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

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

  • Tech startup Sanas has been accused of racism for its ‘accent translation’ expertise.
  • AI and tech trade consultants say the startup’s mission legitimizes racism and is a type of “digital whitening.”
  • However some name middle brokers informed Insider they suppose the expertise would improve their day by day jobs.

“Accent translation” startup ​​Sanas has confronted allegations of racism and discrimination this previous week after being accused of manipulating non-American accents to sound “extra white.” The corporate makes use of speech recognition expertise to vary the person’s accent in near-real time; their major goal appears to be overseas name middle workers. 

Sharath Keshava Narayana, co-founder and COO of Sanas, denied that the startup’s expertise is discriminatory, telling Insider the corporate has all the time supposed to broaden its translation mannequin to incorporate different accents. The demo on its web site, the place the expertise interprets an Indian accent into a typical American one, showcases solely its preliminary mannequin, in line with Keshava Narayana.  

“It is not simply an American having hassle understanding somebody from India, however vice versa as nicely,” Keshava Narayana informed Insider. “As we hold scaling the product, and as we begin seeing increasingly more goal accents, we imagine that this shall be a localized answer.”

Sanas has been testing translation fashions in different international locations, like India and the Philippines, and plans to introduce accent translation to Latin America and Korea as nicely, in line with the startup.

Nevertheless, some consultants inside the tech trade have accused the startup’s product of being a type of “digital whitening.” AI and tech angel investor and CEO of women-led pc programming group FrauenLoop Nakeema Stefflbauer informed Insider the issue with Sanas’ response is that “accents sign energy and belonging.”

“When that is commercialized, there’s just one path by which everybody will get herded,” she mentioned. “It is not about comprehension a lot because it’s about consolation — for the teams that do not need to perceive, empathize, or have interaction with people who find themselves completely different in any respect. This tech does nothing to make sure the hypothetical name middle employee’s consolation.”

She added that till Sanas advertises this expertise to clients within the international South as a device to higher perceive and talk with People and Western Europeans, then “it is a one-way ‘answer’ that reinforces racialized hierarchies, whether or not that is supposed or not.”

AI and tech trade consultants and name middle employees spoke with Insider about what they noticed because the cultural prices, in addition to the potential advantages, popping out of Sanas. Whereas the corporate says its expertise’s goal is to make individuals on the telephone everywhere in the world sound extra “native,” Stefflbauer and others within the AI subject have considerations that it is one other step in the direction of homogenizing the startup world — one thing Silicon Valley has been repeatedly accused of perpetuating. 

“What is that this making an attempt to inform us when it comes to what the long run appears like and the way all of us must be experiencing voices on-line and speaking with individuals?” Stefflbauer mentioned. “Who’re the people who we’re meant to be speaking with, and who’re the people who we by no means hear from?”

Sanas founders

The Sanas founding workforce.


Tech trade consultants say accent ‘translation’ is a type of ‘digital whitening’

Sanas, which has raised $32 million in funding, says its goal is to assist individuals sound “extra native, globally” on its web site. In an interview with the BBC, Keshava Narayana mentioned that 90% of the corporate’s workers and all 4 of its founders are immigrants, and denied the criticism that the corporate is making an attempt to make the world sound “white and American.”

However Mia Shah-Dand, the founding father of Girls in AI Ethics and Lighthouse3, informed Insider that as an immigrant from India with a non-American accent, she discovered the announcement of Sanas “very triggering,” particularly as somebody who has been “teased and discriminated towards due to [their] accent.”

She mentioned the expertise is making an attempt to erase what makes individuals distinctive, and is telling them that who they’re “will not be adequate.”

“It looks like the whole lot in Silicon Valley, so long as it is legitimized by the Stanfords or the MITs, it is OK,” she mentioned. “Folks will settle for racism, settle for sexism, so long as the individuals doing it belong to considered one of these prestigious universities.”

Shah-Dand added that Sanas’ product reinforces an influence dynamic that “harkens again to the time of colonialism.” Moderately than addressing the basis causes of racism and discrimination,  “accent translation” leans right into a type of “whitening” — an influence dynamic seen in lots of international locations that had been traditionally colonized the place individuals felt pressured to make their pores and skin whiter to suit European magnificence requirements.

