Datasets:
The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider
removing the
loading script
and relying on
automated data support
(you can use
convert_to_parquet
from the datasets library). If this is not possible, please
open a discussion
for direct help.
Dataset Summary
The origin dataset from: PhoST
A speech-to-text translation dataset with source audio in English and target sentences in Vietnamese.
In this STVi dataset, we split the origin audio into 1-15s corresponding to each sentence and spelling correction, normalization word, audio segmentation
Data Structure
{'path': 'stvi/train/waves/315814.wav', 'audio': {'path': 'stvi/train/waves/315814.wav', 'array': array([-0.0050354 , -0.00296021, -0.00549316, ..., -0.0085144 ,
-0.00686646, -0.00018311]), 'sampling_rate': 16000}, 'sentence': 'tôi cảm thấy bẽ mặt và xấu hổ đến mức tôi chạy ngay về nhà với mẹ và trách phạt bà vì để tôi mặc cái áo gớm guốc'}
Data Fields
- path: The path to the audio file
- audio: A dictionary containing the path to the downloaded audio file, the decoded audio array, and the sampling rate.
- sentence: The transcript sentence
Data Splits
The speech material has been subdivided into portions for train and test.
The speech was split from a TED Talk, and each speech segment corresponds to a transcript sentence.
| Train | Test | |
|---|---|---|
| Utterances | 294723 | 1054 |
| Duration (in hours) | 425.91 | 1.616 |
- Downloads last month
- 0