Apotierioluwa Owoade had a problem he could not stop thinking about.
He had spent time working at Aforevo, a local streaming and dubbing firm in Lagos, Nigeria, and his experience stayed with him. During his time at the company, from 2022 to 2023, he saw firsthand how cost-prohibitive the dubbing industry could be.
Translating a film into another language costs upwards of $500,000 for a full production, according to Owoade.
Yet, beyond the cost, something frustrated him even more: the lack of nuance that most translators failed to capture in the local tongue. Voice actors, overstretched and underpaid, flattened the emotional texture of scenes they were recording. The existing software tools were no better.
He had seen Yoruba rendered so poorly that the phrase “I am pregnant” came out flatly as “I have a ball,” he explained, his face visibly grimacing over our video call.
He wanted to fix it. He called his friend David Mac-Asore, who was a Computer Engineering undergraduate at the time and a software developer.
Owoade and Mac-Asore had known each other for years, a friendship anchored partly through shared work at Living Faith Church Worldwide International, one of Nigeria’s largest churches. Since 2022, the two have collaborated on projects to bridge the language divide between the church’s English and French-speaking congregations at its headquarters in Ota, Ogun State, in south-western Nigeria, said Mac-Asore.
When Owoade pitched his idea, Mac-Asore was in, but they agreed they needed someone steeped in machine learning.
Assembling a team
Mac-Asore reached out to two of his former schoolmates from Covenant University, a private Christian university in Ota: Maryann Nnaji and Emmanuel Ibiang. Both had graduated in 2024.
The four got on a call. At the time, they did not even have a name for what they were building. According to Owoade, they called it the Hagen Project, a name that made me chuckle.
The ‘Hagen Project’ eventually evolved into Reedapt in 2025.
Nnaji brought the machine learning depth the team needed.
Before joining, she had built a sign language-to-speech and text model as part of her undergraduate thesis, working through the full pipeline from data collection to training, deployment, and testing.
She had noticed the friction the hard-of-hearing (HOH) community faced in everyday interactions and wanted to use technology to address it.
She had to put the work on hold, partly due to a data gap that would later feel very familiar when she began working on Reedapt.
According to Nnaji, most research on sign language recognition was built on Western contexts, not Nigerian or African ones.
“When I started the project, it was just a way to actually see how technology could be a way of bridging this gap,” Nnaji said. “To prove something to myself that this is applicable.”
Ibiang, Reedapt’s product engineer, arrived with a different but equally critical instinct: an obsession with usability. Where the AI engineers on the team reached for accuracy, Ibiang optimised for the user.
“Can the average Joe use your product without having to be walked through?” Ibiang asked rhetorically, almost as if he was expecting a response. “Ease of use of the product—that’s what my role optimises for.”
For a product as technically complex as Reedapt, that perspective has been the team’s internal check against building something impressive that nobody can use.
In a four-man small team where Mac-Asore and Nnaji are the technical engineers, and Ibiang is the product specialist, Owoade was described by his teammates as the person with the ideas.
The four of them, all fresh out of school and under 25, decided to build together.
From a translation tool to a dubbing platform
I first met Owoade and Mac-Asore at the Builders Summit by Founders Connect in May 2025, a networking event for early-stage technology founders held in Lagos, Nigeria.
At the time, they told me they wanted to build a translation tool that allowed users to move between different languages over text or audio, without needing to learn a new language.
I remember joking that their ambition would eventually put Duolingo out of business.
More seriously, I asked them why this needed to exist in a world where we already have DeepL.
They didn’t have a clear answer. I sensed they were still finding the edges of it.
Since that conversation, the vision has sharpened considerably.
Reedapt is now focused on becoming the go-to dubbing and real-time multilingual streaming platform for Nollywood filmmakers, churches, and African content creators who want their work to travel further than the English language will take it.
The startup, Owoade said, has signed two enterprise dubbing contracts with a Nollywood gospel producer, with those projects expected to be completed before the end of 2026.
Reedapt’s paying customers today are a mix of Nollywood producers and churches. It currently serves over 200 active users.
About 94% are individual creators on its consumer tier, while the remaining 6% are enterprise clients who generate the majority of revenue, said Owoade.
Reedapt makes money by charging subscription fees.
Pricing is structured in dollar-denominated tiers: a free plan offering up to 60 minutes of usage, a creator plan at $11 per month, and higher tiers at $39 and $99.
The decision to price in dollars was deliberate, said Owoade.
The team had experimented with Naira pricing early and found that it undercut their credibility with both users and potential investors.
“Most of our costs are not in Naira,” he said. “So it would be significantly disadvantageous to charge in Naira.”
The team is now aiming to reach 50,000 users before the end of 2026.
How the four graduates built Reedapt
Building a speech technology product for African languages is not the same as building one for English or French.
According to Owoade, the tools that exist in the market were not built with Africa as a priority.
“Big Tech isn’t really building with Africa for Africa as a priority,” he said.
