
+ [Needs More Information]
229
+
230
+ ### Licensing Information
231
+
232
+ Public Domain, Creative Commons Attribution 4.0 International Public License ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/legalcode))
233
+
234
+ ### Citation Information
235
+
236
+ ```
237
+ @article{Pratap2020MLSAL,
238
+ title={MLS: A Large-Scale Multilingual Dataset for Speech Research},
239
+ author={Vineel Pratap and Qiantong Xu and Anuroop Sriram and Gabriel Synnaeve and Ronan Collobert},
240
+ journal={ArXiv},
241
+ year={2020},
242
+ volume={abs/2012.03411}
243
+ }
244
+ ```
245
+
246
+
247
+ ### Data Statistics
248
+
249
+ | Duration (h) | Train | Dev | Test |
250
+ |--------------|-----------|-------|-------|
251
+ | English | 44,659.74 | 15.75 | 15.55 |
252
+ | German | 1,966.51 | 14.28 | 14.29 |
253
+ | Dutch | 1,554.24 | 12.76 | 12.76 |
254
+ | French | 1,076.58 | 10.07 | 10.07 |
255
+ | Spanish | 917.68 | 9.99 | 10 |
256
+ | Italian | 247.38 | 5.18 | 5.27 |
257
+ | Portuguese | 160.96 | 3.64 | 3.74 |
258
+ | Polish | 103.65 | 2.08 | 2.14 |
259
+
260
+ | # Speakers | Train | | Dev | | Test | |
261
+ |------------|-------|------|-----|----|------|----|
262
+ | Gender | M | F | M | F | M | F |
263
+ | English | 2742 | 2748 | 21 | 21 | 21 | 21 |
264
+ | German | 81 | 95 | 15 | 15 | 15 | 15 |
265
+ | Dutch | 9 | 31 | 3 | 3 | 3 | 3 |
266
+ | French | 62 | 80 | 9 | 9 | 9 | 9 |
267
+ | Spanish | 36 | 50 | 10 | 10 | 10 | 10 |
268
+ | Italian | 22 | 43 | 5 | 5 | 5 | 5 |
269
+ | Portuguese | 26 | 16 | 5 | 5 | 5 | 5 |
270
+ | Polish | 6 | 5 | 2 | 2 | 2 | 2 |
271
+
272
+ | # Hours / Gender | Dev | | Test | |
273
+ |------------------|------|------|------|------|
274
+ | Gender | M | F | M | F |
275
+ | English | 7.76 | 7.99 | 7.62 | 7.93 |
276
+ | German | 7.06 | 7.22 | 7 | 7.29 |
277
+ | Dutch | 6.44 | 6.32 | 6.72 | 6.04 |
278
+ | French | 5.13 | 4.94 | 5.04 | 5.02 |
279
+ | Spanish | 4.91 | 5.08 | 4.78 | 5.23 |
280
+ | Italian | 2.5 | 2.68 | 2.38 | 2.9 |
281
+ | Portuguese | 1.84 | 1.81 | 1.83 | 1.9 |
282
+ | Polish | 1.12 | 0.95 | 1.09 | 1.05 |
283
+
284
+