For Data Buyers and Customers
Consented, documented speech data in African languages, under a licence that holds up to due diligence.
Who This Is For
AI labs and speech-technology companies building African-language models. Telecoms, banks and fintechs building voice services. Research groups and public-interest projects.
No dataset has been released yet. The first release is planned in a twelve-week pilot, co-designed with one design-partner buyer.
What a Dataset Looks Like
Each release covers one language. It is versioned and documented.
Every item is transcribed, and transcripts are reviewed by native speakers. Audio is FLAC at 16 kHz, mono.
Every item carries the consent choice that was valid on the release date. The datasheet says how the data was collected, who is in it, what it may be used for and its known gaps.
Test sets use speakers who appear in no training split. A withdrawn item is listed as removed in the next changelog.
Speakers appear only as pseudonymous ids. No names, phone numbers, device ids, precise locations or contacts are in any dataset.
Licence Products
Dataset prices are planned to vary with transcription level and language scarcity, with research and commercial rates.
Price levels are to be decided. They will be benchmarked with the design partner before anything is published.
Licence Terms
The planned terms protect members and keep the library open to the community.
Non-exclusive licences are one of the voice network's sovereignty indicators.
Whether voice cloning or text-to-speech training is ever allowed is to be decided. The plan: not under any of the four choices.
Start a Pilot
In the planned pilot, one design-partner buyer co-designs the dataset spec and the licence before collection starts. The first dataset has a customer on day one.
The pilot language is to be decided by buyer demand and member reach. Candidates are Swahili, Hausa or Yoruba.
To discuss a pilot, contact the team.