“Apple released AirPods with a live translation feature, and you see their focus is on languages like English, French, and Spanish.
In fact, they seem to prioritise Spanish even more because they’re catering to Latin American users. So, if I had to say it, Nigeria is out of the conversation.”
Reedapt is trying to change that, but the foundation of any good model is data, and the data problem for African languages is acute.
According to Nnaji, training data for the languages Reedapt needs is either very scarce or too poor in quality to be useful.
The machine learning engineer likened the problem of training a model to raising a child.
If you show a child only one breed of dog, they will struggle to recognise other breeds, said Nnaji, visually describing the challenge.
“If the volume of what you’re feeding it is not enough,” she said. “Your model still performs poorly.”
The team is currently working with open-source licenced data while building pipelines to collect its own higher-quality training sets.
The model architecture builds on existing foundations rather than starting from scratch.
Reedapt had previously used credits from ElevenLabs, one of the biggest names in text-to-speech and dubbing, but has since moved toward independence.
Before the end of Q2 2026, the team plans to release its first in-house model, designed to handle the specific challenges of African speech: code-switching between English and Yoruba mid-sentence, incantations, spiritual language that should not be translated, and names that existing models routinely mangle.
“We aim to have our first model released to the world that is capable of handling a wide variety of tasks, but only for niche cases,” said Owoade.
“It [will be] quite independent. But even [today], we’re not giving our customers’ data to any third parties, and we’re also aware of the fact that it’s a competition.”
The infrastructure built around the model is just as important as the model itself.
Mac-Asore explained one layer of the pipeline that often goes unnoticed: before any audio reaches the model, the team runs it through a diarisation and cleaning process, which helps to separate different speakers and strips background noise from audio.
“We have proper audio engineering before it gets to the model,” said Mac-Asore. “To give it a better form of the initial speech.”
Accuracy is tracked using word error rate, a metric that measures how far a model’s output strays from a correct transcript, and supplemented with human evaluation where team members listen back and flag anything that sounds wrong to a native ear.
According to Owoade, Reedapt’s dubbing already performs at the same accuracy level as ElevenLabs for general content, and exceeds it on African-specific nuances.
The platform includes an editor that allows clients to correct pronunciation errors, and Nollywood producers can upload scripts ahead of a dubbing job to give the model additional context. Voice cloning is handled entirely in-house, said Owoade.
The cost of building all of this is high. Compute costs for training machine learning models are “outrageously pricey,” said Nnaji.
Collecting data at the quality the team needs requires controlled environments and specialised recording equipment.
Across the entire development process, the four graduates have bootstrapped Reedapt to the tune of over $50,000, according to Owoade, drawing on cloud credits from service providers and AI model providers to subsidise a significant portion, with the remainder coming out of their own pockets.
The startup is now seeking to raise $500,000 to accelerate product development.
The weight of building something personal
For Owoade, building Reedapt is an expression of something he has carried for a long time.
He studied Foreign Languages and Literatures at university and worked in the dubbing industry before starting the company.
His reason for building Reedapt, he said, stemmed from a deeply held conviction that African creators deserve to tell their stories in their own languages and be heard across the world without having to route everything through English.
He remembers growing up in a school environment where speaking Yoruba, his native language, was treated as an infraction.
“They dictated the fact that in most schools, when you spoke Yoruba, it was vernacular,” he said. “That’s our mother tongue.”
Reedapt, in his telling, is the rebuttal to that, and his ambition for scale and storytelling in Africa for Africans is unmistakably passionate.
By 2026, he and his team aim for Reedapt to be the top dubbing and real-time multilingual streaming tool in Africa.
Within five years, the startup plans to expand into India and the Philippines, markets where content creation is booming, and creators face revenue disadvantages similar to those of their African counterparts.
By 2030, the team is targeting support for 500 languages.
Owoade asked his teammates to call him a cockroach, a half-serious reference to the insect’s famous resilience.
The journey, he says, has been brutal in ways that are hard to fully convey: days with no clear sense of where the next round of resources would come from. But he is not stepping back.
Yet scaling Reedapt, and betting that a dubbing tool built out of Africa is one people want, is a large bet from a team still counting credits and pulling from their own pockets.
The only thing that would make them stop, Owoade said, is if 50,000 customers told them the product was not worth building.
So far, not one has, he added.
And if, despite everything, Reedapt does not survive, Owoade says he has thought about that, too.
He plans to leave an open-source version behind for someone else to carry forward.
“That’s how passionate we are about the problem we’re solving,” he said. “[Reedapt] is not simply a tool; it’s very personal for us, and that’s how we approach the mission at Reedapt.”
from TechCabal https://ift.tt/vwaLTIB
via IFTTT
I am the CEO of 9JAGISTLAND.My name is Adedeji Timothy and my co-owner is Oyekola Damilare.This blog is born out of our passion to educating to people in all aspect of internet life.We are enthusiastic about writing posts and articles online.
Write your views on this post and share it. ConversionConversion EmoticonEmoticon