'The goal is not to erase accents': Why masking software can make sense

"The goal is not to erase accents, but rather to support human-to-human communication.”

So says TELUS Digital in explaining which it is using AI accent-“softening” technology as part of a broader suite of AI tools in its contact centre operations.

The tech works by modifying the pronunciation of call centre agents in real time, shifting it closer to a standard North American or British English pattern while preserving the agent's underlying voice.

While critics raise questions about possible discrimination, worker autonomy and consumer deception, proponents say employees appreciate the accent “masking” tools because they reduce miscommunication, shorten call times and shield agents from accent-based bias and verbal hostility.

How accent masking works

Sometimes people are unfamiliar with a particular accent and it can be hard for them to understand the speaker, says Shawn Zhang, co-founder and chief technology officer at Sanas in San Jose, Calif., one of the vendors in the space.

“We're hoping that our technology can help bridge that gap and make people feel connected and have conversations just flow a lot smoother.”

Sanas uses proprietary neural networks to analyze speech at the phoneme level — the smallest unit of sound — and translate pronunciation in real time, he says: “All this is done in 200 milliseconds for almost all of our deployments.”

The processing runs directly on the agent's laptop; audio never passes through Sanas servers.

“We do that just so that people are very secure with their data and their data is never sent outside of their ecosystem,” he says, plus it helps avoid delays: "You can't be lower latency than directly on device.”

Zhang compares the technology to noise cancellation, now a broadly accepted workplace tool.

"I don't think you're trying to mask that you're in a noisy scenario or you're trying to hide something there. You're using it because it helps with clarity in that conversation."

The software doesn’t alter words, tone or personality, and agents can’t select a different voice profile, according to Zhang.

"We do not try to change the identity of the speaker. So, in terms of your voice timber, your pitch, your voice fingerprint — we really try to preserve that… You sound like yourself. We're merely trying to translate the phonetic pronunciations.”

TELUS embraces accent softening

TELUS Digital — which uses Tomato.ai for accent softening — frames the technology as one piece of a wider AI-enabled contact centre strategy, from training and real-time knowledge support to post-call coaching and quality assurance.

"The objective is simple: better customer experiences, shorter resolution times and improved support for our team members,” said Richard Gilhooley, director of communications, in a statement to Canadian HR Reporter.

The company uses accent-softening technology to “support clarity and ease of understanding” for both customers and employees, he said.

“In practice, we’ve found conversations are more efficient when customers and agents can understand each other clearly the first time. Agents spend less time repeating information, customers spend less time on calls and, importantly, many agents have shared that the AI technology has helped create more positive and productive customer interactions.”

But there are limits, according to Gilhooley.

"It enhances pronunciation clarity to support ease of understanding — it does not alter the substance of what is being said, nor does it change an individual's personality or identity.”

The tool is not currently used in customer interactions with TELUS subscribers in Canada, he said, though other AI tools such as background noise cancellation are active.

Discrimination against accents

Zhang says the inspiration for Sanas came from a friend who worked as a contact centre agent. Customer complaints about his accent appeared in his performance reviews, lowering both his ratings and his pay.

"We thought that was incredibly unfair. People should be evaluated based off their intelligence or work ethic or talent... They should not need to change their accent for a job that they wish to succeed in."

Christy Zhou Koval agrees. Social science research is clear that people with non-native accents face negative evaluations in hiring, promotion, funding decisions, court decisions and beyond — driven by both outright bias and genuine cognitive disfluency, says the assistant professor of organizational behaviour at Queen's University Smith School of Business who studies accent-based discrimination.

“We're very sensitive to how people sound and we're just kind of hardwired to use language to figure out if someone is an in-group or out-group member. And so if you hear someone who has a different accent, you think they're an out-group member, and that kind of just triggers bias.”

When it comes disfluency, heavier accents can be harder to understand, she says, making it harder for our brains to process the information, which can also lead to a more negative evaluation.

Zhou Koval notes that not all accents are treated equally.

"Some accents are perceived as more prestigious, like the European accents," she says, while others rank lower based on history and general stereotypes about different countries and ethnicities.

