Intron has released Sahara v2.5, a new set of voice AI models designed to improve speech recognition for Africans who regularly switch between languages when speaking. The initiative could make voice technology more useful across business, healthcare, finance and other sectors where conversations often move between English and African languages.
The Lagos-based voice technology company has added code-switching support to 12 African languages, including Zulu, Hausa, Swahili and Luganda. Code-switching allows the system to understand a speaker who moves between languages within the same sentence, such as, “My loan imekataliwa, but I paid yesterday.” In many existing speech recognition systems, the English may be captured while the Swahili is missed, leaving out important information.
Sahara v2.5 also introduces what Intron describes as the world’s first African trilingual speech recognition model. It is designed for Rwanda, where Kinyarwanda, English and French are commonly used in professional settings. Intron says it has also filed US patents covering the algorithms behind the technology.
Building Voice AI Around How Africans Speak
The company argues that code-switching has often received limited attention from global AI developers, even though it is common in African markets. A system may perform well in English and separately in Swahili, but struggle when both languages appear in the same sentence.
Intron’s own benchmarks show Sahara v2.5 recorded an average word error rate of 34.3% across 12 languages of code-switched African speech, compared with 53.8% for Gemini 3.6. The company says Sahara also outperformed Gemini, ElevenLabs and Meta across all 12 languages tested. However, the results still show that the technology has room to improve, with roughly one in three words incorrectly transcribed.
Tobi Olatunji, Intron’s CEO and co-founder, said the problem became clear through work with customers deploying voice AI.
“Code-switching was one of the biggest problems that consistently came up for clients deploying real-world voice AI,” Olatunji said. He added that “Africa needs AI built for how Africans really speak,” without requiring people to change their accents, avoid local expressions or repeat only the English parts of conversations.
Founded in 2020 by Olatunji, a Nigerian-trained doctor, and Kunle Asekun, Intron initially focused on reducing paperwork in hospitals. The company raised $1.6 million in pre-seed funding in July 2024 in a round led by Microtraction. It has since expanded its focus to the wider challenge of making voice AI work effectively across Africa.
The technical challenge is significant because speech recognition must identify sounds, words, accents, context and language without clear markers showing where one language ends and another begins. Intron trains Sahara directly on mixed-language speech rather than simply identifying a language and sending it to a separate monolingual system.
From Financial Services to Healthcare
Intron says Sahara now supports production and research deployments for more than 40 organisations across Nigeria, Kenya, South Africa, Uganda, Rwanda and Ghana.
One example is Branch International, where Sahara-powered collections agents recovered more than ₦1.2 million in delinquent loans in one week. The company reported stronger after-hours and weekend repayments, including better performance on loans more than 356 days overdue.
“Customers engaged naturally even after hours and on weekends,” said Adanne Anene, Head of Product Africa at Branch.
The technology has also expanded beyond speech recognition. Intron says its text-to-speech and voice agents can now handle language mixing across 13 language pairs, while new support for Nupe, Kanuri, Nigerian Fulfulde, Tigrinya, Kikuyu, Dholuo and Somali brings total language coverage to 31.
Streaming speech recognition and streaming text-to-speech have also been added to Intron’s API, allowing developers to build applications such as live captions and other real-time voice services.
The company’s 2026 Africa Voice AI Report argues that better language data alone will not solve the continent’s voice AI challenges. Research capacity, technical skills and the ability to deploy connected AI systems are equally important.
Healthcare provides a clear example. Ambient medical scribes, which listen to doctor-patient consultations and automatically prepare notes, are already used in the US and Europe. In many African clinics, however, consultations can involve two or three languages, making existing systems difficult to use.
By focusing on that gap, Sahara v2.5 could help move voice AI closer to the realities of African businesses and communities, while opening new opportunities for locally relevant technology across the continent.
