AI Startup’s ‘Accent Translation’ Technology Accused by Critics of Racism

• Tech startup Sanas has been accused of racism for its ‘accent translation’ technology.
• AI and tech industry experts say the startup’s mission legitimizes racism and is a form of “digital whitening.”
• But some call center agents told Insider they think the technology would enhance their daily jobs.

“Accent translation” startup ​​Sanas has faced allegations of racism and discrimination this past week after being accused of manipulating non-American accents to sound “more white.” The company uses speech recognition technology to change the user’s accent in near-real time; their primary target seems to be foreign call center employees. 

Sharath Keshava Narayana, co-founder and COO of Sanas, denied that the startup’s technology is discriminatory, telling Insider the company has always intended to expand its translation model to include other accents. The demo on its website, where the technology translates an Indian accent into a standard American one, showcases only its initial model, according to Keshava Narayana.  

“It’s not just an American having trouble understanding someone from India, but vice versa as well,” Keshava Narayana told Insider. “As we keep scaling the product, and as we start seeing more and more target accents, we believe that this will be a localized solution.”

Sanas has been testing translation models in other countries, like India and the Philippines, and plans to introduce accent translation to Latin America and Korea as well, according to the startup.

However, some experts within the tech industry have accused the startup’s product of being a form of “digital whitening.” AI and tech angel investor and CEO of women-led computer programming group FrauenLoop Nakeema Stefflbauer told Insider the problem with Sanas’ response is that “accents signal power and belonging.”

“When this is commercialized, there’s only one direction in which everyone gets herded,” she said. “It’s not about comprehension so much as it’s about comfort — for the groups that don’t want to understand, empathize, or engage with people who are different at all. This tech does nothing to ensure the hypothetical call center worker’s comfort.”

She added that until Sanas advertises this technology to customers in the global South as a tool to better understand and communicate with Americans and Western Europeans, then “it’s a one-way ‘solution’ that reinforces racialized hierarchies, whether that’s intended or not.”

AI and tech industry experts and call center workers spoke with Insider about what they saw as the cultural costs, as well as the potential benefits, coming out of Sanas. While the company says its technology’s aim is to make people on the phone all over the world sound more “local,” Stefflbauer and others in the AI field have concerns that it’s another step towards homogenizing the startup world — something Silicon Valley has been repeatedly accused of perpetuating. 

“What is this trying to tell us in terms of what the future sounds like and how we all ought to be experiencing voices online and communicating with people?” Stefflbauer said. “Who are the people that we’re meant to be communicating with, and who are the people that we never hear from?”

Sanas

Tech industry experts say accent ‘translation’ is a form of ‘digital whitening’

Sanas, which has raised $32 million in funding, says its aim is to help people sound “more local, globally” on its website. In an interview with the BBC, Keshava Narayana said that 90% of the company’s employees and all four of its founders are immigrants, and denied the criticism that the company is trying to make the world sound “white and American.”

But Mia Shah-Dand, the founder of Women in AI Ethics and Lighthouse3, told Insider that as an immigrant from India with a non-American accent, she found the announcement of Sanas “very triggering,” especially as someone who has been “teased and discriminated against because of [their] accent.”

She said the technology is trying to erase what makes people unique, and is telling them that who they are “is not good enough.”

“It feels like everything in Silicon Valley, as long as it’s legitimized by the Stanfords or the MITs, it’s OK,” she said. “People will accept racism, accept sexism, as long as the people doing it belong to one of these prestigious universities.”

Shah-Dand added that Sanas’ product reinforces a power dynamic that “harkens back to the time of colonialism.” Rather than addressing the root causes of racism and discrimination,  “accent translation” leans into a form of “whitening” — a power dynamic seen in many countries that were historically colonized where people felt pressured to make their skin whiter to fit European beauty standards.

“It’s Silicon Valley’s version of digital whitening,” Shah-Dand said. “Instead of technology making the world a better place, it’s amplifying, it’s helping, it’s just monetizing all of that hate and racism rather than actually trying to fix anything.”

Stefflbauer told Insider she found Sanas’ technology “really disappointing and unsettling,” especially in the growing culture of bringing one’s “whole self to work.”

“Only certain people can bring their whole self, everyone who’s outside of this mythical norm is not invited to bring any of themselves,” she said, referencing the 2018 dark surrealist comedy “Sorry to Bother You,” where a Black telemarketer finds that new doors of professional success are opened to him only after he adopts a “white” sounding voice.

“It’s really another example of just what we’re up against in terms of trying to get the tech industry and the products and services that come out of it to reflect the real world,” Stefflbauer said.

She added that she doesn’t see how this technology would actually address racial biases in any way.

“It doesn’t even attempt to approach that in its solution,” she said. “It basically offers support and cover for people who will behave badly with accented individuals that they have any kind of dealings with to continue to do so.”

Call center agents told Insider they face racialized hostility

Sanas’ founders said they came up with the idea for the startup after a college friend from Stanford underperformed at his job at a call center because of his thick Central American accent.

Call center agents Insider spoke with said their jobs can be brutal — and are doubly so if they have a racially-distinctive accent or name.

“Unfortunately in this world, there are a lot of people who are going to feel like they’re better than you or will choose to speak down at you when they hear your accent,” Dafina Swann, who has worked at call centers for more than five years, said.

Swann, who’s from Trinidad and Tobago, said she received a lot of “hostile” and “negative comments” from callers who demanded they speak with someone who’s American. She had also heard of instances where colleagues were called racist names, like the n-word, and told that they were “not human, but Black.”

To minimize the racialized backlash they face, some call center agents told Insider that they already try to mimic customers’ accents, and even change their names. Sometimes, the directive to change their names comes from the agents’ managers or employers.

“After I started introducing myself as Michael O’Connor, my performance ratings from customer surveys increased — all green, green, green,” Osama Badr, a call center agent from Egypt, told Insider.

Sanas co-founder Keshava Narayana said he also had a similar experience when he worked at a call center, where he underwent six weeks of accent training and was told to change his name to “Ethan.”

“There are certain incidents that stay with you for a long time, and this was one of them,” he told Insider.

Some fear that manipulated voices could signal a homogenized future in tech

Shah-Dand said she isn’t convinced by defenses of the technology, saying that people are exposed to and can understand different accents, but it’s only because call center workers are treated as “lesser than” that they receive unfair abuse. 

“There are a lot of folks who have very heavy accents, like Boutros Boutros-Ghali for example,” Shah-Dand said, referring to the former Secretary-General of the United Nations. “But because they’re in powerful positions, you make an effort to understand.”

Stefflbauer said in her work, she’s always thinking about what digital life will be like 10, 20 years into the future, and she’s concerned about what technology like Sanas foretells. 

“I see more and more examples of a digital life where no one is Black, no one is brown, no one has an accent, no one has a history outside of the North American mythical ideal,” Stefflbauer said. “And the question is: do we want to export this mentality and bring this misery to everyone? Because that’s definitely what this is.” Other AI technologies, including facial recognition technologies, have also faced accusations of racism and homogenization.

“Who would be comfortable taking a selfie on Instagram and having their face automatically altered to look like someone of a different race?” she said. “That’s essentially what this is.”

But call center employees who have to deal with racist comments in their daily jobs say that a solution like the one Sanas offers could be a blessing.

“It definitely would’ve made my job easier. Everyone wants to be understood,” Swann said. “There’s a job that needs to be done, and if there’s something that can be implemented to make that job easier, then that’s great.”