Sorry, I don’t speak American
Is accent translation digital whitening? (Photo: Sanas)
My grandma grew up speaking Chinese, and speaks with a thick, Chinese accent when she speaks in English. I used to have a hard time deciphering what she was trying to say when she spoke in English. My eldest sister, however, had an ear for it. Whenever my grandma had a hard time voicing her thoughts, my sister would relay her message to us.
It wasn’t until much later on in life that I realized it wasn’t about having a way with words, but rather having empathy—the empathy to understand how a person speaks, why a person speaks that way, and how to best communicate with them without changing who they are.
I am sad to report that today’s tech doesn’t echo that same sentiment.
Lost in translation
Tech startup Sanas has long been developing “accent translation” AI for call-center agents all over the world, and have already secured $32 million in funding. The company uses speech-recognition technology to manipulate non-American accents to make them sound “more white,” all in near real-time.
The company trained a machine learning algorithm to recognize a person’s speech on one end, and then output the same words with an accent chosen from a list. The supported accent list includes American, British, Spanish, Indian, and Filipino.
The AI was designed to reduce the prejudice of having a non-American accent and improve the overall welfare of call center agents.
But I wasn’t entirely hot on this tech. Based on the marketing on their website, it’s pretty clear that the target market is POCs who work as call center agents. See the interactive demo for yourself.
All of this reeked of “digital whitening,” which initially did not shock me given that it’s coming from a Silicon Valley startup. After all, tech created by Americans will most likely service American ideals. Just a couple of weeks ago, we pointed out the Western-centric database of the US company Open AI’s DALL·E 2 program.
You can only imagine my surprise reading about the diverse team and brand story behind Sanas.
The inspiration behind Sanas’ product began with one of the creators’ friends named Raul. Raul was forced to take a leave of absence from Stanford University and return to Nicaragua. Back home, he started looking for jobs. Though he had high English proficiency, he only had a high school diploma to his name. The highest paying job he could land then was in a call center where he faced discrimination and verbal abuse daily due to his accent. Unsurprisingly, he quit after a few months.
The creators hoped to make a product to “ensure people around the world could take back the power of their own voice.”
Sanas’ Co-Founder and COO Sharath Keshava Narayana denied accusations that their tech was racist.
Narayana shared that the startup’s intention was to expand its translation model to include other accents.
"It's not just an American having trouble understanding someone from India, but vice versa as well," Narayana said. In fact, he and his co-creators envision that scaling up the product will lead to more localized solutions.
The voice of the people
Attempts to debunk allegations of discrimination have not stopped tech industry experts from voicing their concerns. CEO of the women-led computer programming group FrauenLoop Nakeema Stefflbauer flagged Sanas for not understanding that “accents signal power and belonging.”
She claimed that Sanas’ messaging promotes a false sense of community. "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,” she said.
She went on to label the AI product as a one-way solution that reinforces racial hierarchies, whether or not it was grounded on good intentions.
People, like their accents, are cultural products of generations’ worth of history. The way we speak echoes our socio-cultural identity. When we hide our accents, we hide who we are and where we came from.
Tech has and always will reflect the values of its creators and its users. When we support AI tech that virtually erases the nuances that come with unique voices, we enable a society that prefers comfortable conformity over cultural sensitivity.
Why relegate the task of changing accents to tech, when we as a people should just work harder to understand them?
