Teach your AI
how Ghana really is.
Labelled pictures and text, voices in Ghanaian languages, translations and human ratings, made by real people across the country and checked by our team before you pay for a single answer.
- Up to 3 people per item
- Agreement scores per item
- Yours to train on
What we make
Data your model can't scrape
Most of what's online about Ghana is thin, old or in the wrong language. We collect it fresh, from the people who live it.
Image labelling
Upload up to 500 pictures and set the labels. People pick one for each: pothole or clear road, in stock or empty shelf, ripe or not.
Text labelling
Paste your texts and choose the categories: sentiment, topic, complaint or praise, spam or not. Each one is tagged by real readers.
Speech recordings
Your sentences read aloud in Twi, Ga, Ewe, Hausa and more, on everyday phones, in the accents your users really have.
Parallel text
English sentences translated into Ghanaian languages by people who speak them, with up to three versions of each.
Fresh photos
New pictures taken to your brief: shop shelves, price tags, signboards, food, packaging. Taken for your job, not scraped from the web.
Human ratings
Show people your model's answers and ask them to rate or choose between them, so you know how it lands with Ghanaian users.
What labellers see
Small tasks, done with care
A few items at a time keeps people focused, and every answer is checked before it reaches you.
Real screens from the Fairfield app, shown with a demo job.
Labels you can trust
Every answer, in a file your pipeline can read
No screenshots of spreadsheets or PDFs to retype. You download a clean CSV, built straight from the approved answers.
- One row per answer, with the item, the label and an anonymous respondent ID
- top_label and agreement columns, so you can keep only the items people agreed on
- Links to every picture and recording, ready to fetch
- Only approved answers are included: rejected work never reaches you
item_no,item_image,label,respondent,top_label,agreement
1,fairfieldgh.com/files/road_0001.jpg,pothole,3fa91c02e7,pothole,2/2
1,fairfieldgh.com/files/road_0001.jpg,pothole,b7d2410a9c,pothole,2/2
2,fairfieldgh.com/files/road_0002.jpg,no pothole,0c55e8b31d,no pothole,2/2
3,fairfieldgh.com/files/road_0003.jpg,pothole,e19a7f4c20,pothole,3/3
4,fairfieldgh.com/files/road_0004.jpg,no pothole,5b03d9e6aa,no pothole,2/2
5,fairfieldgh.com/files/road_0005.jpg,pothole,9d7c1e2f48,pothole,2/3
6,fairfieldgh.com/files/road_0006.jpg,no pothole,71e4b0c93d,no pothole,2/2
7,fairfieldgh.com/files/road_0007.jpg,pothole,a26f58d01b,pothole,2/2
Speech data
Real voices, real accents
Speech models trained on studio English stumble in a Kumasi market. Ours are recorded the way your users will talk to you: on ordinary phones, in ordinary rooms.
- Read-aloud sentences, or spoken answers to your questions
- Choose 1, 2 or 3 different speakers for every sentence
- Each clip played back and checked before it's approved
“Mepɛ sɛ metɔ nsuo.”
How it works
From your items to a dataset in four steps
Choose and see the price
Pick labelling, voice or translation in your free account. Standard jobs are priced on the spot.
Pay by Mobile Money
Pay with the reference you're given. We confirm it and email your receipt.
Add your items
Upload pictures or paste your texts and sentences, then set the labels and how many people answer each.
Download your data
We check every answer, then you get the CSV and the files. Unused tasks are refunded.
What you get
Ready for your pipeline
CSV file
One row per answer, ready for pandas, Excel or your labelling tool.
Your files
Links to every picture and every recording, in the same rows.
Agreement
For labels: the winning label per item, and how many agreed.
PDF report
A summary to pass on, with how often each label came out on top.
Ideas
What teams build with it
- Pothole detection
- Shelf and price recognition
- Twi speech recognition
- Customer-review sentiment
- Chatbot answer ratings
- Crop and plant photos
- Receipt and signboard reading
- Spam and scam detection
- Local-language search
- Voice assistants
- Product photo matching
- Translation models
Pricing
Clear prices, before you pay
One price for 100 approved responses, shown the moment you choose. No subscription, no setup fee. Rejected work costs you nothing, and anything not filled is refunded.
Labelling
People choose a label for each of your images or texts
GH₵ 125.00
100 tasks of 10 = 1,000 labels
Voice recording
Your sentences read aloud, or questions answered by voice
GH₵ 260.00
100 tasks of 5 = 500 recordings
Translation
Your sentences translated into Ghanaian languages
GH₵ 260.00
100 tasks of 5 = 500 translations
Photo task
Photos of shelves, prices, posters and the like
GH₵ 260.00
for 100 approved responses
Questions
Before you ask us
Can I use the data to train my models?
Yes. Your results are yours to use, including for training. We don't resell them or use them for anything else, and they come without members' names or contact details.
How many people answer each item?
You choose 1, 2 or 3 when you build the job. For labelling, asking 2 or 3 people gives you an agreement score per item, so you can keep only the ones people agreed on.
How big can a job be?
Up to 500 pictures per labelling job, and as many texts or sentences as your paid tasks cover. Bigger than that? Chat with us and we'll plan it in rounds.
Who are the speakers and labellers?
Fairfield members: people across Ghana who signed up with a confirmed email, one account each, and are paid to their own Mobile Money. They choose the jobs they take, so they choose the languages they speak.
Did the speakers agree to be recorded?
Yes. A voice job says clearly that people will be recorded, and members choose whether to take it. Recordings carry an anonymous ID, not a name.
What format do I get?
A CSV with one row per answer: the item, the answer, an anonymous respondent ID and the time. Labelling adds the top label and agreement; voice adds a link to each recording, and pictures have links too.
Want to know how answers are checked? Read how we check the data.
Explore
What else we can do for you
Your model is only as good as its data.
Start with a small labelling or voice job, see the quality for yourself, then scale up.
Free account. See your price before you pay anything.