{
  "id": 3475999,
  "title": "Intron’s new voice model can follow African speakers who switch languages mid-sentence",
  "url": "https://urgent.news/2026/08/26/introns-new-voice-model-can-follow-african-speakers-who-switch",
  "topic": "business",
  "section": "Business",
  "published": "2026-08-26T09:00:51.000Z",
  "source": {
    "name": "TechCabal",
    "slug": "techcabal",
    "url": "https://techcabal.com/2026/08/26/intron-voice-ai/"
  },
  "original_language": "en",
  "account": "The African continent is home to many multilingual speakers, often blending languages within a single sentence. Traditional speech recognition systems typically transcribe English while ignoring other languages, leading to losses in meaning. This is where Intron, an Africa-focused voice technology company, comes in with its new voice model, Sahara v2.5.\n\nSahara v2.5 is designed to handle code-switching, the ability to follow a speaker moving between languages mid-sentence, in 12 African languages including Zulu, Hausa, Swahili and Luganda. It also introduces the world's first African trilingual speech recognition model, supporting conversations in Kinyarwanda, English and French in Rwanda. This is a significant step forward as it specifically addresses the unique challenge of code-switching in African languages.\n\nIntron asserts that code-switching is a distinct technical problem that global labs have generally overlooked. The company claims that its dedicated data, training and evaluation have produced superior results. For instance, on their own benchmarks, Sahara v2.5 showed an average word error rate of 34.3% across 12 languages of code-switched African speech, compared to 53.8% for a Google AI model. In practical terms, this means Sahara gets about one word wrong in three, while the other model gets more than one in two wrong.\n\nFounded in 2020 by Tobi Olatunji and Kunle Asekun, Intron has raised $1.6 million in pre-seed funding to tackle the issue of voice AI that doesn't work across the African continent. Their focus on code-switching is crucial because speech recognition systems typically struggle with this problem due to the complexity of separating language switches in audio.\n\nIntron trains Sahara directly on mixed-language speech, treating switching as ordinary and learning which transitions are possible in each language pair. Their approach is more effective as it uses acoustic evidence and context across the whole utterance rather than separating languages and transcribing them individually.\n\nSeveral organizations across Nigeria, Kenya, South Africa, Uganda, Rwanda and Ghana are already using Sahara. One example is Branch International, a fintech lender that reported significant improvements after integrating Sahara-powered collections agents.\n\nLooking ahead, Intron has published an Africa Voice AI Report, highlighting that while collecting African language data is important, there are many other barriers to reliable voice AI. These include research capacity, system integration and deployment expertise. They argue that Africa needs AI built for how Africans really speak, without flattening accents or requiring speakers to translate themselves for machines.",
  "summary": "Intron has released Sahara v2.5, a set of voice AI models built for when Africans use multiple languages in one sentence.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}