Datasets:

Kencorpus
/

KenCorpus_audio

language:
  - sw
  - luo
  - bxk
  - lri
  - rag
license: cc-by-4.0
task_categories:
  - automatic-speech-recognition
tags:
  - swahili
  - kiswahili
  - dholuo
  - lubukusu
  - lumarachi
  - lulogooli
  - kenyan-languages
  - low-resource-languages
  - african-languages
  - speech-corpus
  - spontaneous-speech
pretty_name: KenCorpus Audio
size_categories:
  - n<1K

KenCorpus Audio: A Kenyan Multilingual Spontaneous Speech Corpus

Dataset Description

KenCorpus Audio is a multilingual spontaneous speech corpus for Kenyan languages. The recordings capture natural, unscripted speech from native speakers in their communities across diverse topics including culture, agriculture, commerce, health, and daily life.

Three languages were selected: Kiswahili, Luhya (dialects: Lumarachi, Logooli, Lubukusu), and Dholuo.

Dataset Statistics

Language	Files	Duration
Dholuo	512	99h 3m 8s
Lubukusu	354	30h 11m
Swahili	104	19h 10m 57s
Lumarachi	138	15h 37m 46s
Logooli	44	12h 26m 55s
Total	1,152	176h 29m 46s

Data Fields

Column	Type	Description
audio	Audio	Embedded audio waveform (decoded on load)
language	string	Language (Swahili, Dholuo, Lubukusu, Lumarachi, Logooli, Luhya_Bukusu, etc.)
genre	string	Genre (Culture, Agriculture, Commerce, Health, etc.)
source	string	Data source (Community, etc.)
title	string	Topic/title of the recording
story_id	string	Unique identifier matching the KenCorpus metadata
duration_seconds	int	Duration in seconds (from metadata)

Example Record

{
    'audio': {'path': '0001_swa.aac', 'array': array([...]), 'sampling_rate': 44100},
    'language': 'Swahili',
    'genre': 'Culture',
    'source': 'Community',
    'title': 'Masharti ya ndoa',
    'story_id': '1',
    'duration_seconds': 1303
}

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("Kencorpus/KenCorpus_audio")

# Access a sample
sample = dataset['train'][0]
print(sample['language'])
print(sample['title'])
print(sample['audio']['sampling_rate'])
print(sample['audio']['array'][:10])

# Filter by language
dholuo = dataset['train'].filter(lambda x: x['language'] == 'Dholuo')
print(f"Dholuo recordings: {len(dholuo)}")

Genres

Culture, Agriculture, Commerce, Health, Religion, News, Song, Story, and more.

Data Collection

Spontaneous speech recordings were collected from native speakers in their respective language communities. Recordings capture natural, unscripted conversations and narratives.

Dataset Curators

Kiswahili: Rose Felynix, Khalid Kitito, Dr. Benard Okal Dholuo: Jotham Ondu Ajiki, Dr. Jackline Okello, Jonathan Muga, Mercy Lavinca Oduoll Luhya (Logooli): Salano Odari, Dr. Phillip Lumwamu Luhya (Bukusu): Mactilda Nekesa Makana, Mulwale Martin Luhya (Marachi): Yonah Weunda

Citation

@article{wanjawa2022kencorpus,
  title={Kencorpus: A Kenyan Language Corpus of Swahili, Dholuo and Luhya for Natural Language Processing Tasks},
  author={Wanjawa, Barack W. and Wanzare, Lilian D. and Indede, Florence and McOnyango, Owen and Ombui, Edward and Muchemi, Lawrence},
  journal={arXiv preprint arXiv:2208.12081},
  year={2022}
}

License

CC-BY-4.0

Acknowledgments

Part of the Kencorpus project for low-resource Kenyan language NLP resources.