Gungbe speech and text data, made properly.
Gungbe (Gun language, ISO 639-3 guw) is spoken every day across southern Benin and in
Badagry, Nigeria, yet almost no usable data exists for it. We produce it with a team of
native speakers who have worked on Gungbe language resources for more than four years.
Gungbe Customer Care Corpus
2,000 customer-service sentences across telecom, mobile money, banking, health, utilities, phone menus, numbers and dates — in English, French and Gungbe — each recorded by at least three different native speakers.
| Item | Detail |
|---|---|
| Sentences | 2,000 (EN · FR · Gungbe) |
| Recordings | about 6,000 clips, 8–10 hours |
| Audio | WAV, 16 kHz, mono, 16-bit |
| Text | UTF-8, Unicode NFC, numbers written as words |
| Speakers | Anonymous codes with gender, age range and region |
Custom collection
Your own sentences or domain — health messages, product names, IVR prompts, conversational speech — translated, recorded and checked by our team, in Gungbe or other languages we work in.
- Studio-quality single-voice recordings for text-to-speech
- Multi-speaker read speech for speech recognition
- Parallel text for machine translation
- Transcription and validation of your own audio
How every item is checked
- A native speaker translates the English/French sentence into Gungbe.
- A different native speaker reviews it: approves, corrects or rejects it.
- Several speakers record the approved sentence on their phones.
- Another speaker listens to each recording; the app does not allow a decision until the whole clip has been played.
- An administrator listens again before any recording is accepted into the dataset.
Anyone on the team can flag a text mistake at any stage; when a text is corrected, recordings of the old wording are removed so audio always matches its transcript.
Consent and licensing
Every contributor is 18 or older and has accepted our contributor terms, which grant Guwit the rights needed to license the data commercially. Datasets contain no names, phone numbers or usernames. We offer:
- A free sample (about 200 clips) for evaluation, under a non-commercial licence.
- Commercial licences for the full corpus, priced per use case.
- Exclusive custom datasets, owned by you, for collections made to your specification.
Every delivery includes
parallel_text.csv— English, French and Gungbe for every sentencemetadata.csv— one row per clip with transcript, domain and speaker detailsspeakers.csv— anonymous speaker codes, gender, age range, region- A datasheet describing how the data was made and its known limitations
Request a sample.
Tell us what you are building and we will send the sample and the datasheet.