Market conversation, Nigerian Pidgin
Natural two-speaker conversations between traders and regular customers, recorded on clip-on mics.
Samples are blurred and watermarked, 30 to 60 seconds. Request the full 10-hour sample on the right.
Consented conversations in Lagos markets: pricing, haggling, small talk. Each speaker consents separately; recorded on two clip-on mics. Transcribed, time-aligned layers available.
- WAV 48 kHz / 16-bit
- JSON metadata per clip
- Transcripts (JSON, time-aligned) when labelled
Trimmed to approved minutes only, with a review score on every clip.
21 keypoints per hand per frame, with a confidence score.
Start and end times for each step of the task, in plain verbs.
A short description of what is happening, every few seconds.
Verbatim transcript per speaker, in the language spoken.
| Field | Type | What it holds |
|---|---|---|
| clip_id | string | Stable clip ID |
| duration_s | number | Clip length in seconds |
| trade | string | Trade slug |
| contributor_id | string | Pseudonymous contributor ID |
| consent_id | string | Consent record ID |
| consent_version | string | Consent text version |
| capture_date | date | Capture date (WAT) |
| device_model | string | Phone model |
| qa_score | number | 0–1 review score |
| language | string | Language code (en, pcm, yo, ig, ha) |
| speakers | number | Number of consented speakers |
Non-exclusive licence for AI training. Every contributor is ID-checked and consented on video in their own language; bystander faces are blurred; every minute is reviewed by a person.
Related datasets
All datasets
Read-aloud prompts, five Nigerian languages
Short scripted clips read by native speakers in English, Pidgin, Yorùbá, Igbo and Hausa.
Tailoring, first-person
Head-mounted video of Lagos tailors measuring, cutting and sewing in their own shops.
Auto mechanic, first-person
Mechanics diagnosing and repairing cars and motorcycles in roadside and workshop settings.