README.md

metadata

language:
  - am
task_categories:
  - automatic-speech-recognition
pretty_name: Amharic ASR Dataset (CC-BY 4.0)
license: cc-by-4.0

Amharic ASR Dataset

Dataset Summary

This dataset is a processed version of the original BDU-speech dataset by Yohannes A. Ejigu.

It contains paired Amharic speech audio and transcriptions, structured for use in automatic speech recognition (ASR) research and model training.

Audio files are decoded as mono. Sampling rates may vary across files.

---

Dataset Info

DatasetDict({
    train: Dataset({
        features: ['audio', 'sentence'],
        num_rows: 32901
    })
    test: Dataset({
        features: ['audio', 'sentence'],
        num_rows: 3139
    })
})

Features

• audio: Audio(sampling_rate=None, mono=True) — decoded as mono, variable sampling rate
• sentence: Value(dtype='string') — Amharic transcription

---

Language

• Amharic (am)

---

Source Dataset

Original dataset by Yohannes A. Ejigu:

• Figshare: https://figshare.com/articles/dataset/Yohannes_A_Ejigu_Amharic_ASR_Dataset_zip/24959727
• Paper: https://arxiv.org/abs/2503.18485

Please cite the original author when using this dataset.

---

Preprocessing Applied

• Paired audio files with corresponding Amharic transcriptions
• Audio decoded as mono (sampling rate varies across files)
• Structured as Hugging Face DatasetDict with train/test splits

---

Usage

from datasets import load_dataset

ds = load_dataset("chappM/amharic-bdu-asr")

# Access a sample
sample = ds["train"][0]
print(sample["sentence"])   # Amharic transcription
print(sample["audio"])      # {"array": ..., "sampling_rate": ..., "path": ...}

Listen to samples in a Jupyter notebook:

from IPython.display import Audio, display

for i in range(5):
    sample = ds["train"][i]
    print(f"\nIndex {i}")
    print("Text:", sample["sentence"])
    display(Audio(sample["audio"]["array"], rate=sample["audio"]["sampling_rate"]))

---

License

This dataset is distributed under the Creative Commons Attribution 4.0 International License (CC-BY 4.0).

You are free to share and adapt the material provided proper attribution is given to the original author.

---

Citation

If you use this dataset, please cite the original work:

@article{Ejigu2024, author = "Yohannes Ayana Ejigu", title = "{Yohannes A.Ejigu Amharic_ASR_Dataset.zip}", year = "2024", month = "1", url = "https://figshare.com/articles/dataset/Yohannes_A_Ejigu_Amharic_ASR_Dataset_zip/24959727", doi = "10.6084/m9.figshare.24959727.v1" }

---

Acknowledgment

This processed version was prepared for ASR research purposes. All credit for data collection belongs to Yohannes A. Ejigu.