At the Deep Learning Indaba in Lagos, a 21 country hackathon tested voice AI built for Africans who switch languages mid sentence.
At this week's Deep Learning Indaba in Lagos, more than 120 teams from 21 countries crowded into the Sahara CodeSwitch Hackathon to test voice AI that can keep up with how Africans actually talk: switching languages mid-sentence.
Intron, the Nigerian voice-tech company behind Sahara, used the Indaba to launch v2.5 of the platform. The release adds bilingual code-switching speech recognition across 12 African languages, including Zulu, Hausa, Swahili, Luganda, Yoruba, and Nigerian Pidgin, and expands total coverage to 31 languages. It also includes what Intron calls the world's first African trilingual speech recognition model, covering Kinyarwanda, English, and French for Rwanda. The company has filed US patents on the trilingual code-switching technology.
Hackathon teams compared Sahara against Gemini, AssemblyAI, and Whisper on code-switched African conversations and reported Sahara performed significantly better, though slightly slower, per the competition results. Fintech, healthtech, agritech, and edutech apps dominated the builds.
Intron's published benchmarks claim Sahara outperformed Gemini, ElevenLabs, and Meta across all 12 languages on code-switched speech recognition, and beat ElevenLabs and Gemini on language-mixing text-to-speech in 11 of 13 languages. Those figures are vendor-published, not independently audited.
The competition continues through September. An Nvidia-hosted masterclass runs August 28.