When Accents Are Erased: Are We Fixing Bias or Fuelling It?

I remember watching a cartoon version of “The Sneetches” by Dr Seuss when I was a child. It was a musical special, animated in that classic saturated and jagged style that was popular in 1973, and I found it fascinating. (If you’ve not seen it, it’s available on YouTube).

It tells the story of a town of yellow bird-like creatures called Sneetches, some of whom have green stars on their bellies. Those without stars are discriminated against by those with stars, until a salesman – Sylvester McMonkey McBean – arrives in town with a machine that can add a star to the belly of those without one for three dollars.

The service is hugely popular with the non-starred Sneetches. But the Star-Bellied Sneetches are in danger of losing their status.

Not to worry. As well as the Star-On machine there is a Star-Off machine that removes the identifier (for ten dollars), and they can remain “special”.

But McBean allows those who have recently been starred to go through machine too, leading to all the Sneetches randomly running between the two machines until “neither the Plain nor the Star-Bellies knew whether this one was that one… or that one was this one… or which one was what one… or what one was who.”

The Sneetches highlights how society places value on arbitrary differences, shaping social hierarchies in the process. Today, technology offers new ways to erase or alter these distinctions – sometimes in pursuit of fairness, sometimes for convenience, and sometimes for profit. One such innovation is a real-life Star-On/Star-Off machine: AI Accent Translation.

This is a tool that changes how people sound in real-time, removing an auditory marker of identity much like McBean’s machine removed (or added) stars. But what does it mean when voices themselves are modified to fit expectations?

What is AI Accent Translation?

AI accent translation for contact centre agents uses artificial intelligence to neutralise accents in real time. The developers claim that the technology can help reduce miscommunications, improves customer satisfaction, and save average handling time – a major resourcing cost in contact centres.

Together with background noise cancellation, contact centre providers – including Krisp, Sanas and Teleperformance – have begun its roll-out. Teleperformance’s Deputy-Chief Executive Officer, Thomas Mackenbrock, believes the software “creates more intimacy, increases the customer satisfaction, and reduces the average handling time: it is a win-win for both parties.”

The AI tool itself raises ethical and social questions. But the deeper discussion about human psychology, in-group/out-group bias, and technology’s role in shaping communication is even more compelling.

The psychology of familiarity

Teleperformance suggest that “when you have an Indian agent on the line, sometimes it’s hard to hear, to understand.”

Whilst it might read as such, this isn’t strictly xenophobic. Humans are wired to feel more comfortable with people who sound, look, and behave like them. Linguistic similarity builds trust and rapport, which is why accents can be a barrier (or a bridge) in communication.

Accent often points to social information like a speaker’s social class, where they’re from, and their sexuality. Often people will rely on accent to make judgements about unrelated traits like intelligence or reliability. This can lead to what linguists have called Accent Bias or linguistic discrimination.

People also tend to categorise themselves and others into in-groups (people like “us”) and out-groups (people unlike “us”). This can lead to favouritism toward those who share linguistic, cultural, or national traits. This is Social Identity Theory.

Accent Bias: Why We Judge People by Their Voices

Accent bias is the tendency to judge people’s intelligence, trustworthiness, or competence based on how they speak. Studies show that people with non-native or regional accents are often perceived as less credible, less intelligent, or less “professional” – even if their language skills are excellent.

In the UK, you might have experienced this in advertising, where certain regional accents that are considered more friendly, trustworthy or competent are used to promote products and services, or to report the news.

There is some truth to this. Gluszek & Dovidio (2010) found that listeners rated speakers with non-native accents as less competent and less socially appealing, even when their speech was clear. A separate study found that people were less likely to believe factual statements when spoken in a foreign accent (Lev-Ari & Keysar, 2010). These have implications for hiring decisions, promotions, customer interactions, and even perceptions of truth and competency.

Interestingly, there is evidence that people naturally modify their speech (consciously or unconsciously) to match their audience in order to gain approval. AI Accent Translation could be seen as an extreme version of this, removing differences instead of bridging them.

This bias is not just social but psychological. Studies indicate that when people hear an accent different from their own, their brains register it as “cognitive effort.” This means they subconsciously associate it with difficulty, even if they understand the words perfectly. This can create an instinctive preference for voices that sound familiar, reinforcing in-group biases and subtly feeding into xenophobic attitudes.

Social Identity Theory: Why We Prefer People Who Are Like Us

Humans have an innate tendency to sort themselves into groups based on shared traits, and language is one of the strongest markers of identity. Social Identity Theory (SIT), developed by psychologist Henri Tajfel, explains how people categorise themselves and others into in-groups (those who share similarities) and out-groups (those who are different).

These divisions influence how we perceive and interact with others, often leading to favouritism toward those who sound like us and subtle discrimination against those who don’t. (For a real-world example, see my article on the Zelensky-Trump press meeting).

The language aspect of ingroup identification is deeply engrained. Toddlers prefer to interact with people who sound like them, suggesting accent bias is not just learned (Kinzler et al., 2007), and people with standard or native accents tend to be rated higher in status and competence compared to those with non-standard accents (Fuertes et al., 2012).

