A call centre can now change how an agent sounds while they speak, without changing the agent. That is a real choice with real trade-offs, and it is worth thinking through before adopting it.
Here is the concrete case. In early May 2026, Telus Digital, the customer experience division of Telus, deployed AI that alters the accents of its customer service agents in real time. When an agent speaks, the audio passes through an AI model that adjusts pronunciation and softens a heavy accent before the customer hears it. The agent is a real person. The voice the customer hears has been processed.
What the technology actually does
The tool Telus uses comes from a company called Tomato.ai. It uses speech-to-speech models that take the agent's live audio and re-render it with a softened accent, while keeping the agent's natural rhythm and voice. It is not a script change and not a chatbot. It is a filter on a human voice, applied during the call.
Vendors in this space also bundle noise cancellation and clarity improvements. The accent change is the part that draws attention, because it touches identity, not just audio quality.
Why companies do this
The stated reason is clarity. Telus describes the goal as reducing "accent-related friction" so customers understand agents more easily. For a contact centre, that is a measurable thing: fewer repeated sentences, shorter calls, higher satisfaction scores.
There is a second reason that vendors state openly. Accent softening lets a centre hire from a wider pool, including offshore agents whose accents some customers find hard to follow. That widens recruitment and can lower cost. So the technology solves a clarity problem and a staffing problem at the same time, which is part of why it is appealing.
The questions it raises
This is where CX leaders need to slow down. The deployment drew criticism quickly. Labour groups called it deceptive because customers were not told the voice had been AI-processed, and a union representative raised it with members of Parliament. Linguists described the practice as linguistic profiling. Rogers and Bell have both confirmed they will not use accent-modifying software.
Three questions sit underneath the criticism. The first is disclosure. If a customer's perception of who they are speaking to is being shaped by a tool, is staying silent about it honest. The second is agent dignity. Telling an agent their natural voice will be altered to be acceptable sends a message, and how it is introduced matters a great deal. The third is whether it solves the real problem. If customers struggle to understand an agent, the cause might be call audio quality, training, pace, or background noise. Accent softening treats one possible cause and may mask the others.
If you are considering it
The technology is not automatically wrong. Clearer calls genuinely help customers, and clarity tools that remove noise or improve audio quality are uncontroversial. The accent-change use is the one that needs care, because it affects trust and how staff are treated.
The takeaway for CX leaders
AI that changes how agents sound is now a real product, in real use, and it does deliver clearer calls. It also carries trust, transparency, and dignity costs that do not show up in a satisfaction score. Treat it as a decision that needs a disclosure policy and agent agreement, not as a quiet settings change. If the only way it works is by keeping customers and agents in the dark, that is the strongest sign to reconsider.
Phidea looks at the underlying problem in the accent gap in voice AI.