Voice technology built for how Kenya actually speaks.
From matatu banter to Swahili on the radio, Kenyan English in the office, and the vernacular spoken at home — one voice engine, not a flattened accent.
Generic Swahili TTS isn't Kenyan speech.
Most text-to-speech providers offer a single "Swahili" voice trained for a generic East African market. It doesn't code-switch and has never heard a vernacular language spoken naturally.
Usemi is built differently — trained on how people in Kenya actually talk: mixing English, Swahili, and vernacular in the same sentence, the way conversation really happens in Nairobi, Kisumu, Nakuru, and beyond.
Four registers. One engine.
Each track is trained on real speech patterns and its own spelling conventions — not translated or approximated from another language.
Sheng
Nairobi's street lingua franca — fast, code-switched, and evolving by the neighborhood.
Kiswahili
Standard Kenyan Swahili, for formal and everyday use across the country.
Kenyan English
The accent and rhythm of English as actually spoken in Kenya — not British, not American.
Vernacular
Kikuyu, Dholuo, Luhya, Kalenjin, and more — starting with the languages our community helps us prioritize.
Built for real deployments, not demos.
Fintech & customer service
IVR and voice bots that customers actually understand — loan reminders, balance alerts, fraud notices in the language they think in.
Health & NGO communication
Appointment reminders, triage lines, and public health messaging that reaches people in their first language.
Content & dubbing
Voiceover for Nairobi creators, ads, and e-learning — natural vernacular narration, not a foreign accent reading a script.
Accessibility
Screen readers and reading tools for users whose first language was never built into mainstream assistive tech.
Built in the open, on purpose.
Before recording a single clip, we published our own orthography spec — a shared spelling standard so informal speech gets written consistently instead of guessed at. It's revised in public as our dataset grows.
We're doing the unglamorous work first: consistent transcription, community-checked spelling, and honest documentation of what we don't know yet.
The road to v1.
We're early, and we'd rather show real progress than overpromise a launch date.
-
Orthography specDone · v0.1
A shared spelling standard published and being refined against real audio.
-
First datasetIn progress
Collecting and transcribing our first hours of Sheng audio.
-
v1 demoTargeting late 2026
First working model, rough but real — shared with early testers first.
-
Pilot partnersOpen now
Onboarding a small number of businesses and NGOs to test real use cases.
Good to know.
What languages will you support first?
We're starting with Sheng, since almost nothing exists for it today. Kiswahili, Kenyan English, and vernacular languages (Kikuyu, Dholuo, Luhya, Kalenjin, Ekegusii, and more) follow as our dataset grows.
When does it launch?
We're building our first dataset now. We'll share a working demo with our waitlist and pilot partners before any public launch — no fixed date yet, and we'd rather be honest about that than guess.
Can I use this for my business?
Yes — we're onboarding a small number of pilot partners now, particularly in fintech, health, and NGO communication. Reach out directly if that's you.
How is my email used?
Only to notify you when we launch or open a pilot. We don't share or sell it, and you can ask us to remove it anytime.
Talk to us directly.
General & pilot inquiries
Fintech, health, NGO, or content use case? We'd like to hear about it.
hello.usemi@zohomail.comFollow along
We're building this in public — updates, demos, and the orthography spec as it evolves.
Be first to hear it.
We're onboarding a small number of pilot partners and early testers ahead of public launch. Leave your email and we'll reach out.
NO SPAM · ONE EMAIL WHEN WE LAUNCH
We only use your email to notify you at launch — never shared, never sold.