9jatesters

Nigerian language data, collected from Nigerians who were paid for it and asked first.

We are Ranked Technologies Limited, a Nigerian company, RC 9522220, registered with the Nigeria Data Protection Commission as a data controller, NDPC/DCP/14538. We trade as 9jatesters.

---

What is here

9jaVoice, read-aloud and spontaneous speech in Nigerian languages, one dataset per language, under CC BY-NC 4.0.

Every clip carries the id of the consent record it rides on. A contributor ticked an optional box, unticked by default, that was never a condition of doing the work or being paid. We hold a great deal more Nigerian speech than we have published. What is here is what cleared that box.

A read-aloud clip contains two people's work: the voice and the sentence. Most of our lines were written by other contributors, so we ask the writer as well as the speaker, and a clip stays out unless both agreed. That rule roughly halved the first release. We think it is the right trade.

What we will not do

We will not publish work under a licence broader than what people were asked. Our contributors agreed to a free, open dataset that anyone can download and keep at no charge. They were not asked whether a stranger could build a business on it, so these datasets are non-commercial. A Creative Commons licence cannot be revoked once someone holds a copy, which is exactly why the narrower one is the honest one.

If you want this data commercially, that is a real conversation and we can license it separately.

What we publish about ourselves

9jaBench is our benchmark for Nigerian-language models. We publish our own model's losses on it, including the benchmarks where better-funded labs beat us. A leaderboard that only shows its owner winning is marketing, not measurement.

NGPT is the Nigerian-language model we train, and the reason this data exists.

Contact

support@9jatesters.com  or  support@ranked.ng

Write to license data commercially, to ask how something was collected, to withdraw your consent if you are a contributor, or to tell us something we have published is wrong. The last one is welcome and we will correct it in public.