Products
From voice to a benchmarked model
Four complementary products connect data, voice, evaluation, and essential services in West African languages.
Compare
Which one do you need?
The four products share the same language-quality standard, but solve different needs. Here is what actually separates them.
| Product | SpeakYodi | CollectData Hub | MeasureLomBench | SupportHLlama |
|---|---|---|---|---|
| Who it is for | App builders, call centers, public services, and media who want to speak their users' language. | Companies and research labs that need reliable African language data, not unverified raw scraps. | Research labs, model builders, and public institutions who want to verify a claim before deploying. | Ministries, health providers, civil-registration bodies, telecom operators, NGOs, and inclusion programmes that need locally understandable, reachable services. |
| Languages | Ewe ↔ French ↔ English available now. Kabiyè ↔ Ewe ↔ French ↔ English in progress. | 12 West African languages, continuously extended through contributors. | Evaluation available across every language in the Eyaa-Tom corpus. | Ewe, Mina, and Kabiyè, with gradual expansion based on language quality and partner needs. |
| What is measured | 90%+ · accuracy measured on tested words | 70+ · active contributors and moderators | 12 · Togolese languages and lingua francas evaluated | 3 · local languages in the initial scope |
| Learn more | Learn more | Learn more | Learn more |
What sets us apart
Large models skip African languages for lack of data. We chose to build that data ourselves, with the communities who speak these languages, instead of waiting for it to exist.
Data validated by native speakers
Every audio recording and every text passes through a community of native contributors and moderators before it trains a model. Contributors get paid for accepted work, not just asked for it.
Languages no one else covers
Ewe, Kabiyè, and ten other Togolese languages, extending into Ghana, Benin, and Nigeria. Fieldwork the big labs never did.
Published methodology, not a black box
Our datasets, models, and Lom evaluation metric are published and documented, presented at AfricaNLP. Anyone can check what we claim.
Locally rooted, not just passing through
Partnerships with Togo AI Lab, universities, and local authorities. Models are built with the institutions that will use them, not shipped in from outside.
Something that doesn't fit a box?
AI and robotics training, robot integration (Pepper, Niryo One, NAO), a joint research project: let's talk about what you're trying to solve.