African languages are emerging as the next frontier in artificial intelligence (AI), and the good news is that space is vibrant with startups pushing to localise the capabilities.
Nigerian startup Intron recently expanded its flagship voice AI model, Sahara, to support 57 languages, including 23 African languages and more than 500 regional accents. The upgrade positions the company as one of the most aggressive builders of voice infrastructure designed specifically for African speech.
According to CEO Tobi Olatunji, the problem the company set out to solve is structural, arguing that most global AI systems were never designed for the linguistic complexity of African communication.
“Sahara v2 proves that when technology is built with deep cultural and linguistic understanding, amazing things can happen,” Olatunji said.
Sahara v2 was trained on over 14 million audio clips representing more than 40 000 speakers across 30 African countries, allowing the system to recognise local names, accents and multilingual speech patterns that international models often miss.
Africa is home to roughly 2 000 languages, many of which exist primarily as spoken languages with limited digital datasets, making them difficult for internationally created mainstream AI models to understand.
Another example is YarnGPT, an open-source speech system built by Nigerian developer Saheed Azeez. The project focused on text-to-speech models capable of generating audio and translating videos in Nigerian-accented English as well as Yoruba, Igbo, Hausa and Pidgin.
According to Azeez, the early release of YarnGPT demonstrated that demand for AI tools built for local languages is real, highlighting that language integration is critical for Africa’s AI ecosystem, because millions of people use local languages and accents that most global models don’t understand.
“If AI can’t process Yoruba, Igbo, Hausa, Pidgin or Nigerian-accented English, it will remain inaccessible to a huge part of the population,” he said.
Pelonomi Moiloa, CEO and co-founder of South Africa’s Lelapa AI estimated that her company has worked with around 35 African languages, but felt a targeted focus was preferable. “Our focus is on eight (languages) and those eight cover 520 million people, so almost half the population of the African continent.”
Lelapa AI focuses on creating resource-efficient small language models and its Vulavula solution specialises in transcription and translation, with a focus on customer service centres for telcos and financial services. “You could have two people having a conversation with each other, both of whom speak different languages,” she said.
Meanwhile, global technology companies are also entering the race. Google has begun expanding African language support across its AI-powered search products, including AI Overviews and AI Mode, adding languages such as Yoruba and Hausa to improve how its systems respond to queries from African users.
Research groups are also building foundational infrastructure. The pan-African NLP collective Masakhane is developing open datasets and language models for dozens of African languages, while Mozilla Foundation’s Common Voice project is crowdsourcing voice recordings to improve speech recognition systems globally.
The surge of activity reflects a broader shift in the AI industry: companies are realising that the next billion users will not interact with technology in English alone but in their local indigenous languages.
With locally trained models like Sahara v2, YearnGPT and Lelapa’s Vulavula, those familiar AI frustrations may soon fade. The everyday phrase “No worry, e go better” may no longer return as the baffling “No war eagle butter,” and names like Wanjiru or Chukwuebuka might finally be heard the way they were meant to be.
Share
