01
The product
LomBench applies the Lom metric, developed by our research team, to objectively evaluate speech recognition, translation, and synthesis models on West African languages. It quantifies a language's actual "AI-readiness", beyond marketing claims.
02
The product story
Measure instead of promise
Announcing that a model “supports” an African language costs nothing. LomBench replaces the announcement with a number: the Lom metric quantifies how ready a language actually is for AI, task by task.
What the evaluation covers
The evaluation runs on the Eyaa-Tom corpus and covers the tasks that matter in real use.
- Speech recognition — does the model understand what was said?
- Translation — does the meaning survive the move between languages?
- Speech synthesis — is the generated voice intelligible to a native speaker?
What goes unmeasured is always, in the end, overstated.LomBench design principle
Reproducible, and published
The Lom metric and the Eyaa-Tom corpus were presented at AfricaNLP 2026 and cover twelve Togolese languages and lingua francas. The protocol is published: a third party can rerun the measurement and land on the same score.
03
Contribute
What a contributor does, step by step.
Choose a model
Yours, or a public model you want to compare.
Select languages
One or more languages from the Eyaa-Tom corpus, depending on your evaluation need.
Run the benchmark
LomBench runs the test suite and computes the corresponding Lom score.
Compare results
Review the ranking and detailed metrics by task and language.
04
Become a client
How data access works, from request to delivery.
Describe your model
Architecture, target languages, intended use case.
We configure the evaluation
Choice of languages, tasks, and test protocol with our team.
Report and recommendations
You receive a detailed Lom score and concrete improvement paths.