Dataset · naijavoices
1,800-hour Nigerian live speech audio corpus
naijavoices/naijavoices-dataset
Large dataset with speech recordings from over 5,000 speakers and text transcripts exceeding 500 GB.
What's inside
The dataset contains 1,800 hours of live speech from over 5,000 different speakers along with the text modality. The total volume of recordings ranges from 1 to 10 million items. The original version takes up more than 500 GB, but the authors also prepared an 84 GB compressed variant with the same data.
What you can do with it
- Training and fine-tuning automatic speech recognition (ASR) models accounting for Nigerian accents
- Developing text-to-speech systems with diverse voices
- Size
- 1M–10M rows
- Downloads this month
- 2.4K
- Likes
- 38
- Access
- on request on Hugging Face
Models trained on it
- Shinzmann/sorottsText → speech 191
- ogunlao/SBPN_multilingual_baseAutomatic speech recognition 171
- ogunlao/SBPN_multilingual_largeAutomatic speech recognition 68
- theelvace/whisper-small-igbo-25kAutomatic speech recognition 25
- samuelolubukun/moonshine-tiny-yoruba-v2Automatic speech recognition 14
- samuelolubukun/moonshine-tiny-igbo-v2Automatic speech recognition 12
- samuelolubukun/moonshine-tiny-hausa-v2Automatic speech recognition 8
- Alvin-Nahabwe/asr-africa-w2v-bert-2.0-naijavoices-yoruba-500hr-v0Automatic speech recognition 0
How to load it
from datasets import load_dataset
ds = load_dataset("naijavoices/naijavoices-dataset")