BantuNomics My BantuNomics

What we have — and what you can use today

A living, continually growing corpus of tonal Bantu homographs, recorded as consented Bantu↔English code-switch. This page is read live from the platform, so the numbers grow as contributors record. ● live

5/12
languages live
2439
consented takes
969
meanings (senses)
422
homograph words
5
rounds / word

By language

LanguageStatusHomographsSensesWord items Sentence itemsRoundsTakesSpeakers
🇿🇲 Bemba (bem) LIVE 244574411 2865 15571
🇿🇦 Xhosa (xho) LIVE 296262 625 3421
🇸🇿 Swati (ssw) LIVE 204646 465 2811
🇰🇪 Swahili (swh) LIVE 102424 245 1441
🇷🇼 Kinyarwanda (kin) LIVE 102020 205 1151
🇿🇼 Northern Ndebele (nde) PLANNED 102222 225 00
🇿🇦 Northern Sotho (nso) PLANNED 306666 665 00
🇿🇲 Chewa (nya) PLANNED 102121 215 00
🇿🇼 Shona (sna) PLANNED 102222 225 00
🇿🇼 Tsonga (tso) PLANNED 102929 295 00
🇿🇼 Venda (ven) PLANNED 102424 245 00
🇿🇦 Zulu (zul) PLANNED 295959 595 00

“Takes” are consented, de-identified recordings (rejected/flushed excluded). Word items are recorded across 5 rounds each for reproducibility; sentence items once.

The 6-in-1 — what each take yields, and what's ready

Bantu ASR live
Code-switch ASR live
Bantu↔English pair live
TTS / voice live
Code-switch boundary live
Tone / F0 roadmap

5 of 6 datasets are usable today from every consented take; Tone/F0 activates per language as its forced-alignment model ships. See it in action on the 6-in-1 page.

Roadmap

Live now: bem, xho, ssw, swh, kin · Recording next / planned: nde, nso, nya, sna, tso, ven, zul · Tone/F0: activates per language as its forced-alignment model ships.

Full corpus (all rounds, all languages, raw audio, bulk) is licensed — start an evaluation. Machine surface: GET /api/v1/tone/stats.