Yoruba Speech Dataset
Yoruba speech recordings with matching transcriptions, covering everyday, health, finance and agriculture topics.
Considerations
Dataset status
This dataset is under active development and not yet final. Releases, corrections, and updates will continue through November 2025. Users should regularly check this repository for the latest versions and documentation.
For attribution, benchmarking, or publications:
- Always cite the latest available version (as of November 2025 or later).
- Do not use or benchmark against older or incomplete releases.
- Reference the current documentation provided in this repository.
We appreciate your patience and collaboration as we continue to expand and refine this resource.
Splits
| Split | Samples | Audio | Speakers | Size |
|---|---|---|---|---|
| Train | 147,686 | 435.0 hrs | 607 | 204.0 GB |
| Dev | 7,717 | 23.0 hrs | 35 | 12.0 GB |
| Dev test | 8,882 | 25.9 hrs | 36 | 12.6 GB |
| Total | 164,285 | 483.8 hrs | 678 | 228.6 GB |
Breakdown
Gender Distribution
Male Voicings
59,626
36.3% of total
Unique Speakers
291
Female Voicings
104,659
63.7% of total
Unique Speakers
400
Balance
Domain Distribution
Everyday Conversation
83,947
51.1% of total
Health Conversation
42,575
25.9% of total
Finance (Business and Money Conversation)
34,764
21.2% of total
Agriculture Conversation
2,999
1.8% of total