Intron launches Sahara-v2, a voice AI model supporting 24 new African languages

Intron, a Lagos-based AI startup has officially launched Sahara-v2, its flagship, second-generation voice AI model, effectively raising the bar for global speech recognition. The model supports 24 new African languages and mastering over 500 distinct African English accents, bringing the company’s total language coverage to an unprecedented 57. 

For years, interacting with global voice assistants like Siri, Alexa, or standard text-to-speech engines has felt like talking to a wall that fundamentally doesn’t speak your language. We’ve all been there, trying to use a voice command only to have a localized phrase like “No worry, e go better” transcribed into the nonsensical “No war eagle butter.” It’s funny at first, but when “Wanjiru” is digitized into “One zero” and “Chukwuebuka” inexplicably becomes “Check wheelchair baker,” the joke fades quickly.

These repeated transcription failures are more than mere technical gaffes; they are an active form of digital exclusion. Built predominantly on Western datasets, leading global AI models completely miss the mark here. They aren’t wired for the tonal richness of our speech, where a single word can mean six different things depending on the pronunciation. They stumble over our natural rhythmic code-switching and cannot comprehend that a single office might employ staff with ten entirely different accents. When voice assistants misunderstand basic words and names, they lock millions of users out of the digital shortcuts that make modern life easier.

Intron

Sahara-v2 bridges this critical gap because it wasn’t built on assumptions or trained in a quiet, sterile studio. Intron built this model by going to the streets and listening. The dataset is massive and hyper-localised: over 14 million audio clips, totalling more than 50,000 hours of speech, sourced from over 40,000 speakers. These voices represent 64 African and diaspora languages across more than 30 countries. Crucially, they were recorded where AI actually needs to function, in crowded clinics, noisy markets, bustling call centers, and courtrooms.

Intron’s Sahara-v2 beat industry leaders

This ground-up, real-world approach has allowed Sahara-v2 to run circles around the biggest names in global tech. When benchmarked against industry heavyweights like Gemini-3, GPT-4, Whisper, ElevenLabs, AWS, and Azure, Intron’s model delivers commanding leads. 

It posts a 68.6% better accuracy rate on African names, organizations, and locations, and is 55.6% sharper at transcribing numbers, fractions, and currencies. Furthermore, it boasts a 36.5% greater resistance to the AI hallucinations caused by background noise and overlapping speakers, alongside a 46.7% performance bump across specialized verticals like healthcare, law, telecommunications, and finance.

Beyond the impressive benchmarks, Sahara-v2 is a serious productivity engine designed for real-world enterprise deployment. Backed by robust APIs that can be deployed in as little as five minutes, the infrastructure is already actively powering voice banking, automated KYC processes, and autofill systems for everything from health data to account opening forms. Early enterprise adopters report that the technology is slashing administrative processing times by up to 4.4 times.

To address the continent’s unique linguistic realities, Intron has rolled out highly specialized regional features. In collaboration with Kenya’s Penda Health, they introduced the world’s first bilingual Swahili-English Automatic Speech Recognition (ASR) model, perfectly capturing how people naturally flip between languages mid-sentence. 

For the market here, they’ve launched a native Hausa text-to-speech (TTS) model, enabling low-latency, 24/7 voice bots. And for governments and enterprises navigating data privacy concerns, the Sahara-v2 rollout includes new offline support, allowing data to remain secure and local for sovereign AI compliance.

Intron’s Sahara-v2

Enterprises and government clients across six countries are already utilising the system. Ayo Oluleye, Head of Data & Insights at ARM Investments, noted, “Using Intron AI models, we’ve seen significant improvement in transcriptions and summaries compared to models we previously explored. Their systems capture context and nuance better, leading to more accurate results.” Sarah Morris, CPO at Audere, echoed this, reporting 99%+ API success rates and excellent accuracy on Southern African accents.

Alongside the Sahara-v2 launch, Intron also released its inaugural 2026 Africa Voice AI Report. The publication offers a comprehensive, first-of-its-kind look at the ecosystem, detailing the lessons learned from deploying voice AI in complex environments to guide startups, investors, and policymakers.

Also read: Google to train AI in 21 African languages, including Yoruba, Hausa and Igbo

The newest languages include African French, Afrikaans, Akan, Amharic, Arabic, Bemba, Fulani, Ga, Hausa, Igbo, Kinyarwanda, Luganda, Oromo, Pedi, and Pidgin. Others are Sesotho, Shona, Swahili, Tswana, Twi, Wolof, Xhosa, Yoruba, and Zulu. With these additions, the door to the digital world is finally swinging wide open. 

As Intron CEO Tobi Olatunji put it, “Sahara-v2 proves that when technology is built with deep cultural and linguistic understanding, amazing things can happen, and we’re just getting started.”

The post Intron launches Sahara-v2, a voice AI model supporting 24 new African languages first appeared on Technext.