naijavoices/naijavoices-dataset

Important Information: please be aware that this version of the dataset is huge (500+ GB) and can therefore be challenging to use. To alleviate this, we’ve provided a compressed version (84GB) to facilitate quicker access here. We suggest using that instead if you have compute/storage constraints. They are both the exact same data.

Introduction

Welcome to the NaijaVoices dataset. The NaijaVoices dataset consisted of 1,500 hours of authentic speech (from over 5,000 diverse speakers!) and expert curated text in Igbo, Hausa, and Yoruba. 500 hours for each of the three languages. It also boasts of adequate female representation and balanced age-range distribution (young to old speakers). For more about the dataset info visit our website: https://naijavoices.com/. You must read and accept our data usage guidelines before using the NaijaVoices dataset.

Metadata

Each row contains the audio data, the transcript, the language of the transcript, the speaker id, along with their gender and age-range. Use the “Dataset Viewer” to look at the dataset.

Usage

To ease usage, we divided the dataset into batches. Specifically, we batched the dataset into their languages, and further into three uniform batches for each language. Each batch should contain approximately 167 hours. We obtained the following batches:

1. igbo-batch-0
2. igbo-batch-1
3. igbo-batch-2
4. yoruba-batch-0
5. yoruba-batch-1
6. yoruba-batch-2
7. hausa-batch-0
8. hausa-batch-1
9. hausa-batch-2

To load a specific batch run (e.g. “igbo-batch-1”:

from datasets import load_dataset

# If the dataset is gated/private, make sure you have run "huggingface-cli login"
dataset = load_dataset("naijavoices/naijavoices-dataset","igbo-batch-1")

If you want to load the whole dataset, then just do

dataset = load_dataset("naijavoices/naijavoices-dataset")

It is advisable to load the batch in streaming mode by setting streaming=True

# Streaming mode
dataset = load_dataset("naijavoices/naijavoices-dataset", streaming=True)

License

Our dataset is licensed under the CC BY-NC-SA 4.0 license. Basically, It's free for non-commercial, research use, with proper credit to the community. For commercial interests, please check out our membership licenses.