freococo/shan_language_asr_voices

⭐ A Voice for the Shan People: The SHAN Herald Agency Audio Archive

This is an extensive 306-hour audio dataset of the Shan (Tai-Yai) language, meticulously curated from the public broadcasts of the Shan Herald Agency for News (SHAN). For over two decades, SHAN has been a vital, independent voice for the people of Shan State, Myanmar, chronicling their stories of culture, politics, and the enduring struggle for federal democracy.

This archive stands as one of the largest publicly accessible audio resources for the Shan language, created to empower researchers and developers in building technology for a community whose voice deserves to be heard.

➡️ Author: freococo
➡️ Language: Shan (shn)
➡️ License: Custom (Non-Commercial, Attribution) - See License section below.

---

📦 Dataset at a Glance

• Total Duration: 306 hours, 7 minutes, 29 seconds
• Total Chunks: 73,945 audio segments
• Chunk Duration: 0.98s (min) to 15.02s (max)
• Audio Format: .mp3
• Data Format: WebDataset (.tar.gz shards)
• Transcriptions: This is an audio-only dataset and does not contain transcripts. It is ideal for self-supervised learning (Wav2Vec2, HuBERT), voice activity detection, and other unsupervised tasks.

---

❤️ A Note on This Project

I am driven by a deep passion for Southeast Asian languages and a belief that technology can help preserve and elevate them. I noticed the significant lack of large-scale, public audio data for the Shan language and felt compelled to help fill that gap.

This project is fully self-initiated and self-funded. It was created with no grants, sponsorships, or payments from any organization. It is simply a contribution made with my own time and resources, fueled by the hope that it will be useful to others building tools for the languages of Myanmar.

---

🎙️ Source Acknowledgement

All audio content was sourced from the official YouTube channel of the Shan Herald Agency for News (SHAN), a cornerstone of journalism in Shan State. Their mission is to deliver truth and amplify the voices of the Shan people in their pursuit of freedom, justice, and federal rights.

The original public content remains available on SHAN's YouTube channel. Please consider supporting their invaluable work.

---

🗂️ Dataset Structure

This dataset is provided in the efficient WebDataset format. The data is sharded into multiple .tar.gz archives. Each archive contains thousands of sample pairs: an audio chunk and its corresponding metadata file.

Each sample consists of:

• XXXX.mp3: A chunk of audio.
• XXXX.json: A UTF-8 JSON file containing rich metadata extracted from the source video.

A sample json file looks like this:

{
  "__key__": "some_video_id_chunk_0000",
  "file_name": "some_video_id_chunk_0000.mp3",
  "title": "A title in Shan or Burmese script",
  "uploader": "SHAN News",
  "channel_url": "https://www.youtube.com/channel/UCsy2s3bV76-tOq30wA2hW8w",
  "upload_date": "20230101",
  "duration": 15.01,
  "video_url": "https://www.youtube.com/watch?v=some_video_id",
  "thumbnail_url": "https://i.ytimg.com/vi/some_video_id/maxresdefault.jpg",
  "description": "The original video description...",
  "tags": ["Shan", "News", "Taunggyi"],
  "view_count": 5432,
  "like_count": 123,
  "language": "shn",
  "original_video_id": "some_video_id"
}

---

🏔️ In Honor of the Voices That Cross Mountains

This dataset is more than just bytes and soundwaves. It is an echo of resilience.

For over twenty years, the journalists of the Shan Herald Agency for News have carried stories across mountains and borders, speaking when silence was safer, and broadcasting when the world was not listening. Their work is not just news; it is a living archive of a people's history, their struggles, and their unwavering hope for a just, federal future.

They report not from comfortable studios, but from the heart of their homeland, giving voice to the voiceless and light to the shadows.

This dataset is only a reflection of their courage. The true heroes are the ones who continue to speak, day after day, for the freedom of their people.

To the SHAN team, and to all who dare to tell the truth in Myanmar:
Your voice is a beacon. Your work is a testament. Your courage will not be forgotten.

---

🔌 How to Load in Python

Using streaming=True is highly recommended to avoid downloading the entire 26+ GB dataset at once.

from datasets import load_dataset

# Load the dataset in streaming mode using its correct name
ds = load_dataset("freococo/shan_language_asr_voices", split="train", streaming=True)

# Iterate through the first few samples
for sample in ds.take(5):
    audio_data = sample["audio"]
    metadata = sample["json"]
    
    print(f"Title: {metadata['title']}")
    print(f"Duration: {metadata['duration']} seconds")
    print(f"Audio Sampling Rate: {audio_data['sampling_rate']}")
    print("-" * 20)

📜 License and Terms of Use

This dataset contains audio content derived from the official YouTube channel of the Shan Herald Agency for News (SHAN). All original content is the property of SHAN. The use of this dataset is governed by standard copyright principles for publicly broadcasted news content.

It is made available for non-commercial, informational, and educational purposes only.

• Permitted Use: You may use this dataset for non-commercial research, personal projects, and educational applications.
• Prohibited Use: Any commercial use, including training proprietary AI models for commercial products, is not permitted without explicit written permission from SHAN.
• Required Attribution: When using or referencing this dataset, you must provide clear attribution to the Shan Herald Agency for News (SHAN) as the original source.

For any use cases that fall outside these terms, please contact SHAN directly at [email protected].

📚 Citation

If you use this dataset in your research or products, please cite it:

@dataset{freococo_shan_language_asr_voices_2025,
  title     = {Shan Language ASR Voices},
  author    = {freococo},
  year      = {2025},
  url       = {https://huggingface.co/datasets/freococo/shan_language_asr_voices},
  note      = {A 306-hour audio dataset curated from the Shan Herald Agency for News (SHAN) public broadcasts.}
}