DigitalUmuganda/Afrivoice_Kinyarwanda

Dataset Card for the image text and voice dataset

Dataset Description

Each datapoint in this dataset consists of a JPEG image, a corresponding audio Webm file describing the image, and when available, the transcription of the audio file.

Domain	Total Hours	Transcribed Hours	Number of Clips	Dataset Size (GB)
Agriculture	467.13	465.40	86,305	30.13
Health	994.32	992.87	179,219	58.53
Finance	564.21	563.11	103,159	38.55
Government	676.10	674.11	122,265	49.22
Education	198.49	197.52	37,377	13.70
Total	2900.24	2893.01	528,668	190.13

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 in two ways:

Direct download of the whole dataset to a local directory

from huggingface_hub import snapshot_download                                                   

snapshot_download(repo_id="DigitalUmuganda/Afrivoice_Kinyarwanda",repo_type='dataset',local_dir='<destination_dir>')

Or, using load_dataset to download a particular domain

from datasets import load_dataset

data = load_dataset("DigitalUmuganda/Afrivoice_Kinyarwanda",name='health')

Dataset Structure

Data Instances

{'creator_id': 'bcSXMYbErjM6pJyAwwLs7NAxA9v2',
 'raw_text': 'Ingagi ihagaze yonyine. Ingagi ni nziza cyane, kuko zikurura ba mukerarugendo bakazana amadovize mu Gihugu cyacu.',
 'duration': 15.06,
 'LUFS': -25.5,
 'image_category': 'Agriculture',
 'image_sub_category': 'Wild Animals',
 'text': 'ingagi ihagaze yonyine ingagi ni nziza cyane kuko zikurura ba mukerarugendo bakazana amadovize mu gihugu cyacu',
 'audio_filepath': 'audio_1751479904-bcSXMYbErjM6pJyAwwLs7NAxA9v2.webm',
 'image_filepath': 'restyf.jpg',
 'age_group': '50+',
 'gender': 'Male',
 'location': 'Musanze',
 'shard_id': 0,
 'image_shard_id': 0}

Data Fields

creator (string): An id for which client (voice) made the recording

raw_text (string): Original audio transcription with punctuation and capitalization

image_filepath (string): name of the image file inside the shard

audio_filepath (string): name of the audio file inside the shard

text (string): normalized audio transcription (i.e: without punctuation and capitalization)

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.

shard_id (int): index of the shard containing the audio file in the audio_filepath column.

image_shard_id (int): index of the shard containing the image in the image_filepath column.

LUFS (int): Loudness of the audio in Loudness Units relative to Full Scale (LUFS). Lower values indicate quieter audio, while higher values indicate louder audio.

Data Splits

Each domain in the dataset is divided into train, validation, and test splits.

Licensing Information

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