freococo/myanmar-english-accent-speech

Myanmar English Accent Speech (PVTV & FOEIM)

This dataset contains English speech by Myanmar speakers, collected from public videos published by PVTV and FOEIM — two media channels operating under the National Unity Government (NUG).

The clips reflect a wide range of spoken English contexts: interviews, announcements, sermons, and educational content. The speakers vary in tone, pace, and emotion — but all share the characteristic sound of Burmese-accented English.

This dataset was created for research, education, and technology building — especially for speech recognition, accent modeling, and accessibility in Myanmar.

This dataset would not exist without the excellent journalists, fantastic teachers, amazing editors, and brave spiritual leaders who speak publicly in English through PVTV and FOEIM — even under immense pressure, risk, and hardship.

They are the real heroes and heroines of this work. Their voices are not just informative — they are courageous, insightful, and deeply human. All credit belongs to THEM.

I claim no ownership over their words. I simply listened, learned, and organized what they already offered — with clarity, dignity, and purpose.

📦 Dataset Contents

Each .tar file contains multiple audio-transcript pairs:

- `.mp3` — Short English utterances (1–10 seconds)  
- `.json` — Includes:
  - "transcript": the English sentence  
  - "duration": float (audio length in seconds)

Files are named numerically (0001.mp3, 0001.json, etc.) and designed for easy streaming.

🔍 Duration Overview

- Total clips: 4,249  
- Total duration: approximately 3.5 hours  
- Min/Max duration: 0.2 sec → 17.0 sec

🚀 How to Use This Dataset

This dataset is suitable for:

- Training automatic speech recognition (ASR) systems  
- Accent-adaptive transcription  
- Pronunciation modeling for Myanmar English  
- Speech and communication research in resistance and education  

Each .tar archive contains multiple audio-transcript pairs:

- `.mp3` — Spoken English audio (typically 1–10 seconds)  
- `.json` — Contains:
  - "transcript": the spoken English sentence  
  - "duration": the audio duration in seconds (as float)

You can extract with the command:

tar -xvf 00001.tar

Each pair will be named like:

0001.mp3  
0001.json

Each JSON contains two fields:

"transcript" and "duration"

To read the contents using Python:

import tarfile, json  
with tarfile.open("00001.tar", "r") as tar:  
  for member in tar.getmembers():  
    if member.name.endswith(".json"):  
      data = json.load(tar.extractfile(member))  
      print(member.name, "→", data["transcript"])

Using WebDataset:

import webdataset as wds  
dataset = wds.WebDataset("00001.tar").decode().to_tuple("mp3", "json")  
for audio, meta in dataset:  
  print(meta["transcript"])

Using Hugging Face Datasets:

from datasets import load_dataset  
dataset = load_dataset("freococo/myanmar-english-accent-speech", split="train")  
streaming = load_dataset("freococo/myanmar-english-accent-speech", split="train", streaming=True)  
for sample in streaming:  
  print(sample["transcript"])

🖥️ Edit Transcripts with GUI (Optional Tool)

This dataset includes a simple local GUI tool for reviewing and improving transcripts:
transcript_editor_gui.py

This optional script lets you:

• Listen to each .mp3 file inside a .tar
• View and edit the corresponding "transcript" in .json
• Save your changes and export an updated .tar file

To use:

1. Download a .tar file (e.g., 00001.tar) from this dataset
2. Download the script: transcript_editor_gui.py
3. Open the script in a text editor
4. Change this line to your local tar file path:
   tar_path = "/path/to/00001.tar"
5. Install the audio playback module:

   pip install playsound

6. Run the script:

  python3 transcript_editor_gui.py

7. A window will appear where you can:
   • Press "Replay" to listen
   • Edit the transcript in the box
   • Press "Save & Next" to continue
   • Click "Export Updated TAR" to save your edits

This runs entirely offline and is a great way to contribute to the dataset or adapt it for your own use.

📝 License & Source

This dataset is released under the CC0-1.0 license (public domain).
You are free to use, modify, redistribute, and build upon it for any purpose.

The original content comes from:

• PVTV – A public-interest journalism platform sharing news, interviews, and announcements in support of democracy.
• FOEIM – A media channel focused on spiritual teachings, community wisdom, and peaceful resistance.

We acknowledge the work of these media teams, speakers, and editors.
This dataset is a curated reflection of their public contributions — not a product of commercial funding or institutional backing.

Public Interest Declaration

This dataset was created independently, without sponsorship, institutional affiliation, or financial compensation.
No donations, grants, or commercial arrangements were involved in its collection or release.

It was assembled solely for public benefit — to support language technology, research, and accessibility for the people of Myanmar.
We especially hope it serves:

- Educators and students  
- Technologists building inclusive tools  
- Linguists and researchers studying Burmese English  
- Communities advocating for digital democracy

We share it freely, and ask only that it be used ethically, respectfully, and in good faith.

📖 Citation

If you use this dataset in your research, training, or tools, please cite it as follows:

@misc{freococo2025myanmarenglish,  
  title        = {Myanmar English Accent Speech (PVTV & FOEIM)},  
  author       = {freococo},  
  year         = {2025},  
  howpublished = {\url{https://huggingface.co/datasets/freococo/myanmar-english-accent-speech}},  
  note         = {English speech dataset from NUG-aligned public media}  
}

Though modest in size, this dataset is meaningful in voice.
May it help machines listen better — and help humans speak more freely.