Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,166 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license:
|
3 |
+
- cc-by-4.0
|
4 |
+
size_categories:
|
5 |
+
ar:
|
6 |
+
- n==1k
|
7 |
+
task_categories:
|
8 |
+
- automatic-speech-recognition
|
9 |
+
task_ids: []
|
10 |
+
pretty_name: MASC dataset
|
11 |
+
extra_gated_prompt: >-
|
12 |
+
By clicking on “Access repository” below, you also agree to not attempt to
|
13 |
+
determine the identity of speakers in the MASC dataset.
|
14 |
+
language:
|
15 |
+
- ar
|
16 |
+
---
|
17 |
+
|
18 |
+
# Dataset Card for Common Voice Corpus 11.0
|
19 |
+
|
20 |
+
## Table of Contents
|
21 |
+
- [Dataset Description](#dataset-description)
|
22 |
+
- [Dataset Summary](#dataset-summary)
|
23 |
+
- [Languages](#languages)
|
24 |
+
- [How to use](#how-to-use)
|
25 |
+
- [Dataset Structure](#dataset-structure)
|
26 |
+
- [Data Instances](#data-instances)
|
27 |
+
- [Data Fields](#data-fields)
|
28 |
+
- [Data Splits](#data-splits)
|
29 |
+
- [Additional Information](#additional-information)
|
30 |
+
- [Citation Information](#citation-information)
|
31 |
+
|
32 |
+
## Dataset Description
|
33 |
+
|
34 |
+
- **Homepage:** https://ieee-dataport.org/open-access/masc-massive-arabic-speech-corpus
|
35 |
+
- **Paper:** https://ieeexplore.ieee.org/document/10022652
|
36 |
+
|
37 |
+
### Dataset Summary
|
38 |
+
|
39 |
+
MASC is a dataset that contains 1,000 hours of speech sampled at 16 kHz and crawled from over 700 YouTube channels.
|
40 |
+
The dataset is multi-regional, multi-genre, and multi-dialect intended to advance the research and development of Arabic speech technology with a special emphasis on Arabic speech recognition.
|
41 |
+
|
42 |
+
|
43 |
+
### Supported Tasks
|
44 |
+
|
45 |
+
- Automatics Speach Recognition
|
46 |
+
|
47 |
+
### Languages
|
48 |
+
|
49 |
+
```
|
50 |
+
Arabic
|
51 |
+
```
|
52 |
+
|
53 |
+
## How to use
|
54 |
+
|
55 |
+
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.
|
56 |
+
|
57 |
+
```python
|
58 |
+
from datasets import load_dataset
|
59 |
+
|
60 |
+
masc = load_dataset("pain/MASC", split="train")
|
61 |
+
```
|
62 |
+
|
63 |
+
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.
|
64 |
+
```python
|
65 |
+
from datasets import load_dataset
|
66 |
+
|
67 |
+
masc = load_dataset("pain/MASC", split="train", streaming=True)
|
68 |
+
|
69 |
+
print(next(iter(masc)))
|
70 |
+
```
|
71 |
+
|
72 |
+
*Bonus*: create a [PyTorch dataloader](https://huggingface.co/docs/datasets/use_with_pytorch) directly with your own datasets (local/streamed).
|
73 |
+
|
74 |
+
### Local
|
75 |
+
|
76 |
+
```python
|
77 |
+
from datasets import load_dataset
|
78 |
+
from torch.utils.data.sampler import BatchSampler, RandomSampler
|
79 |
+
|
80 |
+
masc = load_dataset("pain/MASC", split="train")
|
81 |
+
batch_sampler = BatchSampler(RandomSampler(masc), batch_size=32, drop_last=False)
|
82 |
+
dataloader = DataLoader(masc, batch_sampler=batch_sampler)
|
83 |
+
```
|
84 |
+
|
85 |
+
### Streaming
|
86 |
+
|
87 |
+
```python
|
88 |
+
from datasets import load_dataset
|
89 |
+
from torch.utils.data import DataLoader
|
90 |
+
|
91 |
+
masc = load_dataset("pain/MASC", split="train")
|
92 |
+
dataloader = DataLoader(masc, batch_size=32)
|
93 |
+
```
|
94 |
+
|
95 |
+
To find out more about loading and preparing audio datasets, head over to [hf.co/blog/audio-datasets](https://huggingface.co/blog/audio-datasets).
