One recording, six datasets.
A single consented Bantu→English code-switch noun take — singular and plural — is not one labelled example. Its switch point is logged as switch_ms and it self-segments, becoming ground truth for six things labs otherwise buy from six vendors. The native aligns the word to the meaning, then records the truth — that recording IS the verification. And every take ships with its metadata: FSI syllabification, the full noun equation (class · prefix · composed→surface · concord), the singular/plural class pairing, and the segments boundary — across the whole family.
Code-switch ASR
Natural Bantu→English switching with the exact switch point pre-labelled.
Bantu-accented English
The English half spoken by a native Bantu speaker — accent-robustness data.
FSI alignment
The noun decomposes into its Full Syllable Inventory — the units a forced aligner needs.
Pronunciation & TTS
A consented native pronunciation keyed to meaning and class — a clean synthesis basis.
Singular↔plural class truth
Both class-inflected forms, so the take teaches the noun-class pairing, not just a word.
Aligned concept truth
The noun is anchored to a concept (ref_id) — aligned to the same word in every language.