freococo/nug_myanmar_open_asr_corpus

🇲🇲 NUG Myanmar Open ASR Corpus

The NUG Myanmar Open ASR Corpus is the first large-scale open Burmese speech dataset, containing over 150,000 audio-text pairs totaling ~100 hours of clean, segmented audio. All data was collected from public-service educational broadcasts by the National Unity Government (NUG) of Myanmar and the FOEIM Academy.

This dataset is released under a CC0 1.0 Universal license — fully open and public domain. No attribution required.

---

📦 Dataset Structure (WebDataset Format)

The dataset is stored as a WebDataset-compatible archive with:

• train/00000.tar, ..., train/00015.tar
• Each .tar file contains:
  • XXXXX.mp3: audio chunk
  • XXXXX.json: metadata with fields:
    • "transcript" (string)
    • "duration" (float)

Total: 150,082 audio-text pairs

---

🧪 Example Usage with Hugging Face datasets

from datasets import load_dataset

ds = load_dataset(
    "freococo/nug_myanmar_open_asr_corpus",
    data_dir="train",
    split="train"
)

# Example
print(ds[0]["json"])  # {'transcript': "...", 'duration': 2.3}
print(ds[0]["mp3"]["array"])  # NumPy waveform
print(ds[0]["mp3"]["sampling_rate"])  # e.g., 44100

---

📊 Dataset Stats

• 🧾 Total examples: 150,082
• ⏱️ Duration: ~100 hours
• 📏 Avg. chunk: 2.4 seconds
• 🎙️ Domain: Civic education, teacher voiceovers, narration
• 💬 Language: Burmese (မြန်မာဘာသာ)
• 📦 Format: .tar (WebDataset), .mp3 + .json

---

🕊️ Background & Acknowledgment

This dataset was created from public-service educational videos broadcast by the NUG (National Unity Government) and the FOEIM Academy, designed to reach students under political oppression.

The voices preserved in this dataset taught in hiding — under threat — yet persisted in educating a generation.

---

🎙️ Credit Where It’s Due

The true heroes and heroines of this dataset are the NUG teachers.

They taught online during a revolution. They recorded lessons under threat. Their voices — calm, clear, and determined — are now preserved in this dataset.

• This corpus exists to honor their clarity, their calm — and their courage.

---

⚠️ Known Limitations & Cautions

While this dataset is the first large-scale open Burmese ASR corpus in history, it is not perfect — and developers should be aware of the following caveats:

• Transcripts are machine-aligned and not manually verified.

  • Minor typos, spelling inconsistencies, or segmentation errors may exist.
  • If you're training a general ASR model (e.g., Whisper-style multilingual systems), this noise is generally tolerable.
  • If you're building domain-specific or low-error ASR applications (e.g., dictation, legal, or medical), we strongly recommend human verification and correction.

• Some pronunciation variations and informal phrasing may appear in teacher speech.

• Accent and dialectal bias is unknown — most voices come from urban-educated speakers aligned with Myanmar school broadcasts.

• Despite these imperfections, the dataset offers valuable coverage of clean, read Burmese speech — recorded under real-world public service conditions. It is an unprecedented resource for modeling a once low-resource language.

---

🕊️ A Moment in History

This is more than data — it's a declaration.

The National Unity Government (NUG) of Myanmar may not yet be able to liberate its people from the grip of a low militants junta,
but it has already liberated the Burmese language from being called a "low-resource" language in the global AI landscape.

• 🔓 For the first time in history, 100+ hours of Burmese speech have been openly released — no copyright, no restriction, no paywall.

This dataset stands as a milestone — not just for Myanmar, but for the global fight to make language technology equitable.

Free the voice. Free the language. Free the people.

---

🧭 Future Releases

This dataset represents Volume 1 — approximately one-third of the full corpus.

Planned upcoming releases include:

• 📦 Volume 2 (100+ additional hours of speech)
• 🗂️ Continued additions of segmented, cleaned audio chunks

---

This dataset is freely given, so that Burmese may never again be called a low-resource language.

Free the voice. Free the language. Free the people.

---

📚 Citation

@misc{nug_asr_2025,
  title = {NUG Myanmar Open ASR Corpus},
  author = {freococo},
  year = {2025},
  publisher = {Hugging Face Datasets},
  howpublished = {\url{https://huggingface.co/datasets/freococo/nug_myanmar_open_asr_corpus}},
  note = {Open Burmese ASR corpus from NUG civic broadcasts. CC0 licensed.}
}