Why TemboAI starts with a voice note

By Kelvin Omereshone

A builder needs Hausa voice notes in a WhatsApp agent. Here’s why that practical job belongs in the same product as our community missions and ambition to build our own models.

Gumzo and Griot share a voice waveform across TemboAI’s illustrated landscape.

A builder is working on a WhatsApp agent. People send it voice notes in Hausa, and the agent needs to understand what they said. Before it can do anything useful with the message, it needs a transcript.

That is where I want TemboAI to be useful. There is a recording to send, an output to inspect, and an application waiting on the result.

I’ve written before about not speaking Urhobo or Itsekiri. I wasn’t taught as a child, and I wish I had been. I want to practise, ask questions, and follow a conversation in the languages I wish I spoke. That is still a big part of why I’m building this.

Getting there means doing some very practical work. The Hausa agent is one place to begin.

Give the builder something to use

TemboAI’s Voice API offers hosted speech-to-text in English, Nigerian Pidgin, Yoruba, Igbo and Hausa. The Console gives you somewhere to try a recording, inspect the result, and see your usage before you wire the API into your own application.

Existing providers power the initial transcription service. I want to put a useful integration in a builder’s hands now and learn from the work they are trying to do. An actual application gives us a reason to get the formats, limits and cost right.

A voice note also gives you a very direct way to find a problem. A missing word can change a request. A name can come through incorrectly. The builder can inspect the transcript and decide what their application should do with it. We need to look at those details with the people using the service. That is how we find out what needs work.

New organisations start with 10,000 free API credits, granted once per organisation. After that, 10,000 credits cost ₦2,000. Transcription uses 5 credits per started second, with a 30-second limit per clip. A full 30-second clip costs 150 credits, or ₦30.

The starter grant gives you room to try the API in your own workflow. You can buy more credits when you need them.

I want us to build our own models

That ambition is a big reason the community work matters to me.

The way people actually speak is full of things a clean word list will miss. People switch languages. An expression carries a meaning you cannot get by translating each word separately. The person speaking may know exactly why one greeting fits and another sounds wrong.

I want that knowledge to have a place in what we build. It takes people who know the language, recordings worth keeping, accurate transcripts, and review by people who can judge the work.

That is what the missions are for. Contributors record their own voices and write what they said against a clear brief. Human review checks the contribution. Accepted work earns Points on Points missions or money on funded paid missions, under the terms shown for that mission. Cowries fund mission work; API credits pay for API usage.

Useful integrations tell us where the gaps are. Properly permitted, reviewed data gives us material to work on those gaps as we develop and evaluate our own models. I want the product and the community to keep informing each other.

We need to know what we’re allowed to use

A recording sent for transcription does not, by itself, give TemboAI a training licence. The permission to provide a service and the rights to improve a model have different scopes.

Community contributions follow the agreement recorded for the mission. Partner data can support model improvement from day one when its documented agreement permits that use and the required review clears it. Retention and withdrawal terms remain part of the work too.

I want us to be able to explain where a recording came from, what we may do with it, and why. That is part of building something I can stand behind.

Put it to work

The Hausa WhatsApp agent is a concrete use case for a much larger ambition. I want African languages to be practical choices for the people building the next set of applications. I want the people who speak those languages to help shape the models we build.

For now, we have a Voice API to put to work, a Console to work from, and community missions that people can contribute to. I want to see what you build with them.

Bring your voice to TemboAI.

See open missions