Datasets:
aaparajit02
commited on
Commit
•
f54201f
1
Parent(s):
9495ffd
Update README.md
Browse files
README.md
CHANGED
@@ -13,7 +13,31 @@ dataset_info:
|
|
13 |
num_examples: 39238
|
14 |
download_size: 10866820110
|
15 |
dataset_size: 10917088956.322
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
---
|
17 |
-
# Dataset Card for "punjabi-asr"
|
18 |
|
19 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
num_examples: 39238
|
14 |
download_size: 10866820110
|
15 |
dataset_size: 10917088956.322
|
16 |
+
task_categories:
|
17 |
+
- automatic-speech-recognition
|
18 |
+
language:
|
19 |
+
- pa
|
20 |
+
tags:
|
21 |
+
- punjabi
|
22 |
+
- asr
|
23 |
+
- transcription
|
24 |
+
- translation
|
25 |
+
pretty_name: Punjabi ASR
|
26 |
+
size_categories:
|
27 |
+
- 10K<n<100K
|
28 |
---
|
|
|
29 |
|
30 |
+
# Dataset for Punjabi ASR
|
31 |
+
Shrutilipi is a labelled ASR corpus obtained by mining parallel audio and text pairs at the document scale from All India Radio news bulletins for 12 Indian languages: Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Odia, Punjabi, Sanskrit, Tamil, Telugu, Urdu. The corpus has over 6400 hours of data across all languages.
|
32 |
+
|
33 |
+
```
|
34 |
+
@misc{https://doi.org/10.48550/arxiv.2208.12666,
|
35 |
+
doi = {10.48550/ARXIV.2208.12666},
|
36 |
+
url = {https://arxiv.org/abs/2208.12666},
|
37 |
+
author = {Bhogale, Kaushal Santosh and Raman, Abhigyan and Javed, Tahir and Doddapaneni, Sumanth and Kunchukuttan, Anoop and Kumar, Pratyush and Khapra, Mitesh M.},
|
38 |
+
title = {Effectiveness of Mining Audio and Text Pairs from Public Data for Improving ASR Systems for Low-Resource Languages},
|
39 |
+
publisher = {arXiv},
|
40 |
+
year = {2022},
|
41 |
+
copyright = {arXiv.org perpetual, non-exclusive license}
|
42 |
+
}
|
43 |
+
```
|