Voices
- Audio
- Mono, 16 kHz, AAC in an .m4a file
- Prompt
- The French sentence said aloud
- Text
- Optional verbatim transcription
- Speaker
- Gender and age range
- Setting
- Quiet indoors, outdoors or noisy
15 Nov 2026Our collection tools are built and tested end to end on our own phone. Field collection has not started yet: we are recruiting a first group of speakers. No dataset is for sale yet.
What exists today, and what does not
We would rather show you where we really are.
- BuiltA collector app for Android phones. It works offline and sends each item when there is a network.
- BuiltA review dashboard where our team listens to each recording and accepts or rejects it with a reason.
- BuiltOptional verbatim transcription: a speaker who writes the language types what they said, and a reviewer judges it separately.
- BuiltA word-checking task: speakers confirm or correct French-to-language word translations. Each word is seen by two speakers, with control words to catch careless answers.
- BuiltA database where each collector sees and sends only their own items, and cannot change their review status or earnings.
- BuiltAn export of accepted items with their files, without names, phone numbers or locations, split by speaker.
- Not yetFirst collectors in the field. The app has been tested on our own Android phone, not yet with them.
- Not yetAny recorded data, any published dataset, any sample to send.
What we collect
Collectors see a short sentence in French and say it in their own language. We never put invented sentences in Chadian languages in front of them. These are the fields the app records today.
Sentences and words
- Pairs
- A sentence in the language and its French translation
- Topic
- Market, conversation, health, farming and herding, school, religion, proverbs
- Words
- French-to-language translations checked by two speakers
Photos
- Image
- Market, school, street, home, field, sign or poster
- Text
- A description of the scene, and any writing visible in it, transcribed
Delivery formats (for example JSONL with audio files, laid out for Hugging Face) are agreed with each buyer.
Languages we work in
Codes are ISO 639-3. We open a language when we have speakers to record it and a reviewer who can read it.
Chadian Arabic has no speech recognition in the major open models. The open speech data we know of is about 12 to 15 hours of read speech recorded in Nigeria, where the language is called Shuwa Arabic, under a non-commercial licence. We record everyday speech in Chad, and it can be licensed for commercial use.
Dark green: first waveLight green: pilotGrey: later, on request
From a phone in Chad to your dataset
1ConsentA collector creates an account and accepts the consent terms before sending anything.
2CollectionThey record, translate or photograph, online or not.
3ReviewA reviewer accepts or rejects each item. Rejected items are not paid and never delivered.
4PaymentAccepted items are paid to the collector by mobile money.
What a collector earns per accepted item
A parallel corpus, agreed in principle
Published Bible translations exist in several Chadian languages. Aligned verse by verse with French, they make a useful translation corpus. These translations belong to their publisher, the Alliance Biblique du Tchad.
The Alliance Biblique du Tchad holds the rights to the translations in four of our languages: Chadian Arabic, Sar, Ngambay and Moundang. It has agreed in principle, orally, that we may build this corpus in these four languages and license it, including for commercial use. The Massa and Fulfulde translations belong to the Alliance Biblique du Cameroun and are not covered. We are now putting that agreement in writing. We will offer the corpus only once the written licence is signed, and on its terms.
Clear prices
Data collected to your specification, in USD. About a quarter of each price goes to the speakers. Each order gets a written quote with delivery date and licence terms.
Speech$1,600per validated hour
Transcribed speech$2,000per validated hour
Sentence pairs$700per 1,000 pairs
Described photos$1,400per 1,000 photos
Say hello
Tell us the language and the data you need. We reply in English or French.
contact@innovservices-td.com
You can also open your mail app.