Language	Category	Total number of hours	Total number of transcribed hours	Total number of clips	Total of Size of the dataset in GB
Amharic	Unscripted	351.17	81.67	71119	20.7
	Scripted	113.55	113.55	22412	7.9
	Expert	153.44	24.28	31144	8.1
Afaan Oromo	Unscripted	338.36	80.44	65555	21.8
	Scripted	111.93	111.93	23051	9.7
	Expert	153.7	29.88	29638	8.2
Sidama	Unscripted	348.67	74.94	72801	23.5
	Scripted	114.93	114.93	22717	10.1
	Expert	139.55	35.55	27920	7.5
Wolaytta	Unscripted	339.95	104.29	68671	19.4
	Scripted	108.83	108.83	25200	7.8
	Expert	154.36	5.83	33303	8
Tigrinya	Unscripted	348.86	97.73	69954	23.7
	Scripted	107.25	107.25	26624	10.8
	Expert	144.98	13.6	29830	7.6

Supported tasks

How to use

The datasets library allows you to load and pre-process your dataset in pure Python, at scale. The dataset can be downloaded and prepared in one call to your local drive by using the load_dataset function.

For Full Dataset

Run the following python script

from datasets import load_dataset
data = load_dataset("DigitalUmuganda/Afrivoice_Ethiopia")

For Language dataset

from datasets import load_dataset
data = load_dataset("DigitalUmuganda/Afrivoice_Ethiopia", name="am")

Dataset Structure

Data Instance

{ "voice_creator_id":"hqiqompEWaQa9CICqOwTYjyj3fc2", "transcription_creator_id":"1jMeIatBUBXgmHI1FDjtwIkW1E03", "image_filepath":"Hair-(ጸጉር)_WOLAYTTA_037.jpg", "image_category":"scripted", "image_sub_category":"Free", "category":"scripted", "audio_filepath":"00DioM00qNeWGYsLF7KG.wav", "transcription":null, "script":"Macca naati bantta huuphiyaa dumma dumma hanotan dadisoosona. Hegaappe guyyiyan bantta dadissido huuphiyaappe guyye baggaara qassi aybakkonne huuphiyaa bolla aattoosona.", "age_group":"25-35", "gender":"Female", "project_name":"Digital Umuganda", "locale":"wal_ET", "year":2025, "duration":15.96, "location":null, "uid":"hqiqompEWaQa9CICqOwTYjyj3fc2", "key":"00DioM00qNeWGYsLF7KG", "dir_path":"wal/train/scripted", "chunk_id":0, "workflow":"scripted", "prosodic varieties":"Free", "expert categories":null }

Data Fields

voice_creator_id (string): An id for which client (voice) made the recording
transcription (string): Original audio transcription with punctuation and capitalization,,only for unscripted workflow
script (string): original scripted created which was later read by the contributor, only applies for scripted workflow
image_filepath (string): file path of the image file inside the dataset
audio_filepath (string): file path of the audio file inside the dataset
age_group (string): age range of the audio recorder
gender (string): The gender of the speaker
location (string): geographical location of the audio recorder
duration (int): length in seconds of the audio file
image_category (string): domain of the image (eg: health, agriculture, finance), used as prompt during audio creation.
image_sub_category (string): Sub-domain label of the image (e.g., within agriculture: “seed farming” or “forestry”), used to guide audio creation.
project (string): project name
locale (string): The locale of the speaker
year (int): Year of recording
project_name (string): project name
location (string): location of the project
workflow (string): The workflow of the data collection it can either be scripted or unscripted
unscripted datatype (string): the various unscripted type image prompted, Spontaneous, Expert
prosodic varieties (string): the prosodic varieties used when recording the audio these can be: questions, emphatic, expressive, free, it applies for both scripted and unscripted
expert categories (string): the various categories used in the unscripted expert domain
key (string): key identifier of the data point
dir (string): directory path to the parent directory of the manifest file, and audio and image directory of data point.
chunk_id (int): chunk ID to locate manifest file it belongs to, and audio/image directories to find them

Data splits

Each domain/category in the dataset is divided into train, validation, and test splits, following the following rules: Users on the train split do not appear in the test or validation split. Same goes for test and validation splits.

Licensing Information

All datasets are licensed under the Creative Commons license (CC-BY-4).

Downloads last month: 176