Rumbu
A consented Hausa speech commons, Niger first
What would Hausa speech data look like if it were collected with the people it represents?
Hausa speech collected with consent and fair pay, from the people and places public datasets leave out. Most public Hausa speech data was read aloud from a script, in Nigeria, by people comfortable with a phone app.
Conversation
Two people on separate channels, and calls over ordinary phones.
Consent
Every speaker chooses what their voice may be used for, and is paid.
Coverage
Regional varieties from Niger and Nigeria, including people who do not read.
Most public Hausa speech data was read aloud from a script, in Nigeria, by people comfortable with a phone app. FLEURS, a benchmark many papers report, has a Hausa test set labelled Nigerian Hausa in which 620 of 621 utterances are spoken by women. That is not the Hausa our users speak, and models trained and tested on it learn a narrow version of the language.
Rumbu is our programme to build the Hausa speech that is missing, the way a community keeps a granary: together, with care, and with clear rules about who may draw from it. We will record Hausa from Niger and Nigeria in its regional varieties: conversation between two people on separate channels, speech over ordinary phone calls as well as apps, and people who do not read, prompted by pictures and questions rather than text.
Every speaker will give informed consent in Hausa, choose what their voice may be used for, and be paid for their time. Each dataset will ship with a data statement that says who is in it, how it was recorded and what it cannot support.
We will release what consent and law allow, and say plainly what we keep closed and why.
What we do not claim yet
- Nothing has been collected yet. Rumbu is opening, not a dataset.
- We have not set the pay rate or a release date; both will be published before collection starts.
- No contributor’s face, name or village will appear without their explicit choice.
Who is in Hausa speech data today Every utterance of the FLEURS Hausa test split, in file order, coloured by the gender recorded in the dataset.
QuestionsWhat this programme sets out to answer
- 01
What does a representative Hausa speech sample look like across Niger and Nigeria, and how far are current public sets from it?
- 02
How do you collect spontaneous and conversational Hausa from people who do not read, over ordinary phones, at a quality usable for recognition and conversation research?
- 03
What consent design gives speakers real choices, and keeps those choices attached to every derived file and model?
- 04
What is fair pay for a recording session in Niger and northern Nigeria, and how should it be reported?
ProgressDates appear only on finished steps
- NextDesign and consentConsent form and information sheet in Hausa (audio and text), scope options, pay rate, data-protection review, outside reviewers.
- NextPilotA small multi-dialect pilot with two-channel pairs and phone-line calls; cost per transcribed hour and quality, measured.
- NextTest set firstA speaker-disjoint, dialect- and gender-stratified evaluation set, frozen before training data is collected.
- NextScale where the pilot shows gaps
- NextReleaseData statement, licence tiers by consent, and a withdrawal process.
What we will release
- Rumbu-Hausa v0, released in tiers chosen by speakers’ consent
- A data statement and datasheet
- A phone-based, picture-prompted collection flow
- Paper: collecting conversational Hausa in Niger — consent, cost and coverage
Work with us on Rumbu
We are looking for community partners in Niger and northern Nigeria, people to review the consent design, and organisations that want to fund paid collection.
Get in touchMore research
All researchHira
Speech models that listen while they speak, built for Hausa and judged against how Hausa speakers actually take turns.
Zarma
The first open, consented speech resources and honest baselines for Zarma, spoken by about six million people along the Niger.
Tankaɗe
The methods that turn radio archives, field recordings and crowd audio into Hausa and Zarma data a model can learn from and a reader can trust.
Maʼauni
Test sets, text normalisation, listening protocols and benchmarks that make a Hausa or Zarma result mean something.