Nigerian Startup Intron Launches Enhanced Voice AI for African Multilingualism

Intron, a Lagos-based voice AI company, has released Sahara v2.5, an update to its speech models designed to handle code-switching—the natural mixing of languages within sentences—common in African communication. This new version supports code-switching across a dozen African languages, including Zulu, Hausa, Swahili, and Luganda, addressing a major limitation of global AI models often trained on monolingual datasets.

A key feature of Sahara v2.5 is a trilingual model for Kinyarwanda, English, and French, which Intron claims is an African-language AI first and is seeking U.S. patent protection for. The company's internal benchmarks suggest Sahara outperforms major global AI models like Gemini, ElevenLabs, and Meta in code-switching scenarios, although these claims lack independent verification. A separate evaluation for the Gates Foundation and CLEAR Global also reportedly showed Sahara leading in several Nigerian languages.

Intron highlights several real-world deployments to demonstrate the system's effectiveness. In Nigeria's Ogun State, the judiciary uses Sahara for courtroom transcription, reportedly halving hearing times. Digital lender Branch International saw significant recovery of delinquent loans using Sahara-powered collection agents. In Nairobi, the system is said to transform doctor visits into structured clinical notes rapidly. The update also expands speech recognition to seven more African languages, bringing total coverage to 31, and introduces streaming transcription and text-to-speech capabilities.

The company has also focused on offline models for regions with unreliable internet, deploying them in clinics in Port Harcourt. Intron has grown significantly since its 2024 pre-seed funding, now boasting over 150,000 hours of audio data and more than 40 enterprise customers across six African countries. This development underscores the critical need for AI solutions tailored to the unique linguistic realities of Africa, rather than relying on systems developed for Western contexts.