Why I started Sauti Labs
It began with a speech model that didn’t understand the people I was building for. What I found when I went looking for alternatives, and why I think the distance between African-language research and usable products is worth building a company around.
Sauti Labs started with a practical problem, not a plan to start a company.
In July 2026 I was building Scribe AI, an offline system that listens to a church service and puts the scripture being referenced on screen. Everything in it depends on speech recognition, and for that I used OpenAI’s Whisper. On the African languages and accents the system needed to handle, it disappointed me. I have written separately about why: Whisper’s training data is overwhelmingly English, Swahili appears with roughly five hours of transcribed speech, and Dholuo — the language that first drew me into this — is not among its languages at all.
So I went looking for something better: a model I could access, integrate, and rely on. That search took longer than I expected, and it changed what I was looking for.
Arriving in a field that already existed
The first thing I learned is that I was arriving late to serious, sustained work. Masakhane (opens in a new tab), a grassroots community whose stated mission is to strengthen and spur NLP research in African languages, has spent years building open datasets, models, and benchmarks. Its MasakhaNER project, for example, released the first large, publicly available named-entity recognition dataset for ten African languages, with the code and models alongside it.
Closer to home, the Maseno Centre for Applied AI is collecting voice data in five Kenyan languages, Dholuo among them, as part of African Next Voices — a multi-country effort, funded mainly by the Gates Foundation, to build speech datasets for African languages. In South Africa, Lelapa AI (opens in a new tab) builds products directly: its Vulavula platform transcribes and translates contact-centre conversations in languages such as isiZulu and Sesotho, and its researchers have released InkubaLM, an open small language model for five African languages. Global labs have moved too: Meta’s speech models now claim coverage of more than sixteen hundred languages.
This is not a field waiting to be discovered. It has years of difficult, often under-resourced work behind it, and whatever I build stands on that work.
The distance I kept running into
And yet, as a developer with a concrete problem, I kept running into the same distance. A dataset existed, but not a model trained for my use. A model listed my language, but its error rates on that language were far from what a product needs. A model worked, but its licence ruled out commercial use, or it needed hardware my users did not have. The products that did exist had been built, sensibly, around particular languages and particular markets — which is how products should start, but which left the speech I needed uncovered.
Over time that experience hardened into a conviction. I sometimes put it bluntly:
Africa doesn’t need another conference paper. It needs products that are accessible to the people and work for them in the most efficient and natural way.
I should be clear about what I mean, because it is easy to read as a criticism of research. It is not. Papers, datasets, and benchmarks are how a field learns what works; without them there would be nothing for products to stand on, and research and product are not opposites. My point is about where the work tends to stop. Knowledge has its largest effect when someone carries it the rest of the way: into a model trained on real speech, served behind an interface a developer can call, packaged so that a business can deploy it, and good enough that the person on the other end stops noticing it is there. That last stretch is often nobody’s job. It is the stretch I want to work on.
Why speech
I chose speech and voice because I think they will become one of the main ways people interact with AI and with digital services generally — and many African languages are underrepresented in precisely the systems that make those interactions possible.
That underrepresentation is more than a technical inconvenience. It limits what people can build: a developer who wants to create a voice service for their own community — a clinic line, an advisory service for farmers, a support agent for a local business — cannot do it if there is no capable model underneath. It shapes how people reach services and how naturally they can speak to the technology around them. And when I listen to AI-generated speech in African languages today, I often hear the gap: pronunciation that is slightly off, an accent that belongs to someone else, a delivery without the rhythm of a real speaker.
Something I watched recently made the stakes concrete. It was a demonstration of a system that let a person who had lost the ability to speak talk again, in a synthetic version of their own voice. Programmes such as ElevenLabs’ Impact Program now offer this kind of voice restoration to people living with conditions like ALS. What stayed with me was not the technical achievement, impressive as it is, but what was returned to that person. I found myself asking what the equivalent would mean for someone whose language no such system yet speaks.
What I’m building towards
ElevenLabs is the clearest example I know of carrying research the rest of the way. Its work spans speech generation, voice cloning, dubbing, and conversational agents, and what I find inspiring is less any single model than the range of things its products let people do. I want Sauti Labs to make the same connection between serious technical work and useful products — for languages, accents, and settings that existing systems were not designed around, and with the linguistic, social, and technical challenges those bring.
Sauti Labs (opens in a new tab) is an AI research and products company. Its research teams work across speech, language, language resources, model intelligence, and responsible AI, and its first product, Sauti Cloud, is in early access. The ambition is broader than any one capability — speech recognition, natural speech synthesis, voice cloning and dubbing, conversational systems and voice agents, and the datasets, models, and infrastructure beneath them — and broader than any one language. Dholuo shaped how I came to this work, but the work is meant for African languages generally.
None of that is finished. Those capabilities are a direction, not a description of what exists today, and I am still learning: from the communities and companies that came before, and from the work itself.
Why take the risk
Speech and language AI is expensive and research-intensive. It needs data, compute, funding, expertise, and time — exactly the things many young Africans who see these problems clearly do not have. I decided to take the risk anyway, because I think the problems are important enough to justify being early and under-resourced, and because I want people like me to be among those who build these systems, not only among those who receive them already designed around other languages and other contexts.
I want African languages to take a full part in whatever AI becomes. That is why Sauti Labs exists.
Sources
- Masakhane (opens in a new tab) — masakhane.io
- MasakhaNER: Named Entity Recognition for African Languages (Adelani et al., 2021) (opens in a new tab) — Transactions of the ACL
- African languages for AI: the project that’s gathering a huge new dataset (Marivate, Adebara, Wanzare) (opens in a new tab) — The Conversation
- Lelapa AI and Vulavula (opens in a new tab) — lelapa.ai
- InkubaLM: A small language model for low-resource African languages (opens in a new tab) — arXiv
- Omnilingual ASR: Open-Source Multilingual Speech Recognition for 1600+ Languages (opens in a new tab) — arXiv
- Robust Speech Recognition via Large-Scale Weak Supervision (Radford et al.) (opens in a new tab) — arXiv
- ElevenLabs Impact Program announcement (opens in a new tab) — ElevenLabs