Clear communication is crucial in today’s global workforce, but for some, it’s easier than for others.
On Tuesday, Krisp, an audio AI software developer, introduced an Accent Conversion tool to break down language barriers by altering how a person’s voice comes through to listeners.
While the technology promises inclusivity, it also raises concerns about cultural biases, cultural identity, and what it means to “sound professional.”
Krisp’s AI Accent Converter aims to enable more inclusive and equitable hiring practices by eliminating perception bias and ensuring that a person’s accent does not influence how professional or trustworthy customers perceive them.
According to Krisp co-founder Arto Minasyan, a recent study revealed a significant perception gap between customers and contact centers.
“In our study, 79% of U.S. customers said they ask non-native English speaking agents to repeat themselves at least once per call, yet only 33% of contact centers believe this happens,” Minasyan told Decrypt. “This gap leads to lower satisfaction and frustration for both sides of the meeting.”
Minasyan noted that companies either reject candidates with strong accents, disqualifying an average of 64% of agents, or have to invest in costly accent training.
From then to now
Launched in 2017 by Minasyan and Davit Baghdasaryan, Krisp is an AI-based audio processing software company based in Berkeley, California. It specializes in real-time noise and voice suppression technology.
Minasyan explained that the tool doesn’t distinguish between accents but instead converts speech to match a specific target accent. During the conversion process, Krisp captures characteristics from the user’s voice and incorporates them into the final output.
“This approach helps maintain a more natural-sounding voice and preserves the emotional tone of the original input,” he said.
While users can toggle the AI accent converter on and off in real time, Minasyan said the option to adjust the conversion level is not currently available.
Even though AI accent tools may improve employment opportunities, they also carry risks of misuse.
These concerns prompted the FBI to warn in May 2024 about scammers leveraging AI-powered voice software to carry out phishing scams.
Leading up to the 2024 Presidential election, the FCC banned Robocalls that use AI-generated candidates’ voices after a deepfake of then-President Joe Biden was used to try and trick voters in New Hampshire into not voting.
When asked about the potential for misuse by cybercriminals, Minasyan acknowledged the risks, noting that Krisp’s tools, like cellphones, laptops, or the Internet, can be exploited despite their intended purpose.
“While we recognize that misuse is a concern, it is not feasible to fully monitor how the tool is used without compromising user privacy,” he said.
Still, there is hope that the technology will be used primarily for good rather than for other purposes.
“We haven’t directly consulted with sociolinguists, but we’ve tested the technology in real-life settings with customers before making it widely available,” Minasyan said.
The founder pointed to several internal metrics, including a 99% improvement in a net promoter score, a more than 26% increase in sales conversion rates, and a 12.2% boost in its agent promoter score, showcasing a “higher agent confidence.”
“Seeing these improvements on both the user and customer sides has strengthened our confidence that the technology does not introduce major cultural or business downsides,” he said.
Edited by Sebastian Sinclair
Generally Intelligent Newsletter
A weekly AI journey narrated by Gen, a generative AI model.
