BantuNomics My BantuNomics
B BANTUNOMICS Stories
The connected-speech corpus

Page-length natural speech, in three modes, for every Bantu language.

Words and syllables are units. Stories are the only category where a model hears language connected — a whole page, read or spoken by a native voice. We capture every language on the same three modes, so the structure is legible before you download a single file.

01
Bantu only

Page-length native Bantu read aloud — gold-aligned connected speech and TTS voice.

02
Bantu & English

Interleaved Bantu↔English exactly as people speak — a labeled code-switch boundary in every take.

03
English only (accented)

The same neutral English text read by Bantu L1 speakers — the accent is the asset; frontier ASR trips on it.

3Languages
193Consented takes
218.2Minutes of speech
5Native speakers
The proof point

Frontier ASR trips on the accent.

Mode 3 reads the same neutral English text across many Bantu L1 speakers — a matched benchmark. We ran a frontier speech-to-text model over it and scored word-error rate against the known text. It mishears Bantu-accented English at a measurable rate. That gap is the asset.

Same text, different mouths

Content is held constant; accent is the only variable — so the WER is comparable across languages.

Mostly substitutions

The model hears the audio and picks the wrong word — a measured accent failure mode, not silence.

Bantu-accented English · Bemba
9.0%Word-error rate
91.0% word accuracy 11575 reference words scored (20 takes) Model: google-cloud-stt en-US/latest_long
Hear a public sample
The Morning at Home
Bemba-accented English · 214.6s · 449 words · ASR WER 9.6%

The value model

One recording does the work of six datasets.

A single consented Bantu–English story take is not one labelled example. Captured in one continuous take, its switch point is logged as switch_ms — so the record self-segments with no manual annotation, and one take becomes the ground truth for six things a lab would otherwise buy from six separate vendors.

Consented Take 48kHz Mono

Hear it proven live — a real consented take, and the datasets it yields: accent benchmark · the full 6-in-1.

License the connected-speech corpus, not a dataset.

Browse

Start with a language.

Every language is presented on the same three-mode frame. Empty modes stay visible as the next thing to record.

License the connected-speech corpus.

All three modes, every take, raw audio, and the reproduce-the-WER eval set — consented, de-identified, and growing. Evaluate it on your own models, then license the catalogue.