|
96 |
+
|
97 |
+
### Example scripts
|
98 |
+
|
99 |
+
Train your own CTC or Seq2Seq Automatic Speech Recognition models on MASC with `transformers` - [here](https://github.com/huggingface/transformers/tree/main/examples/pytorch/speech-recognition).
|
100 |
+
|
101 |
+
## Dataset Structure
|
102 |
+
|
103 |
+
### Data Instances
|
104 |
+
|
105 |
+
A typical data point comprises the `path` to the audio file and its `sentence`.
|
106 |
+
|
107 |
+
```python
|
108 |
+
{'video_id': 'OGqz9G-JO0E', 'start': 770.6, 'end': 781.835, 'duration': 11.24,
|
109 |
+
'text': 'اللهم من ارادنا وبلادنا وبلاد المسلمين بسوء اللهم فاشغله في نفسه ورد كيده في نحره واجعل تدبيره تدميره يا رب العالمين',
|
110 |
+
'type': 'c', 'file_path': '87edeceb-5349-4210-89ad-8c3e91e54062_OGqz9G-JO0E.wav',
|
111 |
+
'audio': {'path': None,
|
112 |
+
'array': array([
|
113 |
+
0.05938721,
|
114 |
+
0.0539856,
|
115 |
+
0.03460693, ...,
|
116 |
+
0.00393677,
|
117 |
+
0.01745605,
|
118 |
+
0.03045654
|
119 |
+
]), 'sampling_rate': 16000
|
120 |
+
}
|
121 |
+
}
|
122 |
+
```
|
123 |
+
|
124 |
+
### Data Fields
|
125 |
+
|
126 |
+
`video_id` (`string`): An id for the video that the voice has been created from
|
127 |
+
|
128 |
+
`start` (`float64`): The start of the audio's chunk
|
129 |
+
|
130 |
+
`end` (`float64`): The end of the audio's chunk
|
131 |
+
|
132 |
+
`duration` (`float64`): The duration of the chunk
|
133 |
+
|
134 |
+
`text` (`string`): The text of the chunk
|
135 |
+
|
136 |
+
`audio` (`dict`): A dictionary containing the path to the downloaded audio file, 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]`.
|
137 |
+
|
138 |
+
`type` (`string`): It refers to the data set type, either clean or noisy where "c: clean and n: noisy"
|
139 |
+
|
140 |
+
'file_path' (`string`): A path for the audio chunk
|
141 |
+
|
142 |
+
"audio" ("audio"): Audio for the chunk
|
143 |
+
|
144 |
+
### Data Splits
|
145 |
+
|
146 |
+
The speech material has been subdivided into portions for train, dev, test.
|
147 |
+
|
148 |
+
The dataset splits has clean and noisy data that can be determined by type field.
|
149 |
+
|
150 |
+
|
151 |
+
|
152 |
+
|
153 |
+
### Citation Information
|
154 |
+
|
155 |
+
```
|
156 |
+
@INPROCEEDINGS{10022652,
|
157 |
+
author={Al-Fetyani, Mohammad and Al-Barham, Muhammad and Abandah, Gheith and Alsharkawi, Adham and Dawas, Maha},
|
158 |
+
booktitle={2022 IEEE Spoken Language Technology Workshop (SLT)},
|
159 |
+
title={MASC: Massive Arabic Speech Corpus},
|
160 |
+
year={2023},
|
161 |
+
volume={},
|
162 |
+
number={},
|
163 |
+
pages={1006-1013},
|
164 |
+
doi={10.1109/SLT54892.2023.10022652}}
|
165 |
+
}
|
166 |
+
```
|