If British (or native English-speaking) customers subconsciously feel more comfortable with voices that sound “like them,” AI-modified accents may reduce friction in interactions. However, might they also reinforce the idea that non-native accents are a problem to be fixed?

Prejudice: When identity gets lost or confused

At what point does accommodating communication turn into erasing identity? There’s a thin line between making speech clearer and reinforcing biases against non-native accents.

Prejudice isn’t just about conscious dislike or discrimination. It often operates at a subconscious level, shaping how we perceive and interact with others before we even realise it. Linguistic prejudice is a prime example of this: we instinctively make snap judgments about people based on how they speak, often without recognising the biases at play and causing xenophobic views.

At its core, xenophobia is the fear or distrust of those perceived as “outsiders” – people who seem different due to their nationality, culture, or even the way they speak. While overt xenophobia manifests as discrimination or hostility, a more subtle version appears in the form of linguistic prejudice.

Research shows that accents don’t just signal where someone is from. They also influence how much we trust them, how likable we find them, and how much authority we assign to their words. In many Western societies, native or “standard” accents (such as Received Pronunciation in the UK or General American in the US) are seen as more professional or prestigious, while non-native accents are sometimes perceived as a barrier to understanding – even when the speaker is perfectly fluent.

The future: Tech’s Role in Shaping Perceptions

AI is already modifying how we sound, look (deepfakes, filters), and even think (algorithmic influence). This article has likely been shown to you (or not) based on AI predicting whether you will find it useful and whether this content is “worthy” of inclusion in your feed.

The world is globalised and getting smaller. Is real-time accent translation a natural extension of globalisation, or does it risk sanitising cultural differences?

Could this tech eventually go beyond accents and alter tone, speech patterns, or even emotion to match the listener’s expectations? Would that make interactions more artificial or more human?

In Douglas Adam’s The Hitchhiker’s Guide to the Galaxy, there is a small, bright yellow fish, which can be placed in someone’s ear in order for them to be able to hear any language translated into their first language instantly. When this Babel fish moves from science fiction to science fact (and we’re not far away from that point), we create a world where neither geography nor language fluency are a barrier to advisers handing contact centre calls.

But what nuance and complexity do we put at risk?

Final thought

At the end of The Sneetches, after running back and forth between the machines, the birds are left in confusion – no longer sure who originally had stars and who didn’t. Their differences had been erased; but so had their sense of identity.

This struggle over perceived superiority is not so different from the way accents function in human society. Just as the Sneetches judged each other based on external markers, people make subconscious judgments about intelligence, competence, and credibility based on how someone speaks.

And now, with AI accent translation, we have a modern-day Star-On/Star-Off machine. One that doesn’t alter physical traits but, instead, modifies the invisible markers of identity embedded in our voices. It doesn’t eliminate bias; it merely disguises it, smoothing over the surface without addressing the deeper issue.

The push for clarity and efficiency in communication is understandable. There are potential benefits in improved accessibility for people with hearing difficulties and clearer communication in high-stakes situations.

But when technology erases accents to meet the expectations of those in power, it raises a deeper question: Are we striving for genuine understanding, or are we just making differences more palatable for the sake of OPEX savings and share price?

Much like the Sneetches, we risk ending up in a world where accents – at least on the phone – are neutralised into a single, homogenised voice. And in doing so, we might lose something far more valuable than efficiency: the ability to embrace and appreciate the full range of human diversity.

By altering the accents of call centre workers in real-time, AI is not addressing a communication problem but rather a perception problem. It’s attempting to work around deep-rooted prejudices rather than addressing them directly.

Instead of challenging the bias that certain accents are “less clear” or “less professional,” technology is adapting to these prejudices, reinforcing the idea that some voices are more acceptable than others. AI is amplifying our human biases.

The question then becomes: Are we using AI to remove barriers – or just making them less noticeable?

The key moral of the story of The Sneetches is that superficial differences shouldn’t determine social hierarchy or worth. But there is another, more subtle, moral.

The only real winner in this story is Sylvester McMonkey McBean: the salesman selling the machine and the promise of a better future for all.

References:

Fuertes, J. N., Giles, H., Lawless, K. A., & Houts, P. S. (2012). Language attitudes and social identity: The case of British and American accents. International Journal of Applied Linguistics, 22(3), 349-367. https://doi.org/10.1111/j.1473-4192.2012.00306.x

Gluszek, A., & Dovidio, J. F. (2010). The role of accent and dialect in social identity and intergroup perception. Journal of Language and Social Psychology, 29(1), 111-132. https://doi.org/10.1177/0261927X09351756

Kinzler, K. D., Dupoux, E., & Spelke, E. S. (2007). The native language of social cognition. Proceedings of the National Academy of Sciences, 104(30), 12577-12580. https://doi.org/10.1073/pnas.0705343104

Lev-Ari, S., & Keysar, B. (2010). Why don’t we believe non-native speakers? The influence of accent on credibility. Journal of Experimental Social Psychology, 46(6), 1093-1096. https://doi.org/10.1016/j.jesp.2010.05.023

Tajfel, H. (1982). Social identity and intergroup relations. Cambridge University Press.