Accents are deeply personal, shaped by geography, language, socioeconomic class, gender, and by one’s own identity. Just as voice prints can identify individuals, accents reveal where they are from: in ways that can impact how they’re heard, perceived, and reacted to and engaged with by others.
Class and accents are central to the plot of the film My Fair Lady, based on George Bernard Shaw’s play Pygmalion, in which, to win a bet, Dr. Henry Higgins tries to turn London Cockney flower-seller Eliza Doolittle into an upper-class “lady.”
My family loved the film and had the soundtrack’s LP. For good reason. We’re from northern England, which is noted for a distinct accent (and its class variations), that one can hear on many movies and shows, notably Coronation Street, which is reportedly the world’s longest-running TV soap opera.
Not surprisingly, accents can impede customer experience (CX) in the contact center, causing what can be termed as “accent friction” and leading to misunderstandings and frustration on both sides of conversations.
Contact centers have long sought to address this issue through accent training. Now there are also automated AI-based tools that can help people understand each other across accents.
However, these methods and tools have their own challenges and controversies, like the risk of reinforcing customer biases against agents who speak “differently.” AI applications also hold the potential to fail, creating friction points of their own.
Sanas, a speech AI company, has developed “Real-Time Accent Translation” technology. To learn more about accent issues, and the benefits and challenges of accent translation, we had a virtual conversation with company CEO and co-founder, Sharath Keshava Narayana.
Q. Let’s set the baseline. What kinds of understandability challenges do both customers and agents face during voice interactions? How do these issues affect CX and customer loyalty?
Customers and agents struggle with more than just accents. Noise, call quality, fatigue, and speaking pace all influence how clearly we understand one another.
When people can’t hear, or be heard, they repeat themselves, interactions take longer, and frustration rises on both sides. Over time, these moments affect trust and reduce customer satisfaction and loyalty. Clear communication isn’t just a nice-to-have; it’s foundational to a positive CX.
Q. In terms of scale and impact on CX, how do agent and customer accents compare with other challenges, like audio quality?
Technical issues like jitter or low bandwidth can distort a sentence, but accent-related misunderstandings can distort the entire relationship.
Infrastructure fixes can resolve audio quality issues. Accent friction is different; it’s a human barrier that persists even when every technical variable is perfect. It often shows up in subtle ways, like hesitation or misinterpretation.
Q. It has been pointed out that geographical/regional accents especially, have been fading away thanks to mass media and migration. So, in the face of this, is the accent issue through the contact center still strong and relevant and if so, why?
You’re right about global media and mobility flattening regional accents in many social and professional contexts, but contact centers operate under very different conditions.
These are short, high-stress, audio-only transactional interactions, often between people from very different cultural backgrounds. In those environments, subtle phonetic differences can increase cognitive load and create real friction.
“Clear communication isn’t just a nice-to-have; it’s foundational to a positive CX.” —Sharath Keshava Narayana
Customers aren’t reacting to accents themselves so much as the effort required to understand and be understood. That added effort compounds quickly, increasing frustration, slowing resolution, and lowering overall satisfaction.
As global CX has scaled, the volume of cross-accent interactions has grown faster than accents have converged. In practice, accent-related friction hasn’t faded; it’s become more concentrated in the moments that define CX.
Accent Translation, Accent Modification
Q. Describe accent translation. What is it and what is its objective?
Accent translation enhances clarity by analyzing the sound waves in incoming speech and adjusting them, so listeners hear a clearer, more familiar pronunciation along with the original tone and intonation.
When it comes down to it, accents are simply tiny variations in sound waves. This technology helps eliminate noise, reverb, and echo in real time to ensure the most human, authentic experience possible on both ends of the call.
Q. Is accent translation different from accent modification and if so, how?
In my view, “accent modification” and Sanas’ term “accent translation” can be used to describe the same basic idea: adjusting speech so it’s easier for listeners to understand across accents.
The difference is mainly historical and contextual. Accent modification is an older phrase rooted in human coaching and may imply permanently changing the way someone speaks through training.
We coined the term accent translation to describe technology that transforms the acoustic patterns of speech while enabling the speaker to talk naturally and preserving meaning, intent, and voice identity.
In practice, both terms can describe the same outcome, but we prefer accent translation to make it clearer that the change happens with technology, not in the person.
Q. Hasn’t accent translation been around for some time, such as through agent training?
Traditional accent training can take months and often pressures people to change how they naturally speak. Accent translation technology enables clarity instantly and consistently while also preserving the speaker’s unique voice characteristics like pitch, cadence, and emotion.
AI Application
Q. We understand that you use AI in your solution. What are its benefits and describe how it achieves them? How is AI technology deployed and how are agents trained to use it?
The biggest benefit is that AI improves clarity instantly, with no special training, months of coaching, or pressure to change how someone speaks.
With this technology, agents can speak naturally while the system improves clarity in the background, so they can focus on the customers, not their accents, thereby shortening handle times and reducing escalations.
Unlike traditional accent training, AI adapts in real time to each customer, accent, and environment. This leads to faster resolution. With a single click, AI-enabled accent translation runs on the agent’s existing device, with low latency and no complex setup.
Q. What are the downsides of AI-driven accent translation? Isn’t AI prone to errors, like through hallucinations?
All AI needs is thoughtful design, and accent translation is no exception. It must be engineered carefully to avoid sounding artificial or overcorrected.
Unlike generative AI, accent translation isn’t creating new words or meanings. Rather, it enhances the natural speech that’s already there. And because it’s speech-to-speech rather than speech-to-text or text-to-speech, there’s no intermediate transcription step where meaning can get lost or distorted.
The real challenges are ensuring consistency across diverse accents and noisy environments, which is why real-world evaluation and continuous performance monitoring matter so much.
Customer Accent Issues
Q. You’ve discussed accent translation from the agent end. But what about from the customers’ end? Can it be used there so that agents can understand them better?
The technology could theoretically work in both directions: helping agents understand customers as well as the reverse. Sanas already offers omnidirectional noise cancelation, which removes background noise on both sides of the call.
“The goal is to break down communication barriers without asking agents to speak in a way that doesn’t feel authentic.”
Our goal is always to remove communication barriers to promote understanding, and any future application would need strong ethical guardrails to ensure the purpose remains grounded in clarity, inclusion, and understanding.
Q. To what extent does accent modification or translation, regardless of the technology used, risk concealing customer bias instead of confronting it?
Could this method also contribute to the broader forms of identity neutralization, including gender or regional or national differences?
It’s important to acknowledge that accent translation isn’t a cure for bias. The goal is to break down communication barriers without asking agents to speak in a way that doesn’t feel authentic.
We designed this technology to enhance clarity, not erase identity. Our vision has always been to enable agents to speak comfortably in their natural voice while the technology simply adjusts the acoustic patterns that can create friction. It’s about being understood, not being someone else.
Recommendations
Q. What steps would you recommend that contact centers take to improve mutual understanding between agents and customers? Which actions should they prioritize and how can AI-driven accent solutions best support those efforts?
I always advise to start with the basics: clean audio, reliable equipment, and agent coaching on pacing and empathy. Then add technology that reduces friction, such as real-time accent translation, for clearer speech.
We’ve found the technology works best when it complements strong fundamentals, turning good conversations into effortless ones that help both agents and customers alike feel understood.