metadata
dataset_info:
- config_name: afaan_oromo
features:
- name: audio_id
dtype: string
- name: speaker_id
dtype: string
- name: audio
dtype:
audio:
sampling_rate: 16000
- name: audio_duration
dtype: float64
- name: transcription
dtype: string
- name: gender
dtype: string
- name: age_group
dtype: string
- name: locale
dtype: string
splits:
- name: validation
num_bytes: 940133880
num_examples: 1610
- name: test
num_bytes: 1174688045
num_examples: 2022
- name: train
num_bytes: 3969659858
num_examples: 6742
download_size: 5537062936
dataset_size: 6084481783
- config_name: all
features:
- name: audio_id
dtype: string
- name: speaker_id
dtype: string
- name: audio
dtype:
audio:
sampling_rate: 16000
- name: audio_duration
dtype: float64
- name: transcription
dtype: string
- name: gender
dtype: string
- name: age_group
dtype: string
- name: locale
dtype: string
- name: language
dtype: string
splits:
- name: train
num_bytes: 25911133921
num_examples: 45945
- name: validation
num_bytes: 4534078046
num_examples: 8097
- name: test
num_bytes: 5656271339
num_examples: 10119
download_size: 31681034828
dataset_size: 36101483306
- config_name: amharic
features:
- name: audio_id
dtype: string
- name: speaker_id
dtype: string
- name: audio
dtype:
audio:
sampling_rate: 16000
- name: audio_duration
dtype: float64
- name: transcription
dtype: string
- name: gender
dtype: string
- name: age_group
dtype: string
- name: locale
dtype: string
splits:
- name: validation
num_bytes: 890896389
num_examples: 1630
- name: test
num_bytes: 1105571353
num_examples: 2006
- name: train
num_bytes: 5988421547
num_examples: 10839
download_size: 6990827143
dataset_size: 7984889289
- config_name: sidama
features:
- name: audio_id
dtype: string
- name: speaker_id
dtype: string
- name: audio
dtype:
audio:
sampling_rate: 16000
- name: audio_duration
dtype: float64
- name: transcription
dtype: string
- name: gender
dtype: string
- name: age_group
dtype: string
- name: locale
dtype: string
splits:
- name: validation
num_bytes: 915876080
num_examples: 1605
- name: test
num_bytes: 1130712469
num_examples: 2014
- name: train
num_bytes: 3825056496
num_examples: 6726
download_size: 5306637429
dataset_size: 5871645045
- config_name: tigrinya
features:
- name: audio_id
dtype: string
- name: speaker_id
dtype: string
- name: audio
dtype:
audio:
sampling_rate: 16000
- name: audio_duration
dtype: float64
- name: transcription
dtype: string
- name: gender
dtype: string
- name: age_group
dtype: string
- name: locale
dtype: string
splits:
- name: validation
num_bytes: 872101097
num_examples: 1622
- name: test
num_bytes: 1101900725
num_examples: 2065
- name: train
num_bytes: 6021893910
num_examples: 11093
download_size: 6802038899
dataset_size: 7995895732
- config_name: wolaytta
features:
- name: audio_id
dtype: string
- name: speaker_id
dtype: string
- name: audio
dtype:
audio:
sampling_rate: 16000
- name: audio_duration
dtype: float64
- name: transcription
dtype: string
- name: gender
dtype: string
- name: age_group
dtype: string
- name: locale
dtype: string
splits:
- name: validation
num_bytes: 914973446
num_examples: 1630
- name: test
num_bytes: 1143277287
num_examples: 2012
- name: train
num_bytes: 6105554835
num_examples: 10545
download_size: 7045974123
dataset_size: 8163805568
configs:
- config_name: afaan_oromo
data_files:
- split: validation
path: afaan_oromo/validation-*
- split: test
path: afaan_oromo/test-*
- split: train
path: afaan_oromo/train-*
- config_name: all
data_files:
- split: train
path: all/train-*
- split: validation
path: all/validation-*
- split: test
path: all/test-*
- config_name: amharic
data_files:
- split: validation
path: amharic/validation-*
- split: test
path: amharic/test-*
- split: train
path: amharic/train-*
- config_name: sidama
data_files:
- split: validation
path: sidama/validation-*
- split: test
path: sidama/test-*
- split: train
path: sidama/train-*
- config_name: tigrinya
data_files:
- split: validation
path: tigrinya/validation-*
- split: test
path: tigrinya/test-*
- split: train
path: tigrinya/train-*
- config_name: wolaytta
data_files:
- split: validation
path: wolaytta/validation-*
- split: test
path: wolaytta/test-*
- split: train
path: wolaytta/train-*
license: cc-by-4.0
task_categories:
- automatic-speech-recognition
- audio-classification
language:
- am
- om
- ti
size_categories:
- 10K<n<100K
Ethio Speech Copus — Afrivoices Ethiopian
📌 Overview
The Ethio Speech Corpus dataset is a multilingual speech corpus containing audio–text pairs across five Ethiopian languages. It is designed to support the development of speech-to-text technologies for low-resource languages.
This dataset is part of the Afrivoices initiative — a collaborative effort to create a large-scale ASR dataset for African languages. The broader goal of the initiative is to collect 600 hours of speech data, including both: Image-prompted speech and Scripted speech.
Afrivoice Ethiopian is a joint project involving:
- Digital Umuganda — through the Open Dataset for All initiative
- Africa Next Voice Initiative — a consortium aimed at creating open, ethical, and community-driven African speech datasets.
🗣️ Languages
The dataset contains audio and transcriptions for several Ethiopian languages, including:
- Amharic
- Tigrinya
- Wolaytta
- Sidama
- Afaan Oromo
📦 Dataset Structure
Each entry includes:
| Field | Description |
|---|---|
| audio_id | Unique ID for the audio sample |
| audio | Audio file |
| audio_duration | Length of the audio sample in seconds |
| transcription | Human-generated transcription |
| speaker_id | Anonymized speaker identifier |
| gender | Speaker gender |
| age_group | Speaker age bracket |
| locale | Locale/language code |
Splits
- train
- validation
- test
🎯 Intended Use
The dataset is designed for:
- Automatic Speech Recognition (ASR) training
- Research on low-resource speech technologies
- Developing voice-enabled tools for Ethiopian languages
📥 Loading the Dataset
from datasets import load_dataset
dataset = load_dataset("badrex/ethiopian-speech-flat")
print(dataset["train"][0])
⚠️ Limitations & Ethical Use
- Data quantities vary by language
- Environmental noise may affect uniformity
📜 Citation
Please acknowledge the Afrivoice Ethiopian Initiative, Digital Umuganda, and the Africa Next Voice Initiative when using this dataset.