Intron launches Sahara v2.5, expanding African voice AI to 31 languages with code-switching support
The AMW Read
Independently benchmarked language-coverage and code-switching upgrade for an existing regional voice-AI player with multi-vertical commercial traction, but not a top-tier entrant or debate-resolving event.
Intron launches Sahara v2.5, expanding African voice AI to 31 languages with code-switching support
Nigerian voice-AI company Intron released Sahara v2.5, adding bilingual code-switching speech recognition across 12 African languages and expanding overall model coverage to 31 languages, including new support for Nupe, Kanuri, Nigerian Fulfulde, Tigrinya, Kikuyu, Dholuo and Somali. Text-to-speech and voice-agent capabilities now cover 13 languages. Intron reports a 34.3% average word error rate across the 12 code-switching languages versus 53.8% for Gemini 3.6, a 36% relative reduction, and cites Gates Foundation/CLEAR Global testing that ranked Sahara ahead of Gemini and Meta's Omnilingual model on five of seven Nigerian languages evaluated.
Code-switching — speakers shifting languages mid-sentence — is the everyday failure mode for voice models trained mainly on high-resource languages. Per the AI Market Watch index, this is the first Intron item the pipeline has logged in 90 days, against one in the prior period (coverage limited to name-matched, pipeline-ingested sources) — a pre-seed-stage player converting a narrow linguistic capability into revenue across three procurement buyers: Branch International's loan-collection voice agents, the Ogun State Judiciary's transcription rollout (now nine courts, up from one pilot), and a Nairobi clinical-documentation workflow, plus an offline deployment on Nvidia hardware at PAMO Clinics for connectivity-constrained settings.
For builders, the offline, on-device deployment is the more transferable detail — a template for markets with weak connectivity or data-residency requirements, rather than a head-on benchmark contest with hyperscaler models. For investors, 40-plus enterprise customers across six countries on $1.6 million in pre-seed funding since 2024 signals a capital-efficient early motion, though a single funding round and mostly self-reported benchmarks limit how far that efficiency can be projected before a larger raise or broader third-party validation.
