Browse Datasets

Common Voice (Swahili & Hausa)

Crowdsourced Read Speech for African Languages

asr speech-recognition accent-classification Swahili ~20GB; Hausa smaller CC0 1.0
Access Dataset

Description

Mozilla Common Voice scripted-speech recordings for Swahili (~20GB) and Hausa, with speaker demographics. Crowdsourced read speech released into the public domain (CC0). Stream from Hugging Face or pull a language subset to stay within Colab limits.

Quick Start

Python 
from datasets import load_dataset

# Swahili ("sw") or Hausa ("ha")
ds = load_dataset("mozilla-foundation/common_voice_17_0", "sw", split="train", streaming=True)
print(next(iter(ds)))

Project Ideas

Hausa ASR Fairness

Train a Hausa ASR model and measure accuracy gaps across self-declared gender and age.

Intermediate

Accent Classifier

Build a model that classifies speaker accent from the audio.

Intermediate

Ethical Considerations

  • Demographic gaps (e.g. few female Hausa speakers in some releases) degrade performance for under-represented groups.