Browse Datasets

African Next Voices

Multilingual South African Speech (3000+ hrs)

asr speech-recognition 3,000+ hours CC BY 4.0
Access Dataset Read Paper

Description

A large multilingual speech dataset for seven South African languages, combining scripted and unscripted speech across agriculture, health, and general-purpose domains. Gated on Hugging Face (accept terms first). Very large; stream or use per-language subsets.

Quick Start

Python 
from datasets import load_dataset

# Accept the dataset terms on Hugging Face first
ds = load_dataset("dsfsi-anv/za-african-next-voices", split="train", streaming=True)
print(next(iter(ds)))

Project Ideas

isiZulu Whisper

Fine-tune Whisper on a 10-hour isiZulu subset to adapt to local pronunciation.

Advanced

Scripted vs Unscripted

Compare ASR accuracy on scripted vs unscripted speech.

Intermediate

Ethical Considerations

  • Dataset terms prohibit text-to-speech, voice cloning, or voice synthesis use.
  • Community-collected data; respect consent and avoid biometric profiling.