Upsides and downsides to AI

Accent-softening tools could theoretically help on both fronts, says Zhou Koval — reducing the discrimination and rudeness directed at agents, easing anxiety for non-native speakers and improving comprehension.

“That could be one potential upside to it… would this give them a shield, potentially, to make them feel more confident? Possibly.”

Her deeper concern is what the technology implicitly communicates.

"It kind of perpetuates this idea that there is a standard accent, that there is a neutral accent,” says Zhou Koval. “It does not help to fundamentally reduce the bias. We need to be exposed to a variety of different accents and the way people speak to know that it's normal… it's part of someone's cultural heritage that should be celebrated."

She also raises a more practical question about whether accents is the right problem to solve. Sometimes, what gets attributed to an agent's accent is actually a function of call quality, reception or other infrastructure issues: "A lot of times it's just the quality of the call themselves — the reception, other infrastructure."

ROI of accent 'softening'

Zhang says Sanas tracks outcomes across three groups: the business, the customer and the agent. On the business side, clients report lower average handle times, higher call closure rates and improved customer satisfaction scores. The dynamic is straightforward, he says: fewer requests to repeat, fewer mid-call hang-ups, fewer transfers.

"Previously, you had calls that would be hung up in the middle. Sometimes customer might ‘rage-quit,’ hang up and just [say] ‘I can't understand you, I need to be transferred to a new representative.'”

On the employee side, agent satisfaction scores tend to shift significantly once the tool is in place, even among those who were initially resistant, says Zhang.

"Definitely there's a lot of skeptics in the very beginning [thinking] ‘I'm not sure if this is really going to change how I work’ or ‘Is this really going to change my performance?’”

But post-deployment surveys, which Sanas encourages all enterprise clients to conduct, typically tell a different story, he says.

“Believe me, the contact center industry, the customer service industry, it is not an easy industry. You really have to build a tough skin — I've definitely seen that firsthand.

“So, when you are being verbally abused eight hours a day — it happens a lot — I think something like this can be a huge breath of fresh air.”

Consent and communication 

For HR professionals, how the tool is introduced may matter as much as the tool itself. Zhang says Sanas typically begins with a 30-day pilot, with training provided to help agents understand the purpose and expected results.

“And if people want to opt out, then I would say more freedom and more choice is more powerful," he says, though he acknowledges some clients prefer to deploy it across the board.

Zhou Koval says that a blanket deployment approach is the wrong one.

"The customer service reps themselves should have the choice to decide — do they want to opt in or opt out of having their accent masked? Because there are individual differences. Some people feel their accent is part of their heritage. They [may not] feel comfortable having it erased, they don’t feel comfortable having it erased. But someone else could say… ‘Knowing that I have an accent makes me feel anxious, so having this mask, it gives me confidence.’”

Transparent communication with staff about the rationale also matters, she says. "It's a way of talking about why we're doing this but also showing we're giving the agents their own agency to decide."

Disclosing software to customers

One of the more unsettled issues is whether it might be considered deceptive — or even discriminatory — if customers are not told about the accent masking.

A group of unions representing roughly 32,000 workers at Bell, Rogers and TELUS recently testified before the House of Commons Standing Committee on Industry and Technology, calling for federal limits on how AI is introduced in the sector. Unifor telecommunications director Roch Leblanc said the use of AI to mask the accents of offshore agents could "mislead Canadians" into thinking they were speaking with Canadian-based employees.

Zhou Koval draws a direct parallel to existing disclosures around call recordings, when callers are told “This conversation is being recorded for training purposes.”

"Why not disclose and say, ‘This conversation has been subject to voice processing?’” she says. "Essentially, if you don't tell customers you're masking the accent, you're basically deceiving customers. How would they react if they find out?"

Zhang says the issue of disclosure is up to the regulators and employers using the tech, and TELUS has indicated it will move toward transparency, according to Gilhooley.

“We will be providing clearer disclosure messaging to customers regarding the use of AI-assisted technologies in contact centre interactions, including tools designed to improve voice clarity and transmission quality."