A Ghanaian voice AI passes its real usability test when a conversation can move from one person to another without forcing everyone to restart, simplify their language, or guess what the assistant heard. Passing the phone across the room should feel ordinary, because that is how many family conversations already work.
In 1970, Apollo 13’s crew faced rising carbon dioxide inside the lunar module. The command module had square lithium hydroxide cartridges. The lunar module used round openings. NASA engineers in Houston, led by Ed Smylie, had to devise an adapter from materials already available to the crew, before anyone knew the fix would work. NASA History documents the problem and the improvised solution.
The danger was not a lack of equipment. The crew had cartridges. The problem was that the pieces could not meet in the form the moment demanded.
A conversation changes hands before it changes topic
Picture a phone on speaker in a home in Kumasi, London, or Bronx. One person begins: “Please help me put this message together for Auntie. Tell her sɛ yɛbɛba later.”
Then the phone goes to someone else across the room.
“Add that I’ve spoken to Kofi too, but he said the parcel has not come.”
Then an older relative leans in with the missing detail. The conversation is still one conversation. The language may move between Twi and Ghanaian English. The speaker changes. The urgency stays.
A voice assistant that only performs when one person speaks in tidy, isolated turns has missed the setting where it has to earn trust. The test is not a polished demo. The test is the interruption, the correction, the handoff, and the sentence that begins in Twi before landing on an English word everyone in the room uses.
That is why mixed-language conversation matters beyond transcription. The assistant has to keep the thread while people speak naturally. For a closer look at what gets lost when that thread breaks, read What Happens When Your Voice Assistant Loses the Thread Between Twi and English?.
The handoff reveals the hidden assumptions
Most voice products are designed around a single, stable speaker. That assumption stays invisible until real life pushes against it.
A family member may interrupt to correct a name. Someone may speak from farther away. A child may repeat the question in a different mix of languages. The person holding the phone may need to stop, listen, then continue. These are ordinary actions, not edge cases to explain away.
Nkomo is built for natural Twi and Ghanaian English conversation by voice or text. In hands-free voice mode, you hold to speak, can interrupt naturally, and hear the reply. The useful test is simple: after the phone moves, can the next person continue the task in the words they would actually use?
That test should also include failure. If the app cannot understand a part of the request, the person using it needs to see that clearly. Nkomo shows errors instead of swallowing them. In a shared conversation, a visible error gives the room a chance to correct the message before a bad answer quietly becomes the basis for a family decision.
Privacy needs to survive the same real-world test
Passing a phone also changes who is present. A request that began as a private message can become a conversation with siblings, parents, or visitors nearby. That makes clear control over voice data part of usability.
Nkomo lets you choose whether cloud use is never allowed, asked about each time, or allowed for the current session. Its history stays on the device under your control, and turning history off purges it immediately. You can export your data or delete your account in one tap.
Those controls do not make a shared room private. They make the product’s behaviour clear before someone speaks. Ghana’s communications ministry has identified ethics, data privacy, governance, and inclusion as central responsibilities for AI. A voice product serving Ghanaian conversations should make those responsibilities visible in the moment people need them.
Build for the room, then test in the room
Apollo 13’s adapter worked because the people solving the problem started with the actual interfaces available to the crew. They did not ask the crew to find a different cartridge.
Voice AI needs the same discipline. Test it with a phone being passed between people. Test a request that changes from Twi to English halfway through. Test an interruption. Test a correction from the person who did not begin the conversation. Test what the screen says when the system cannot complete the request.
The result is a more honest standard: the conversation should accommodate the people in the room. People should not have to reshape their language and habits around the assistant.
Comments
No comments yet.