Nigerian AI Startup Brings Speech Recognition to 57 Languages

When a doctor in Kenya switches from English to Swahili mid-conversation with a patient, most voice recognition systems get confused. Not anymore.

Intron, a Nigerian AI startup, just expanded its Sahara platform to recognize 57 languages, including 23 African languages like Hausa, Yoruba, Swahili, and isiZulu. The technology understands more than 500 distinct African accents and can even follow conversations when people code-switch between languages.

For a continent where nearly 2,000 languages exist mostly as spoken tongues with little written form, this matters enormously. People can finally interact with technology the way they naturally speak, not the way Silicon Valley designed systems to work.

Founded in 2020 by Tobi Olatunji and Olakunle Asekun, Intron built Sahara using over 14 million audio clips from more than 40,000 speakers across 30 African countries. Much of this data didn't exist before, so the team recruited contributors, compensated them fairly, and built the datasets from scratch.

The results speak for themselves. In company benchmarks using African voice datasets, Sahara outperformed major systems like GPT-4, Gemini, and Whisper by up to 64% when recognizing African names, organizations, and locations. It also performed 35% better with numbers and 20% stronger in noisy environments.

Real organizations are already seeing the impact. The Ogun State Judiciary uses Sahara for documentation. Audere, a South African company, transcribes WhatsApp voice messages across multiple local accents. Penda Health in Kenya partnered with Intron to develop the world's first bilingual Swahili-English model that handles rapid language switching during patient visits.

The Ripple Effect

The platform now serves enterprise and government clients across Nigeria, Kenya, South Africa, Ghana, Rwanda, and Uganda. Healthcare providers can document patient visits faster. Call centers can deploy voice bots that speak local languages. Legal services can transcribe proceedings accurately in multiple dialects.

Intron made its datasets publicly available so anyone can test global AI models on African speech. The company is also releasing its first Africa Voice AI Report to help raise standards across the industry.

More languages are coming soon. Intron is developing additional bilingual models for Yoruba, Hausa, Zulu, and Kinyarwanda. The startup plans to raise $3 million this year to expand coverage and continue building technology that actually works for how Africans communicate.

For regions with limited internet, Sahara now runs fully offline through partnership with Nvidia, with entry-level devices costing around $250. Technology that once excluded millions is now being rebuilt to include them from the ground up.