DigitalUmuganda/AfriVoice

Dataset Card for the image text and voice dataset

Dataset Description

Dataset Summary

This dataset consists of a unique JPEG image, a corresponding audio WAV file describing the image, and when available, the transcription of the audio file. The Shona dataset has a total of 574.16 hours of audio; out of which, 100 hours have transcriptions and the remaining 474.16 hours do not have a corresponding transcribed text. For Lingala, the dataset is 517.13 hours long, with 100.98 hours transcribed and 416.15 hours with no transcriptions.

Languages

Shona, Lingala

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.

To download the config, specify the language code (i.e., "sn" for shona, and "ln" for lingala):

from datasets import load_dataset
data = load_dataset("DigitalUmuganda/AfriVoice", "sn")

Dataset Structure

Data Instances

{'creator': 'digital_umuganda',
 'project_name': 'shona_data_collection',
 'speaker_id': '2Eud8lyLlsMcciYhmlkwVRtBwi82',
 'audio_path': '/root/.cache/huggingface/datasets/downloads/extracted/9347eb035e3ae38aaf793efa152ba1c93a4336471afce2bbd00ac8c0f67e9066/small_data/audio/I7L1YJVKIRL4.wav',
 'image_path': '/root/.cache/huggingface/datasets/downloads/extracted/9347eb035e3ae38aaf793efa152ba1c93a4336471afce2bbd00ac8c0f67e9066/small_data/image/I7L1YJVKIRL4.jpeg',
 'transcription': 'Varume vaviri vari kukandirana bhora. Varume ava vakapfeka zvipika zvine ruvara rutema neruchena. Zvikabudura zvine ruvara rutema. Bhora ravanokandirana rine ruvara rweyero neruchena nerwebhuruu. Vari kutambira munhandare ine ivhu. Kumashure kwavo kwakagara vanhu.',
 'locale': 'sn_ZW',
 'gender': 'Female',
 'age': ' ',
 'year': '2023'}

Data Fields

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

image_path (string): The path to the audio file

path_audio (string): The path to the image file

transcription (string): The sentence the user was prompted to speak

age (string): The age of the speaker

gender (string): The gender of the speaker

project_name (string): Name of the project

locale (string): The locale of the speaker

year (string): Year of recording

Data Splits

Currently to data not yet split ie to access you must precise the train option, however the dataset will be split into train, dev, and test at some point in the future.