From Noise to Meaning: Krisp’s Next Step in Voice AI

Listen to the AI generated audio article. 

Krisp became widely known during Covid, when remote work turned video calls into a daily routine and background noise into a real workplace problem. For many people, it was the tool that made barking dogs, street noise and home-life interruptions disappear from Zoom calls. It was one of those rare products that solved a new problem at exactly the right moment, and it helped an Armenian tech company secure a clear place in the global voice technology market.

That moment has passed, but the conditions that helped Krisp grow haven’t. Remote and hybrid work didn’t disappear after the pandemic, they became part of everyday professional life. And with them came a continued demand for tools that make digital communication smoother and more efficient. Over the past few years, Krisp has grown well beyond noise cancellation, building products around what it knows well: how people sound, and how they understand one another online.

Its new feature, Accent Understanding, launched on March 3, is one such example. It’s designed to help listeners understand accented English in real time, without asking speakers to change the way they talk. That may sound like a natural next step, but it also suggests something bigger: Krisp has moved beyond cleaning sound and into shaping how speech is heard. That’s a bold move, and one that comes with ethical questions of its own.

Who Is Accent Understanding Built For?

Accent Understanding works on the listener’s side. It is meant to make accented English easier to understand and follow in real time, without asking or expecting the speaker to adjust the way they talk. That may sound like a small tweak, but it’s addressing what has become a familiar challenge of global work: when audio is perfectly clear, yet understanding still takes effort. In Krisp’s demos, you can automatically hear how the accent is improved without changing the speaker’s voice entirely; it makes the speech sound closer to standard American or British English.

In fast-moving teams, people can lose a surprising amount of time simply trying to fully catch what others are saying. It’s rarely dramatic enough to be called a problem, but over time it adds up, slowing conversations, breaking concentration and pulling attention away from the actual work. This is the kind of challenge Krisp wants to reduce, which is why the feature is likely to appeal most to companies working across languages, accents and markets.

Origins of the Idea

Asti Pili, Krisp’s Head of Product and Growth, says that once a company commits to a certain product direction, new ideas often begin to appear naturally around it. She goes on to say that features like this usually emerge at the intersection of what the company knows how to build and what customers are still struggling with. Once a team goes deep into a certain product direction, researchers and engineers become so familiar with the technology that they begin to see new applications. At the same time, customer conversations reveal adjacent problems that have not yet been solved. Accent Understanding, Pili says, grew out of exactly that overlap.

Before this launch, Krisp had already been working on accent conversion, a tool created for international call centers, where agents often speak English with a wide range of accents. That product aimed to help agents convert accented English into a version that would be easier for customers to understand. As the team continued refining the technology behind it, another question followed: if this could be done from the speaker’s side, could something similar be built for the listener’s side instead?

That question became the basis for Accent Understanding.

The Criticism That Comes With Voice Technology

Still, innovation rarely arrives without criticism, and when the subject is people’s voices, the response tends to be especially personal.

When Krisp first introduced accent conversion, it drew immediate backlash. For many, the idea of changing someone’s accent felt uncomfortable, even offensive. Critics argued that an accent is not just a way of speaking, but part of identity, background and ethnicity. And beneath that sits an even harder question: who decides what the “standard” accent is supposed to be? What does “neutral” English even mean?

Accent Understanding inevitably touches the same nerve. The technical process may now happen on the listener’s side rather than the speaker’s, but the broader idea still raises similar concerns. In both cases, the technology is intervening in how accented speech is heard.

Pili does not dismiss that sensitivity. She says Krisp takes this kind of feedback seriously and approaches criticism as part of the process of building voice technology and notes that the company has never stopped a launch because of backlash alone. Having said that, they have made changes to products in response to customer concerns.

In the case of Accent Understanding, Pili argues that the goal is not to “fix” the speaker. The feature is positioned instead as a tool that helps the listener better follow what is being said. In her view, the purpose of a meeting is to communicate an idea clearly, and any tool that reduces friction can help keep the focus on the content of what someone is saying rather than on how they sound.

There is also a practical argument behind it: time. Learning to alter pronunciation in a second language can take years, and for many people it never fully happens. Researchers in second-language speech have long argued that the sound system shaped by a person’s first language continues to influence how they hear and produce sounds in another language throughout life. In simple terms, people tend to filter unfamiliar sounds through the categories they already know. That helps explain why pronunciation can remain distinctive even when grammar and vocabulary become highly advanced.

From Krisp’s perspective, that matters. The company argues that reaching the kind of clarity promised by Accent Understanding through traditional means would often require accent coaching, significant time, and money, and even then, the result might not be as effective as the tool aims to be.

Pili’s final point is broader. AI has already become deeply embedded in how people communicate through writing, editing, and translation. So why, she asks, should audio be treated differently? If AI can already help people write more clearly and efficiently, why should it not also help them speak and listen more effectively?

What Comes Next

For now, what is clear is that Accent Understanding fits neatly into the direction Krisp has been moving in for years. From its early breakthrough in noise cancellation to AI note-taking, meeting recording, and now accent-related tools, the company has consistently built around one central idea: improving communication.

Accent Understanding feels less like a side release and more like a natural continuation of where Krisp has been heading for years. The company built its name by making speech easier to hear. Now it is trying to make speech easier to follow.

Whether users will welcome that shift remains to be seen. In the end, it’s the users who decide whether a product truly works in the real world. If the demand is there, then the feature clearly responds to a real need, even if the ethical questions around it do not disappear.

Creative Tech