metadata

license: cc0-1.0
language:
  - kar
language_bcp47:
  - ksw
pretty_name: Sagaw Karen ASR
tags:
  - automatic-speech-recognition
  - audio
  - karen
  - burmese
  - myanmar
  - webdataset
  - public-domain
task_categories:
  - automatic-speech-recognition
  - audio-to-audio
  - audio-classification
language_creators:
  - found
source_datasets:
  - original

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

Sagaw Karen ASR

This dataset contains audio recordings and aligned metadata in the Sagaw Karen language (ISO 639-3: ksw), a major Sgaw Karenic language spoken throughout southern and eastern Myanmar. The language is sometimes also referred to as Sgaw Karen or Sakaw Karen in English transliterations.

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. These news programs regularly feature content in various ethnic languages, including Sagaw Karen, as part of a multilingual broadcast strategy.

The dataset includes over 8.8 hours of segmented and labeled audio, prepared in WebDataset format with .audio and .json file pairs, suitable for training automatic speech recognition (ASR) models.

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.audio — the audio chunk (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.audio",
  "video_id": "YouTubeVideoID",
  "title": "Original broadcast title from PVTV",
  "url": "https://www.youtube.com/watch?v=YouTubeVideoID",
  "duration": 13.24
}

Usage Example

You can load and stream this dataset using the Hugging Face datasets library:

from datasets import load_dataset

# Load the dataset in streaming mode
dataset = load_dataset(
    "freococo/sagaw_karen_asr",
    split="train",
    streaming=True
)

# Iterate through samples
for sample in dataset:
    print(sample["audio"])        # Audio object (decoded)
    print(sample["file_name"])    # Audio chunk file name
    print(sample["duration"])     # Duration in seconds
    print(sample["title"])        # PVTV news broadcast title
    print(sample["url"])          # Original YouTube video URL

Each sample includes:

audio: the audio chunk (.audio format, usually MP3)
file_name: original filename of the chunk
title: title of the original PVTV news program
url: direct YouTube source link
video_id: YouTube ID
duration: duration of the audio segment in seconds

Known Limitations

This dataset was automatically segmented from PVTV news broadcasts using pause-based and fixed-length chunking strategies. As a result, the following limitations should be noted:

No transcriptions are included. The dataset is intended for pretraining or weakly-supervised learning.
Some chunks may contain only music, background ambiance, or non-speech segments (e.g., intro jingles or broadcast effects).
There is no speaker labeling, noise classification, or manual filtering of content.
Variation in speaker accent, age, gender, and recording quality may affect ASR performance.

Despite these limitations, this is the first open dataset of its kind in Sagaw Karen and is useful for low-resource ASR development, language identification, or pretraining pipelines.

Licensing & Use

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

This means you are free to:

✅ Use the data for any purpose
✅ Modify or adapt the content
✅ Train models (open or commercial)
✅ Share or redistribute it

No attribution is required, though acknowledgment of PVTV as the original broadcaster is encouraged in academic or ethical contexts.

The original recordings were published publicly by PVTV (People’s Voice Television) — the ethnic news and public outreach arm of Myanmar's National Unity Government (NUG). These materials were repurposed solely for public-interest research and language technology development in underserved languages.

📚 Citation

If you use this dataset, please cite:

Freococo (2025).
Sagaw Karen ASR
Hugging Face. https://huggingface.co/datasets/freococo/sagaw_karen_asr
Dataset compiled from PVTV ethnic news broadcasts in the Sagaw Karen language.
Released under CC0 1.0 (Public Domain).