Intron, an Africa-focused voice technology company founded in Lagos, has released Sahara v2.5, a set of models built for that specific scenario. The release adds code-switching support—the ability to follow a speaker moving between languages inside a single sentence—to 12 African languages, including Zulu, Hausa, Swahili and Luganda. 

It also introduces what Intron calls the world’s first African trilingual speech recognition model, which handles conversations moving between three languages. It is built for Rwanda, where Kinyarwanda, the national language spoken by almost the entire population, coexists with English and French in everyday professional life. The company says it has filed US patents on the algorithms behind it.

Intron believes that code-switching is a distinct technical problem that global labs have treated as a rounding error. A model can be excellent at English, competent at Swahili, and still fall apart when the two appear in the same breath. Intron thinks that fixing this requires dedicated data, dedicated training, and dedicated evaluation, and that a small company with a narrow problem can beat a large one with a global average.

On Intron’s own benchmarks, Sahara v2.5 recorded an average word error rate of 34.3% across 12 languages of code-switched African speech, against 53.8% for Gemini 3.6, a Google AI model. In practice, that means Sahara gets about one word wrong in three, while Gemini gets more than one in two.

The company says Sahara beat Gemini, ElevenLabs and Meta on all 12 languages it tested. However, this still means roughly one word in three comes out wrong. 

Founded in 2020 by Tobi Olatunji, a Nigerian-trained doctor, and Kunle Asekun, Intron raised $1.6 million in pre-seed funding in July 2024, led by Microtraction, to solve a paperwork problem in hospitals but is now trying to figure out how to make voice AI work across the continent. 

Why code-switching is hard

Speech is harder than text here, as a model reading a written sentence can see where each word starts and stops. A model listening to audio gets only sound. It has to work out the sounds, the words, the accent, the context, and which language it is hearing—all at once with nothing to mark where Yoruba ends and English begins. The switch can come between sentences, inside a sentence, or around a single word.

Other systems handle this by first identifying the language, then routing each segment to a monolingual recogniser—a model trained to transcribe a single language. That approach holds up on long, clean segments and breaks down when a switch lasts one or two words. Intron says it trains Sahara directly on mixed-language speech, so the model treats switching as ordinary, learns which transitions are possible in each language pair, and resolves them using acoustic evidence and context across the whole utterance.

There is a second problem, and it starts with tokenisation. Before a model can process speech, it breaks the sound into small units called tokens, each one matched to a piece of language it has seen before. If African-language forms make up only a small share of the training data, the model has few African tokens to match against. So it forces an unfamiliar African sound into the nearest English or French token it knows. The result is either a hallucination—a word the speaker never said—or a stretch of speech that vanishes from the transcript altogether.

The training data itself is also scarce. Recordings of natural code-switched speech are hard to come by, and artificially mixed audio does not capture how people actually switch.

“Code-switching was one of the biggest problems that consistently came up for clients deploying real-world voice AI,” Tobi Olatunji, Intron’s CEO, told TechCabal in an interview. “Africa needs AI built for how Africans really speak. People should not have to translate themselves for a machine, flatten their accent, avoid local expressions or repeat only the English part of what they said.”

Who is using Sahara? 

Intron says its model now supports production and research deployments for more than 40 organisations across six countries: Nigeria, Kenya, South Africa, Uganda, Rwanda and Ghana.

The commercial case Intron shared is Branch International, a fintech lender, where Sahara-powered collections agents recovered more than ₦1.2 million ($891) in delinquent loans in one week, with record after-hours and weekend repayments and stronger performance than human agents 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, in a statement shared with TechCabal.

Text-to-speech, which converts written words into spoken audio, and voice agents now handle language mixing across 13 language pairs, beating ElevenLabs and Gemini in nine of them on Intron’s tests, the startup said in a statement. 

New speech recognition support for Nupe, Kanuri, Nigerian Fulfulde, Tigrinya, Kikuyu, Dholuo, and Somali takes total coverage to 31 languages. The company has also added streaming speech recognition and streaming text-to-speech to its application programming interface (API)—the connection developers use to plug Intron’s models into their own products—for live captions and real-time applications.

Intron has also published a 2026 Africa Voice AI Report, which argues that collecting African language data is only one of the barriers to reliable voice AI. Research capacity, the ability to string systems together, and the expertise to deploy them matter just as much.

An example is the ambient medical scribe—software that listens to a doctor-patient consultation and automatically writes up the notes. It is widely used in the US and Europe, but largely unworkable in African clinics, where a single consultation may involve two or three languages.

True scale demands moving beyond surface-level integrations to robust execution. We’ve filtered the noise out of Moonshot 2026, optimising the conference strictly for high-calibre connections between startup founders, global financial operators, enterprise leaders, and individuals rewiring Africa’s technical frameworks. Get 20% off Early Bird tickets for a limited time.