Vambo AI

Language technology for Africa, built in Africa.

Vambo AI builds speech and language models for African languages: text-to-speech, speech recognition, and small language models that run on the devices and connections people actually have. We work in Shona, isiNdebele, isiZulu, isiXhosa, Swahili, Kinyarwanda, Setswana, Hausa, Yoruba, Igbo, Nigerian Pidgin, Afrikaans and more, and we add languages with the communities that speak them.

What we do

• Speech. Text-to-speech and speech recognition for African languages, served through the Vambo voice API and used in our own products.
• Language models. Small, from-scratch language models trained African-first, evaluated in-language with our own multilingual safety and capability harness, and packaged to run on a laptop or a phone. Research previews live in this organisation as private repositories until they are ready to release.
• Data. Collection tools and datasets built with speakers, with provenance, licences and datasheets published alongside the models.

Principles

• Measure in the language, not in translation.
• Publish the failures next to the results.
• Release what runs offline. Connectivity is a design constraint, not an afterthought.

Contact

• Web: vambo.ai
• Data collection: collect.vambo.ai
• Partnerships, research collaborations and native-speaker reviewers: through vambo.ai