Datasets:

freococo
/

raw_1hr_myanmar_asr_audio

license: mit
pretty_name: Raw 1-Hour Burmese ASR Audio Dataset
dataset_type: audio
task_categories:
  - automatic-speech-recognition
language:
  - my
tags:
  - Myanmar
  - Burmese
  - Speech
  - RawAudio
  - PVTV
  - NUG
  - ASR

🇲🇲 Raw 1-Hour Burmese ASR Audio Dataset

A 1-hour dataset of Burmese (Myanmar language) spoken audio clips with transcripts, curated from official public-service media broadcasts by PVTV Myanmar — the media voice of Myanmar’s National Unity Government (NUG).

This dataset is intended for automatic speech recognition (ASR) and Burmese speech-processing research.

➡️ Author: freococo
➡️ License: MIT
➡️ Language: Burmese (my)

📦 Dataset Summary

Duration: ~1 hour
Chunks: Short utterances (0.84s to 25.66s)
Total Samples: 583
Audio Format: .mp3 mono files
Transcription Source: Aligned manually using .srt files
Structure: file_name, transcript, duration_seconds

The dataset was created entirely from public content with no modification or noise reduction applied.

⚠️ Data Quality Notes

This dataset contains raw speech audio extracted from public media without denoising or filtering.
Some chunks contain background music, instrumental intros/outros, or ambient reverb.
Transcripts were manually aligned via subtitle files (.srt) and are mostly accurate.
Estimated transcription error rate: 1–9%, due to:
- Minor typos or spacing issues in Burmese script
- Occasional missing particles or honorifics

These conditions reflect real-world media audio and are left untouched to improve robustness in training and evaluation.

💬 Motivation

I created this dataset because I'm crazy about languages — especially Myanmar language technology.
I noticed a severe shortage of public, modern Burmese audio datasets for speech recognition and wanted to help fix that.

This project is fully self-initiated and unfunded — no grants, sponsorships, or institutional backing. Just passion, time, and a lot of cleaning 😄
If you find it helpful, let me know — I’d love to collaborate or help with related research!

🎙️ Source Acknowledgement

All audio was derived from PVTV Myanmar — a public voice media channel established by Myanmar’s National Unity Government (NUG).
Their mission is to amplify the people's voice in pursuit of freedom, justice, and federal democracy.

⚠️ This dataset contains raw audio, including background music or ambiance. It is not denoised or processed — intended to reflect real-world conditions.

The original public content remains available on PVTV’s YouTube channel.

🗂️ Dataset Structure

Each row in metadata.csv includes:

Column	Description
`file_name`	Relative path to audio file (e.g., `audio/my_audio_001.mp3`)
`transcript`	Burmese-language transcription
`duration_seconds`	Duration of the audio file in seconds

The audio files are mono .mp3 files stored in the audio/ folder.

🌟 In Honor of the Voices Behind the Revolution

This dataset would not exist without the tireless, fearless voices of PVTV Myanmar —
🎙️ the journalists who speak truth,
✍️ the editors who shape it,
📢 and the citizens who carry it forward.

They speak not from studios, but from shadows,
not for fame, but for freedom.
Their words echo through uncertainty,
yet land on ears yearning for light.

This dataset is only a shadow of their work —
the real heroes are the ones who dare to speak when silence is safer.

To the PVTV media team and all those risking safety to tell the truth:
Your voice is our history. Your courage is our future.

🇲🇲🕊️ Long live the Spring Revolution.

🔌 How to Load in Python

from datasets import load_dataset, Audio

ds = load_dataset("freococo/raw_1hr_myanmar_asr_audio")
ds = ds.cast_column("file_name", Audio())
ds[0]

📚 Citation

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

@dataset{freococo_myanmar_asr_2025,
  title     = {Raw 1-Hour Myanmar ASR Audio Dataset},
  author    = {freococo},
  year      = {2025},
  url       = {https://huggingface.co/datasets/freococo/raw_1hr_myanmar_asr_audio},
  note      = {Curated from PVTV Myanmar public media, licensed under MIT}
}