40,000 voices, 500 Accents. 24 New Languages. Intron Sahara-v2 Sets the Standard for African Voice AI
Built on 14 million+ audio clips, 50,000 hours, from 40,000+ African speakers, the new release outperforms leading global models on several benchmarks, and introduces the inaugural 2026 Africa Voice AI Report
“Sorry I didn’t catch that” – a familiar frustration for millions of Africans using voice technology like Siri and Alexa, where common phrases and names like “No worry, e go better” becomes “No war eagle butter”, “Wanjiru” becomes “One zero” and “Chukwuebuka” becomes “Check wheelchair baker”. These repeated failures aren’t just technical gaffes, they become exclusionary when, despite speaking clearly, voice assistants misunderstand basic words and names, locking you out of digital shortcuts that make life easier.
African languages are richly nuanced; a single word can mean six different things depending on pronunciation, and a single office can employ staff with 10 different accents. Global voice AI models don’t get this. They were not built for the tonal richness, code-switching patterns, and accent diversity of African speech.
Today, Intron raises the bar for African voice AI, making sure every voice is heard. Sahara v2, its flagship second-generation speech recognition model, now covers 24 African languages, adding support for Hausa, Swahili, Zulu, Yoruba, Kinyarwanda, Twi, Igbo, Xhosa, African French, and many more — bringing total coverage to 57 languages. The model understands over 500 distinct African English accents, capturing the speech patterns that make African languages unique. This model delivers industry leading performance across several African names, numbers, fractions, currency, and domain specific terms where precision is critical.
The release also introduces the world’s first bilingual Swahili-English ASR model, developed in collaboration with Penda Health, Kenya to support rapid switching between English and Swahili, better reflecting how people naturally speak. Intron also expands its suite of fluent, low-latency speech-to-speech systems with the release of its first local language Hausa TTS model, enabling native language voice bots for conversations 24/7. Also, with new offline support, models can be deployed securely to support sovereign AI, alleviating privacy concerns.
Compared to leading models including Gemini-3, GPT-4, Whisper, ElevenLabs, AWS, and Azure, Sahara v2 delivers 68.6% better performance on African names, organizations, and locations, 55.6% stronger performance with numbers, 36.5% greater hallucination robustness with background noise, silence, and overlapping speakers, and 46.7% better performance across verticals such as health, legal, finance, and telecommunications, positioned as the preferred pan-African voice AI partner for regional and global organizations seeking reliable and private African voice technology.
This performance did not emerge from assumptions about Africa, it came by listening. Sahara v2 is trained on over 14 million audio clips, spanning 50,000+ hours, from 40,000+ speakers representing 64+ African and diaspora languages from over 30 African countries, recording their voices in busy clinics, streets, courtrooms, call centers, and real environments where voice AI must perform.
Backed by robust APIs designed for real-time and asynchronous production deployment in as little as five minutes, Sahara v2 powers key services such as voice banking, voice-powered autofill for account opening forms, KYC forms, health data forms, application and admission forms, and daily reports reducing processing time by 4.4×. Enterprises, startups, and government clients from six countries are already deploying Sahara v2 to power key services such as voice bots, medical ambient listening solutions, courtroom transcription, and call center solutions.
“Using Intron AI models, we’ve seen significant improvement in transcription and summaries compared to models we previously explored. Their systems capture context and nuance better, leading to more accurate results,” Ayo Oluleye, Head of Data & Insights at ARM Investments.
“In our testing, accuracy was excellent on several Southern African accents and APIs were robust with 99%+ success rates.” said Sarah Morris, CPO at Audere.
Alongside the launch, Intron is releasing its inaugural 2026 Africa Voice AI Report, covering the landscape, insights, and lessons from building and deploying real-world voice AI across Africa, providing guidance for governments, enterprises, investors, startups, researchers, and regulators shaping the future of inclusive AI.
“Sahara v2 proves that when technology is built with deep cultural and linguistic understanding, amazing things can happen, and we’re just getting started,” said Tobi Olatunji, CEO, Intron.
The newest languages include African French, Afrikaans, Akan, Amharic, Arabic, Bemba, Fulani, Ga, Hausa, Igbo, Kinyarwanda, Luganda, Oromo, Pedi, Pidgin, Sesotho, Shona, Swahili, Tswana, Twi, Wolof, Xhosa, Yoruba, Zulu.
Intron builds voice AI infrastructure for Africa, enabling organizations to deploy speech recognition and text-to-speech solutions that truly understand the continent’s linguistic diversity. With coverage across 57 languages and 500+ accents, Intron is making voice technology accessible to millions of African speakers previously underserved by global AI systems.
