The proof · live data
One spelling. Many meanings. Already recorded.
Real, consented native audio. Each homograph is recorded per meaning as a
Bantu↔English code-switch, and each word across 5 rounds (reproducibility).
Every take is time-segmented at switch_ms → it self-labels into the
6-in-1. Full numbers on
what we have.
The homograph corpus opens at evaluation
Publicly, hear the problem itself on the 6-in-1 page — the ulebomba demo. Start an evaluation to browse every homograph across 6 languages: each meaning, the 5 rounds, and playable audio.
Start an evaluation →Meanwhile, see what we have — public, live coverage & roadmap.