Intron Sahara v2.5 Targets Code-Switching in Voice AI

In Short

Intron has released Sahara v2.5, a voice AI update focused on African code-switching. That is when people mix languages mid-sentence.

What Happened

Lagos-based Intron says Sahara v2.5 improves speech recognition for conversations that switch between languages, a common pattern in many African countries. Sahara is a speech model, meaning software that turns spoken audio into text.

The company says the update supports code-switching across 12 African languages, including Hausa, Swahili, Luganda, and Zulu. It also includes a trilingual model for Kinyarwanda, English, and French. Intron says it has filed for US patent protection for that trilingual work.

Intron shared its own benchmark results, claiming Sahara performs better than Google Gemini, Meta models, and ElevenLabs on its tests. The company reported an average word error rate of 34.3% for Sahara versus 53.8% for Gemini 3.6 across the 12 languages it evaluated. Word error rate is a standard accuracy measure, it counts how many words a system gets wrong compared to a correct transcript.

A separate evaluation run via Gooey.ai for the Gates Foundation and CLEAR Global reportedly found Sahara ahead on five of seven Nigerian languages tested, according to Intron.

Intron also pointed to live deployments. It says the Ogun State judiciary has used Sahara for courtroom transcription for more than a year, expanding from one pilot court to nine. It also claims Sahara-powered collections at Branch recovered over ₦1.2 million in delinquent loans in a week, and that a Nairobi Swahili-English clinical visit can be converted into a structured note in under 30 seconds.

Why It Matters

Code-switching is a practical problem for voice AI in Africa. When transcription fails, it slows down customer support, healthcare documentation, and legal records.

Still, many of Intron’s performance and impact claims are self-reported and not independently verified. For buyers, the key questions are the test data used, how the model handles accents and noise, and whether accuracy holds up outside pilot environments.

If Sahara’s gains translate at scale, it could make African call centers, clinics, and courts more comfortable using voice automation without forcing users into one language.