African voice technology company Intron has launched Sahara v2.5, an updated voice AI platform designed to understand how people across the continent naturally switch between English, French, and local languages within the same conversation. The release expands the company’s speech recognition, text-to-speech, and voice-agent capabilities while increasing Sahara’s overall coverage to 31 African languages. Intron says the upgrade is aimed at reducing the communication gaps that occur when conventional speech systems lose context or accuracy during multilingual conversations.
Advancing African Code-Switching
A central feature of Sahara v2.5 is bilingual code-switching speech recognition across 12 African languages, including Zulu, Hausa, Swahili, Luganda, Yoruba, and Nigerian Pidgin. The company has also introduced a trilingual model that can recognize switches among Kinyarwanda, English, and French, using technology for which Intron says it has filed U.S. patents. According to benchmarks published by the company, Sahara outperformed competing systems from Gemini, ElevenLabs, and Meta across the 12 languages tested for code-switched speech recognition.
Expanding Voice AI Capabilities
The platform now supports language-mixing in both text-to-speech and voice-agent applications, enabling businesses to build systems that can respond more naturally to multilingual users. Intron said Sahara performed better than ElevenLabs and Gemini in 11 of 13 languages tested for these speech-generation tasks, while new recognition support for Nupe, Kanuri, Nigerian Fulfulde, Tigrinya, Kikuyu, Dholuo, and Somali brings total language coverage to 31. The update also introduces streaming speech recognition and text-to-speech APIs intended for live captions, real-time assistants, customer service tools, and other high-volume applications.
Enterprise Use Cases and Deployments
Intron is positioning Sahara v2.5 for sectors where accurate speech processing can directly affect service delivery, including healthcare, finance, legal services, government, and contact centers. Existing deployments include Branch International’s collections operations, voice interaction tools developed with Audere Africa, transcription systems used by the Ogun State Judiciary, and medical documentation workflows in East Africa. The company also has offline models running on Nvidia hardware at PAMO Clinics in Port Harcourt, giving institutions an option for local deployment where privacy, connectivity, or regulatory requirements limit reliance on cloud-based AI services.
Benchmarking and Developer Adoption
The company reported an average word error rate of 34.3% across 12 code-switched languages, compared with 53.8% for Gemini 3.6, which Intron described as a 36% relative reduction in errors. Separate testing conducted through Gooey.ai for the Gates Foundation and CLEAR Global found Sahara performed better on five of seven Nigerian languages evaluated, according to the release. The technology was also showcased at the 2026 Deep Learning Indaba in Lagos, where more than 120 teams from 21 countries participated in the Sahara CodeSwitch Hackathon and built applications across fintech, healthtech, agritech, and edtech.
Building an African Voice AI Platform
Intron says Sahara currently supports more than 40 enterprise customers across six African countries, including Nigeria, Kenya, South Africa, Uganda, Rwanda, and Ghana. Since raising $1.6 million in pre-seed funding in 2024, the company has expanded its training dataset to more than 150,000 hours of African-language audio from over 53,000 speakers, covering 64 languages and more than 500 accents. Chief Executive Officer Tobi Olatunji said the company’s goal is to make voice AI adapt to how Africans already communicate rather than requiring users to simplify their accents, vocabulary, or language choices.
Sahara v2.5 strengthens Intron’s push to build voice AI around the multilingual realities of African markets, with code-switching positioned as a core requirement rather than a niche feature. By combining broader language coverage, real-time APIs, offline deployment options, and enterprise use cases, the company is targeting organizations that need more accurate speech technology in everyday operational settings. The release also highlights a broader challenge for global AI developers: effective localization depends not only on language datasets, but also on models that reflect how people actually speak across different cultural and linguistic contexts.