README.md

---
license: other
language:
- kyu
pretty_name: RFA Karenni Language Voices
tags:
- audio
- asr
- speech-recognition
- karenni
- kayah
- webdataset
- myanmar
task_categories:
- automatic-speech-recognition
- audio-classification
- voice-activity-detection
language_creators:
  - found
source_datasets:
  - original
---

# RFA Karenni (Kayah) Language Voices

This dataset contains **17 hours** of audio in the **Karenni (Kayah)** language, sourced from news broadcasts by **Radio Free Asia (RFA) Burmese**. This is one of the largest publicly accessible audio resources for the Karenni language family, designed to support research in low-resource automatic speech recognition (ASR), voice activity detection, and other speech-related tasks.

This dataset was created by **freococo**.

The audio has been automatically segmented into **4,170** manageable chunks and prepared in the efficient [WebDataset](https://github.com/webdataset/webdataset) format. It includes rich metadata extracted from the original YouTube videos, but **does not contain transcriptions**, making it ideal for self-supervised pre-training or tasks that do not require text.

*Note on Language Codes: The source material is labeled as "Karenni." The specific dialect (e.g., Eastern or Western Kayah) is not specified. For clarity and precision, the ISO 639-3 code `kyu` (Western Kayah) has been used as a primary identifier, though the content may encompass broader Karenni language varieties.*

### Acknowledgments

This dataset was made possible by the journalists and broadcasters at **Radio Free Asia (RFA)** who produce and publish high-quality content in the languages of Myanmar.

## Dataset Structure & Format

This dataset follows the [WebDataset](https://github.com/webdataset/webdataset) format. Each sample consists of two paired files inside a `.tar.gz` archive:

- `XXXX.mp3` — the audio chunk (up to 15 seconds)
- `XXXX.json` — the corresponding metadata (UTF-8 JSON)

🟢 **Minimum chunk duration:** 1.15 seconds  
🔴 **Maximum chunk duration:** 15.01 seconds

Each `.json` file contains a rich set of fields preserved from the original YouTube video:

```json
{
  "__key__": "ZYmHR116xH0_chunk_0007",
  "file_name": "ZYmHR116xH0_chunk_0007.mp3",
  "title": "ကရင်နီတိုင်းရင်းသားဘာသာ အစီအစဉ်",
  "uploader": "RFA လွတ်လပ်တဲ့အာရှအသံ",
  "channel_url": "https://www.youtube.com/channel/UCE75dgnEYPacknHHg3a3sJg",
  "upload_date": "20200811",
  "duration": 15.012,
  "video_url": "https://www.youtube.com/watch?v=ZYmHR116xH0",
  "thumbnail_url": "https://i.ytimg.com/vi_webp/ZYmHR116xH0/maxresdefault.webp",
  "description": "ကရင်နီတိုင်းရင်းသားဘာသာ အစီအစဉ်...",
  "view_count": 314,
  "like_count": 1,
  "language": "kyu",
  "original_video_id": "ZYmHR116xH0"
}
```

## Usage Example

You can easily stream this dataset using the Hugging Face `datasets` library. The `streaming=True` mode is highly recommended.

```python
from datasets import load_dataset

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

for sample in dataset.take(5):
    print(sample["audio"])          # Audio object with decoded waveform
    print(sample["json"]["title"])  # Access metadata from the JSON
    print(f"Duration: {sample['json']['duration']}s")
```

## Known Limitations

This dataset was created using an automated pipeline and has the following characteristics:

- **No Transcriptions:** This dataset is audio-only and does not include text transcriptions.
- **Background Noise:** As the audio is sourced from news broadcasts, some segments may contain background music, jingles, or other non-speech sounds.
- **No Speaker Labels:** The dataset does not differentiate between different speakers within or across broadcasts.

Despite these limitations, this resource is invaluable for pre-training models like Wav2Vec2, developing voice activity detection systems, or performing other unsupervised speech research for the Karenni language.

## Licensing & Use

This dataset contains audio content derived from the official YouTube channel of **Radio Free Asia (RFA) Burmese**. All original content is the exclusive property of RFA and is protected by copyright.

The use of this dataset is governed by the official [RFA Terms of Use](https://www.rfa.org/about/termsofuse).

In summary, the content 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 but not limited to training proprietary AI models for commercial products, is not permitted without explicit written permission from RFA.
- **Required Attribution**: When using or referencing this dataset, you must provide clear attribution to **Radio Free Asia (RFA)** as the original source of the content.

By using this dataset, you agree to abide by these terms.

## 📚 Citation

```
@dataset{freococo_karenni_language_asr_audio_2025,
  author    = {freococo},
  title     = {RFA Karenni (Kayah) Language Voices},
  year      = {2025},
  url       = {https://huggingface.co/datasets/freococo/karenni_language_asr_audio},
  note      = {Dataset compiled from Radio Free Asia (RFA) Burmese news broadcasts.}
}
```