kha2612/ViSL-News

---
language:
- vi
pretty_name: ViSL-News
task_categories:
- translation
- text-generation
- video-text-to-text
size_categories:
- 10K<n<100K
license: other
annotations_creators:
- machine-generated
- expert-generated
source_datasets:
- original
---

# ViSL-News

## Dataset Summary

**ViSL-News** is a sentence-level Vietnamese Sign Language dataset built from news videos with Vietnamese sign language interpretation.
The dataset was created from **HTV Tin Tức** news videos published during **2024–2025**, and each sample corresponds to **one complete sentence** paired with a short video clip. :contentReference[oaicite:3]{index=3} :contentReference[oaicite:4]{index=4}

The dataset was constructed using **ViSL Tool**, a semi-automatic pipeline that supports:
- data filtering,
- automatic sentence-level labeling,
- signer clustering,
- metadata generation,
- and human post-checking. :contentReference[oaicite:5]{index=5} :contentReference[oaicite:6]{index=6}

ViSL-News is intended to support research on:
- Vietnamese Sign Language Translation,
- sentence-level video-to-text learning,
- dataset construction for sign language research,
- and data-centric methods for sign language processing.

## Languages

- **Sign language:** Vietnamese Sign Language (as interpreted in broadcast news videos)
- **Text language:** Vietnamese

## Dataset Structure

Each sample is sentence-level and typically contains:
- a short video clip,
- a Vietnamese sentence,
- start and end timestamps,
- and signer-related metadata when available. :contentReference[oaicite:7]{index=7} :contentReference[oaicite:8]{index=8}

Example fields may include:
- `sample_id`
- `video`
- `text`
- `start_time`
- `end_time`
- `signer_id`
- `split`

The exact field names may vary depending on the released version.

## Dataset Creation

### Source Data

The dataset was collected from **HTV Tin Tức** news videos on YouTube that include Vietnamese sign language interpretation.
Within the scope of the thesis, the data source is limited to **HTV Tin Tức (2024–2025)** news videos to ensure feasibility and quality control. :contentReference[oaicite:9]{index=9}

### Creation Pipeline

ViSL-News was built with a semi-automatic workflow:
1. collect source videos,
2. filter suitable signing segments,
3. generate sentence-level labels,
4. align timestamps,
5. group signer information,
6. perform post-checking on selected data,
7. export the final dataset. :contentReference[oaicite:10]{index=10}

The system was designed to reduce manual workload while preserving a human review step for quality assurance. :contentReference[oaicite:11]{index=11} :contentReference[oaicite:12]{index=12}

### Annotation Process

The dataset combines:
- **automatic processing** from the ViSL Tool pipeline, and
- **expert/human post-checking** for quality control. :contentReference[oaicite:13]{index=13} :contentReference[oaicite:14]{index=14}

A portion of the data was post-checked to build more reliable validation and test subsets. :contentReference[oaicite:15]{index=15}

## Dataset Statistics

According to the thesis, ViSL-News includes:
- **26,395 samples**
- **59 hours** of video
- **9,918-word vocabulary**
- video resolution of **326×426**
- **25 FPS** :contentReference[oaicite:16]{index=16}

## Intended Uses

This dataset is released for:
- academic research,
- educational use,
- benchmarking and experimentation,
- and non-commercial scientific work on Vietnamese sign language processing.

Example use cases:
- sign language translation,
- sentence-level video-to-text modeling,
- data-centric studies on alignment and annotation,
- dataset construction research.

## Out-of-Scope Uses

This dataset is **not** intended for:
- commercial deployment,
- commercial redistribution,
- monetized reuse of source videos,
- or any use that violates the rights of the original content owner.

## Copyright and Usage Notice

### Copyright

The **original broadcast videos and source content remain the property of HTV Tin Tức**.

This dataset is derived from HTV Tin Tức news videos for research purposes.
The dataset publisher does **not** claim ownership of the original source videos or broadcast content.

### Usage Restriction

This dataset is provided **for research and non-commercial use only**.

By using this dataset, you agree that:
- you will use it only for **research or educational purposes**,
- you will **not** use it for commercial or business purposes,
- you will **not** redistribute or exploit the original source videos for commercial gain,
- and you will properly acknowledge **HTV Tin Tức** as the original content source.

If the copyright holder requests modification or removal of the dataset, the maintainer will cooperate accordingly.

## Limitations

- The dataset is limited to the **news domain**, so it may not represent all Vietnamese Sign Language usage contexts. :contentReference[oaicite:17]{index=17}
- The data comes from interpreted broadcast videos rather than unconstrained real-world conversations.
- Sentence alignment and labeling are supported by a semi-automatic pipeline, so some noise may still remain.
- Signer diversity is constrained by the available source videos and broadcast setting. :contentReference[oaicite:18]{index=18}

## Citation

If you use this dataset, please cite the corresponding thesis or dataset release.

```bibtex
@misc{visl_news_2026,
title = {ViSL-News: A Sentence-Level Vietnamese Sign Language Dataset from HTV Tin Tuc News Videos},
author = {Nguyen Hieu Kha},
year = {2026},
note = {Research-use-only dataset derived from HTV Tin Tuc source videos}
}