“It is Silicon Valley’s model of digital whitening,” Shah-Dand mentioned. “As a substitute of expertise making the world a greater place, it is amplifying, it is serving to, it is simply monetizing all of that hate and racism quite than really making an attempt to repair something.”

Stefflbauer informed Insider she discovered Sanas’ expertise “actually disappointing and unsettling,” particularly within the rising tradition of bringing one’s “complete self to work.”

“Solely sure individuals can convey their complete self, everybody who’s outdoors of this legendary norm will not be invited to convey any of themselves,” she mentioned, referencing the 2018 darkish surrealist comedy “Sorry to Trouble You,” the place a Black telemarketer finds that new doorways {of professional} success are opened to him solely after he adopts a “white” sounding voice.

“It is actually one other instance of simply what we’re up towards when it comes to making an attempt to get the tech trade and the services and products that come out of it to mirror the actual world,” Stefflbauer mentioned.

She added that she would not see how this expertise would really tackle racial biases in any method.

“It would not even try to method that in its answer,” she mentioned. “It mainly affords help and canopy for individuals who will behave badly with accented people that they’ve any form of dealings with to proceed to take action.”

Name middle brokers informed Insider they face racialized hostility

Sanas’ founders mentioned they got here up with the concept for the startup after a school good friend from Stanford underperformed at his job at a name middle due to his thick Central American accent.

Name middle brokers Insider spoke with mentioned their jobs may be brutal — and are doubly so if they’ve a racially-distinctive accent or title.

“Sadly on this world, there are lots of people who’re going to really feel like they’re higher than you or will select to talk down at you after they hear your accent,” Dafina Swann, who has labored at name facilities for greater than 5 years, mentioned.

Swann, who’s from Trinidad and Tobago, mentioned she obtained a number of “hostile” and “unfavourable feedback” from callers who demanded they communicate with somebody who’s American. She had additionally heard of cases the place colleagues had been referred to as racist names, just like the n-word, and informed that they had been “not human, however Black.”

To attenuate the racialized backlash they face, some name middle brokers informed Insider that they already attempt to mimic clients’ accents, and even change their names. Generally, the directive to vary their names comes from the brokers’ managers or employers.

“After I began introducing myself as Michael O’Connor, my efficiency rankings from buyer surveys elevated — all inexperienced, inexperienced, inexperienced,” Osama Badr, a name middle agent from Egypt, informed Insider.

Sanas co-founder Keshava Narayana mentioned he additionally had an analogous expertise when he labored at a name middle, the place he underwent six weeks of accent coaching and was informed to vary his title to “Ethan.”

“There are specific incidents that stick with you for a very long time, and this was considered one of them,” he informed Insider.

Some concern that manipulated voices might sign a homogenized future in tech

Shah-Dand mentioned she is not satisfied by defenses of the expertise, saying that persons are uncovered to and may perceive completely different accents, however it’s solely as a result of name middle employees are handled as “lesser than” that they obtain unfair abuse. 

“There are a number of people who’ve very heavy accents, like Boutros Boutros-Ghali for instance,” Shah-Dand mentioned, referring to the previous Secretary-Common of the United Nations. “However as a result of they’re in highly effective positions, you make an effort to grasp.”

Stefflbauer mentioned in her work, she’s all the time fascinated about what digital life shall be like 10, 20 years into the long run, and he or she’s involved about what expertise like Sanas foretells. 

“I see increasingly more examples of a digital life the place nobody is Black, nobody is brown, nobody has an accent, nobody has a historical past outdoors of the North American legendary best,” Stefflbauer mentioned. “And the query is: will we need to export this mentality and convey this distress to everybody? As a result of that is undoubtedly what that is.” Different AI applied sciences, together with facial recognition applied sciences, have additionally confronted accusations of racism and homogenization.

“Who could be comfy taking a selfie on Instagram and having their face mechanically altered to appear to be somebody of a distinct race?” she mentioned. “That is primarily what that is.”

However name middle workers who must take care of racist feedback of their day by day jobs say {that a} answer just like the one Sanas affords may very well be a blessing.

“It undoubtedly would’ve made my job simpler. Everybody needs to be understood,” Swann mentioned. “There is a job that must be achieved, and if there’s one thing that may be carried out to make that job simpler, then that is nice.”