README.md

---
license: cc-by-nc-4.0
task_categories:
- translation
source_datasets:
- PhoST - VinAI
language:
- vi
size_categories:
- 100K<n<1M
pretty_name: Speech to Text Translation Vietnamese
dataset_info:
  features:
  - name: path
    dtype: string
  - name: audio
    dtype:
      audio:
        sampling_rate: 16000
  - name: sentence
    dtype: string
  splits:
  - name: train
    num_examples: 294723
  - name: test
    num_examples: 1095
---

## Dataset Summary
The origin dataset from: [PhoST](https://github.com/VinAIResearch/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 |