freococo/3hr_myanmar_asr_raw_audio

📚 3-Hour Burmese Speech Dataset from FOEIM Academy (ASR-ready)

This is a curated ~3-hour dataset of Burmese-language audio-transcript pairs derived from the official public-service educational media of FOEIM Academy, a civic platform affiliated with FOEIM.ORG.

It is structured for fine-grained automatic speech recognition (ASR) training and testing.
All data is aligned from timestamped subtitle files (.srt) and segmented into high-quality .mp3 mono files with aligned transcripts.

➡️ YouTube Channel (FOEIM Academy)
➡️ Author: freococo
➡️ License: MIT
➡️ Language: Burmese (my)

---

🔍 Highlights

• 3,200 clips of spoken Burmese
• ~2.90 hours of verified speech data
• Short-form segments ranging from 0.25s to 19.34s
• Cleaned, aligned, and normalized for use in speech models
• Transcripts drawn directly from public .srt subtitles
• Uniform speaker accent across the dataset for consistency

---

📦 Data Format

Each record in metadata.csv contains:

Column	Description
file_name	Relative path to audio (e.g., audio/audio_20250605_0123.mp3)
transcript	Unicode Burmese transcription
audio_duration	Duration of the audio file in seconds (float)

All audio lives inside the /audio/ directory as .mp3 mono files, sampled from long-form content aired on FOEIM’s educational media.

---

📈 Dataset Stats

• 🧩 Total files: 3,200
• ⏱ Total duration: 2.90 hours
• 🪵 Avg. clip length: 3.26 seconds
• 🪶 Shortest clip: 0.25s
• 🎙️ Longest clip: 19.34s

Ideal for small-footprint ASR models, segment-level tokenization tasks, or phrase-level translation experiments in Burmese.

---

📜 License & Credit

This dataset is offered under the MIT License, as a gift of public knowledge, civic empowerment, and linguistic preservation.

Credit to FOEIM.ORG and their educators, producers, and speakers —
whose voices form the foundation of this dataset.

Please retain attribution when adapting or publishing downstream models trained on this dataset.

---

💡 Why This Matters

In a time of erasure, every documented syllable becomes an act of memory.

This dataset is not just for training speech models.
It is for remembering how our people speak, teach, and resist.

Use it with care — and amplify it with purpose.

🇲🇲 This is for the ones who speak, even when unheard.

🧭 What Makes This Dataset Unique?

Unlike most Myanmar datasets which are either religious (e.g., Bible/Qur’an), parliamentary, or scraped from lecture videos, this dataset:

• 📻 Comes from a public-educational voice source
• 📜 Reflects clean Burmese diction with minimal code-switching
• 🧠 Emphasizes civic education, history, ethics, and citizenship

---

🛠 Usage Example

from datasets import load_dataset, Audio

ds = load_dataset("freococo/3hr_myanmar_asr_raw_audio")
ds = ds.cast_column("file_name", Audio())
print(ds[0])

📚 Citation

@dataset{freococo_myanmar_asr_2025_foeim,
  title     = {3-Hour Burmese ASR Dataset (FOEIM)},
  author    = {freococo},
  year      = {2025},
  url       = {https://huggingface.co/datasets/freococo/3hr_myanmar_asr_raw_audio},
  note      = {Curated from FOEIM Academy public videos. Licensed under MIT.}
}