Back in university, way back in 1990, I took a class called Black Studies. (At various points it was also called African American Studies, Afro-American Studies, Africana Studies… the name changed with the times, but the questions we explored were timeless.) One project we did still sticks with me today.
We recorded different students in the class reading the exact same sentence, unfortunately, I cannot remember the sentence anymore. We then played those recordings for other students and asked them to describe the speakers.
Here is what we found: everyone was uncomfortable describing people based on how they sounded. And they did it anyway.
Their discomfort did not stop them from forming impressions, on race, class, geography, education, just from the sound of someone’s voice. The point is not that these reactions are inherently malicious or deliberate. The point is that our brains are wired to react, even when we wish they were not.
Fast forward to today. AI is moving rapidly from text-based interfaces to voice-based interactions. Voicebots, digital assistants, and conversational AI are already answering customer service calls, facilitating public services, and increasingly, supporting health, financial, and education delivery across the Global South.
And that brings me back to our old classroom question—just with a new twist: What do AI voices sound like? And who decides?
We are seeing a growing trend of AI companies offering “neutral” or “unaccented” voices. But here’s the thing: there is no such thing. All voices carry accents. The only question is: whose accent sounds neutral to whom?
Tech companies, many still rooted in the U.S. and Europe, are shaping the default tones and patterns of speech for tools being used globally. Sometimes this happens without much thought. Other times, it is deliberate. Accent softening, real-time accent conversion, and even “accented AI” tools are being developed and deployed, often in the name of clarity or customer experience.
But let’s be honest: what we are really talking about is comfort. Whose voice sounds professional? Trustworthy? Competent? And whose does not?
A recent article in the Washington Post explores how Indian call centres are now training AI to change how real people sound by processing their speech in real time to “sound American,” complete with regional quirks and softened accents. The author describes how Indians in the call centres are provided “relief” from the stress of not being understood. And articles in CX Today and ZDNet talk about “accent reduction” as a feature, not a red flag.
At Digital Frontiers, we are paying close attention to this shift. And not just because we are a learning platform rooted in the digital economy, but because voice matters in how people feel seen, heard, and respected.
We have worked across 35+ countries and engaged learners from all across Africa. That means making real decisions about who our audiences hear. The voices we use in our courses are still recorded by real people —and they reflect the richness of African accents. Nigerian, Kenyan, Ugandan, South African. Francophone and Anglophone. Urban and rural. We do this on purpose.
We also hear from our alumni and facilitators, who navigate these questions every day. Whether leading a digital skills training in Lusaka, delivering a DPI (Digital Public Infrastructure) workshop in Addis, or producing a gender equity module for West African audiences, they understand that tone, cadence, and pronunciation affect trust and uptake just as much as content.
It is easy to think of AI as neutral. But it reflects the values, assumptions, and biases of the people building it. And when it comes to voice, we are entering a space where decisions made today, about how machines sound, will shape how billions of people are treated, understood, or overlooked tomorrow.
That does not mean every AI voicebot needs to sound like it is from Kampala or Kigali. But it does mean we should be asking better questions:
- Who gets to sound like an authority?
- What does a “professional” voice mean in a global context?
- And how do we make sure AI isn’t quietly reinforcing the very biases we have spent decades trying to undo?
Because at the end of the day, as my classmates and I learned all those years ago: we do respond to how people sound. Even if we do not want to. Even if we do not admit it. And as voice AI becomes more embedded in our daily lives, we need to understand what attitudes we might be perpetuating without even noticing. Let’s make sure we are listening, and let’s make sure we are hearing everyone.
Written by Todd Malone (Chief Executive Officer, Digital Frontiers)