freococo/western_poe_karen_asr

This is the first public Western Poe Karen language ASR dataset in AI history.

Western Poe Karen ASR

This dataset contains audio recordings and aligned transcriptions in the Western Poe Karen language (also known in linguistic literature as Western Pwo or Delta Pwo, ISO 639-3: pwo), a Karenic language spoken primarily in the Ayeyarwady Delta region of Myanmar. Although linguists commonly refer to this language as Western Pwo Karen, the community and this project prefer the spelling Poe to reflect local usage more accurately.

All audio segments in this dataset were sourced from publicly available news broadcasts published by PVTV (People’s Voice Television), the official ethnic news program of the National Unity Government (NUG) of Myanmar. The news program regularly features content in various ethnic languages, including Poe Karen, as part of its commitment to multilingual public communication.

The dataset includes over 18 hours of segmented and labeled audio, prepared in WebDataset format, with paired .mp3 and .json files suitable for training automatic speech recognition (ASR) systems.

Acknowledgments

Special thanks to:

• PVTV (People’s Voice Television) for producing and sharing multilingual content freely
• National Unity Government (NUG) for promoting ethnic language inclusion
• Community volunteers and researchers working on low-resource ASR

This dataset is part of an ongoing effort to build open, inclusive, and community-respectful speech datasets for the languages of Myanmar.

Dataset Structure & Format

This dataset follows the WebDataset format. Each training sample consists of two paired files inside a tar archive:

• XXXX.mp3 — the audio chunk (in MP3 format)
• XXXX.json — the corresponding metadata (UTF-8 JSON)
• 🟢 Minimum chunk duration: 2.04 sec 🔴 Maximum chunk duration: 15.05 sec

Each .json file contains the following fields:

{
  "file_name": "XXXX.mp3",
  "video_id": "YouTubeVideoID",
  "title": "Original broadcast title from PVTV",
  "url": "https://www.youtube.com/watch?v=YouTubeVideoID",
  "duration": 13.24
}

Usage Example

You can use this dataset for training speech recognition models, language modeling on transcriptions, or alignment tasks.

Here's a simple example using Python and WebDataset:

from datasets import load_dataset

dataset = load_dataset(
    "freococo/western_poe_karen_asr",
    split="train",
    streaming=True
)

for sample in dataset:
    print(sample["audio"])
    print(sample["file_name"], sample["duration"])
    print(sample["title"], sample["url"])

Known Limitations

This dataset was segmented automatically from publicly available PVTV news broadcasts using fixed or random chunking strategies. As such, the dataset has the following limitations:

• No transcripts are available. The .json files include only metadata such as title, url, and duration.
• Some audio chunks may contain background music, music-only intros, ambient noise, or non-speech segments.
• There is no filtering or labeling to distinguish between speech and non-speech chunks.
• Speaker variation, audio quality, and noise levels are unbalanced due to the real-world broadcast nature of the source material.

Despite these issues, the dataset offers a valuable foundation for pretraining, domain adaptation, or experimentation on low-resource speech recognition in the Poe (Western Pwo Karen) language.

Licensing & Use

All audio and transcription data in this dataset are released under the Creative Commons Zero (CC0 1.0 Universal) public domain dedication.

The original content was broadcast by PVTV (People’s Voice Television), the ethnic news program of the Myanmar National Unity Government (NUG). The source materials are publicly accessible and were used exclusively for the purpose of linguistic preservation, open-access speech technology, and education.

You are free to:

• Use
• Modify
• Train models
• Distribute or remix

without attribution, for both commercial and non-commercial purposes.

Please respect the context and dignity of the speakers and original producers when using this dataset in public or production systems.

📚 Citation

Freococo (2025).
Western Poe Karen ASR
https://huggingface.co/datasets/freococo/western_poe_karen_asr
Dataset compiled from PVTV ethnic news broadcasts in the Western Poe Karen (Pwo) language.
Public domain (CC0 1.0).