Nigerian AI startup Intron has launched Sahara v2.5, introducing an African trilingual speech-recognition model capable of understanding code-switching between English and local languages.
Unveiled at the 2026 Deep Learning Indaba in Lagos, Sahara v2.5 supports bilingual speech recognition across 12 African language pairs, alongside trilingual Kinyarwanda-English-French recognition.
In published benchmarks, Sahara v2.5 recorded an average word error rate of 34.3% across 12 tested languages — including Zulu, Hausa, Swahili, and Luganda — representing a 36% relative reduction in errors compared with global models.
The upgraded platform expands its speech recognition, text-to-speech, and voice-agent capabilities to 31 African languages.
Separate benchmarking conducted for the Gates Foundation and CLEAR Global found Sahara performed better in five out of seven evaluated Nigerian languages.
The technology improves mixed-language speech recognition for enterprise clients in healthcare, financial services, legal affairs, and customer support, reducing transcription errors across more than 40 enterprise customers in six African countries.
Africa needs voice AI built around how Africans naturally speak, because poor handling of code-switching causes content loss and errors, said Tobi Olatunji, Founder and CEO of Intron.
At Branch International, Sahara-powered voice agents recovered over ₦1.2 million ($892) in delinquent loans within one week while improving customer engagement through local language support.
In Ogun State, the software is deployed across nine courts, with plans to expand to all 18 high courts.
In South Africa, health technology firm Audere Africa uses Sahara for voice-based clinical interactions with Zulu and Afrikaans speakers.
In Kenya, the platform converts 14-minute Swahili-English consultations into structured clinical notes in under 30 seconds.
Intron’s systems were trained on more than 150,000 hours of African-language audio sourced from over 53,000 speakers, covering 64 languages and 500 distinct accents, Olatunji said.
The regional enterprise voice market is seeing increased activity, with Nigerian startup Orinode also developing voice AI models for major local languages and code-switching applications.
Share
