Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -1,4 +1,9 @@
|
|
1 |
---
|
|
|
|
|
|
|
|
|
|
|
2 |
dataset_info:
|
3 |
features:
|
4 |
- name: audio
|
@@ -17,3 +22,34 @@ configs:
|
|
17 |
- split: train
|
18 |
path: data/train-*
|
19 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
license: cc-by-nc-sa-4.0
|
3 |
+
task_categories: ["automatic-speech-recognition", "text-to-speech"]
|
4 |
+
language: ["vi"]
|
5 |
+
pretty_name: FPT Open Speech Dataset (FOSD)
|
6 |
+
size_categories: ["10K<n<100K"]
|
7 |
dataset_info:
|
8 |
features:
|
9 |
- name: audio
|
|
|
22 |
- split: train
|
23 |
path: data/train-*
|
24 |
---
|
25 |
+
|
26 |
+
# unofficial mirror of FPT Open Speech Dataset (FOSD)
|
27 |
+
|
28 |
+
released publicly in 2018 by FPT Corporation
|
29 |
+
|
30 |
+
100h, 26k samples
|
31 |
+
|
32 |
+
official link (dead): https://fpt.ai/fpt-open-speech-data/
|
33 |
+
|
34 |
+
mirror: https://data.mendeley.com/datasets/k9sxg2twv4/4
|
35 |
+
|
36 |
+
DOI: `10.17632/k9sxg2twv4.4`
|
37 |
+
|
38 |
+
pre-process:
|
39 |
+
- remove non-sense strings: `-N` `\r\n`
|
40 |
+
- remove 4 files because missing transcription:
|
41 |
+
- `Set001_V0.1_008210.mp3`
|
42 |
+
- `Set001_V0.1_010753.mp3`
|
43 |
+
- `Set001_V0.1_011477.mp3`
|
44 |
+
- `Set001_V0.1_011841.mp3`
|
45 |
+
|
46 |
+
usage with HuggingFace:
|
47 |
+
```python
|
48 |
+
# pip install -q "datasets[audio]"
|
49 |
+
from datasets import load_dataset
|
50 |
+
from torch.utils.data import DataLoader
|
51 |
+
|
52 |
+
dataset = load_dataset("doof-ferb/fpt_fosd", split="train", streaming=True)
|
53 |
+
dataset.set_format(type="torch", columns=["audio", "transcription"])
|
54 |
+
dataloader = DataLoader(dataset, batch_size=4)
|
55 |
+
```
|