AI: NKENNEAi launches Swahili voice models trained on 65,000 hours of audio
NKENNEAi, the artificial intelligence division of the African language learning platform NKENNE, announced its first voice models on September 25, 2026, starting with Swahili, according to TechCabal. These include a speech-to-text transcription model and a text-to-speech synthesis model, which complement the text translation tools already offered by the company.
Trained on over 65,000 hours of audio
According to TechCabal, the models were trained on more than 65,000 hours of recordings. They are designed to account for dialectal variations of a language spoken by over 100 million people in East and Central Africa, as well as tones, the pitch variations that change the meaning of words in languages like Yoruba or Igbo.
The company, led by its founder Michael Odokara-Okigbo, relies on validated linguistic data from its application, which has over 400,000 learners across 15 African languages. A public demonstration is scheduled for September 29, 2026.
Funding and partnerships
NKENNEAi has secured a $1 million grant from the U.S. National Science Foundation under Phase II of its Small Business Innovation Research program. In Nigeria, the company is working with the National Information Technology Development Agency (NITDA) on an offline AI deployment, and with the operator T2 Mobile for customer services.
A challenge of linguistic sovereignty
Most major global voice tools remain designed for English, French, or Mandarin, leaving out hundreds of millions of African speakers. Models built from African data pave the way for banking, agricultural, or health services accessible in the users' own languages, including for those who cannot read.
The next steps announced concern Nigerian Pidgin, Yoruba, Igbo, and Somali, according to TechCabal.
This article was written in French with the help of artificial intelligence from the sources cited below, then translated from French automatically. Read the original.
