Maloba in LoBAI
The Maloba models are presented in conversation in LoBAI, the Congo Digital Services interface, and through an OpenAI-compatible API. The access point is being commissioned.
News
The project journal
What the team has done, measured and published. One article every two weeks, with its sources.
Three validators, 310 samples
What human validation says about our data: meaning is preserved in 96.5 % of cases, but validators agree little on what counts as “correct” or “partial”.
MMS learns to write ɔ and ɛ
Trained in 1 h 50 on 0.23 % of its parameters, a language adapter for MMS reaches the same word accuracy as Whisper, with lighter errors at character level.
The Maloba speech corpus, version 3.2
58 hours of Lingala speech, 18,767 transcribed segments, harmonised spelling and leak-free splits: the reference version of the speech corpus is published.
Measuring honestly
Why we redid our measurements, and what changed: a leak-free test set, a reproducible evaluation suite, and figures that are less flattering but accurate.
A French → Lingala translation model
NLLB-Maloba-h1 gains 8.6 BLEU points from 288 human-translated pairs. Adding 4,700 machine-produced pairs changed nothing.