Your frontier model fails the “wahala” test.
Nigerian Pidgin is the lingua franca of 75M West Africans and ranks 0 on every frontier model's eval card. Below is a transcript preview of what a properly-sourced Pidgin ASR + RLHF training set will look like — real Nigerian speakers, all 6 Nigerian zones, scripted + spontaneous mix. Our first 10-hour pilot batch ships to the first labs that commit.
Pilot status: the 10-hour batch is in production against confirmed buyer briefs — we haven't yet posted a free public bundle. Sample clips below are transcript-only previews. Use the “Scope a paid pilot” CTA to commit; we ship the .wav + JSONL bundle once your brief is signed.
Pilot batch target
10 hours
Pidgin · scripted + spontaneous mix · first drop on buyer commit
Speaker panel
Real Nigerians
Recruit from our existing testing panel · all 6 NG zones · age-stratified
Format
16-kHz mono .wav
+ JSONL manifest with prompt, duration, anonymised contributor id
Sourcing
Naira-paid, consent first
AI-reuse consent recorded before anyone records · real phones, real networks · no emulators, no VPNs
We don't serve the .wav files from this page — they ship in the full request bundle to keep cold-traffic bandwidth sane. Each clip below shows the Pidgin transcript and a parallel English gloss for evaluators who don't read Pidgin directly.
Pidgin transcript
“Today na Monday so the market dey loaded well-well. Customer plenty pass last week.”
English gloss
Today is Monday so the market is very busy. There are far more customers than last week.
Pidgin transcript
“I transfer the money for OPay, but e never enter the person account.”
English gloss
I transferred the money via OPay, but it hasn't arrived in the recipient's account.
Pidgin transcript
“My pikin school don close for term break, so we go visit family for village this weekend.”
English gloss
My child's school has closed for term break, so we'll visit family in the village this weekend.
Pidgin transcript
“Traffic for third mainland bridge no get part two — I waka two hour today.”
English gloss
The traffic on the Third Mainland Bridge is unbelievable — I was stuck for two hours today.
Pidgin transcript
“Doctor talk say make I drink the medicine three times for day, after I chop food.”
English gloss
The doctor said I should take the medicine three times a day, after eating.
{
"clip_id": "clip-001",
"audio_path": "audio/clip-001.wav",
"duration_seconds": 12.3,
"language": "pidgin",
"transcript": "Today na Monday so the market dey loaded well-well...",
"transcript_english": "Today is Monday so the market is very busy...",
"style": "spontaneous", // or "read"
"topic": "commerce",
"speaker": {
"anonymised_id": "c08af3...",
"gender": "F",
"age_range": "25-34",
"state": "Lagos",
"occupation": "trader"
},
"recording": {
"sample_rate_hz": 16000,
"channels": 1,
"codec": "pcm_s16le"
},
"license": "9jatesters-standard-data-licence-v1"
}Compatible with Common Voice schema after a 6-line key remap. We ship the remapper script with the full drop on request.
| Feature | 9jatesters | Common Voice | NaijaVoices | Scale / Surge |
|---|---|---|---|---|
| Languages covered | Pidgin · Yoruba · Igbo · Hausa · Nigerian English | Limited Yoruba/Igbo only | Yoruba · Igbo · Hausa (no Pidgin) | English + a few major non-African |
| Speaker sourcing | Real Nigerians on real phones — no emulators, no VPNs | Self-declared | Demographic survey | Email + IP |
| Contributor pay | Paid in naira to verified Nigerian bank accounts — withdrawals settle in 3 to 5 business days + transparent platform spread (funds QA) | Unpaid contributions | Grant-funded · capped pay | |
| Pidgin RLHF preference data | ||||
| Standard data licence | CC0 (no commercial protection) | Varies per dataset | ||
| Pilot delivery target | Negotiated per brief · typically 2-3 weeks for 5+ audio-hours | N/A (download anytime) | N/A | Custom |
Bootstrap or fine-tune a Pidgin-aware speech-to-text. Bring scripted + spontaneous mix for robustness.
Lock a single identity-consistent speaker, 100+ scripted utterances, one consistent human voice across the dataset.
Bilingual annotators rate model outputs in Pidgin/English code-switching for SFT/DPO data.
Sentence-aligned English ↔ Pidgin pairs with regional-variant tagging — TMX/JSONL/CSV.
Every clip comes from a real Nigerian on a real phone and network — no emulators, no VPNs, no synthetic personas. Contributors are recruited from our existing testing panel, consent to AI-data reuse before they record, and the dataset ships under a standard licence with a contributor-rights schedule.
They're paid in naira to verified Nigerian bank accounts — withdrawals settle in 3 to 5 business days. The data is self-hosted on hardware we own and control — not a third-party cloud — with an NDPA-aligned privacy posture, and a transparent platform spread funds QA and AI-summarised review. Honest provenance, no opaque proxy chain.
See /trust for the full security + compliance posture and /trust/data-residency for the sub-processor list.
Tell us your role + intended use. We'll respond within 24 hours with a brief proposal and a pricing scope for the first batch (ASR $30/hr, TTS $100/hr, RLHF $120/annotator-hr, red-team $120/hr — full rate card on /ai-data). Delivery follows once the brief is signed.
Standard NDA available on request · region-pinned (EU-only) delivery available for Enterprise · custom licence terms negotiable for academic vs commercial use.
Future versions of this preview will also be published on HuggingFace for direct audio playback + parquet download — email [email protected] to be notified when the drop goes live.