AI and the End of Accents

It all began, as these things often do, with an Instagram ad. “No one tells you this if you’re an immigrant, but accent discrimination is a real thing,” said a woman in the video. Her own accent is faintly Eastern European—so subtle it took me a few playbacks to notice.

The ad was for BoldVoice, an AI-powered “accent training” app. A few clicks led me to its “Accent Oracle,” which promised to guess my native language. After I read a lengthy phrase, the algorithm declared: “Your accent is Korean, my friend.” Smug. But impressive. I am, in fact, Korean.

I’ve lived in the US for more than a decade, and my English isn’t just fluent. You could say it’s hyperfluent—my diction, for one, is probably two standard deviations above the national average. But that still doesn’t mean “native.” I learned English just late enough to miss the critical window for acquiring a native accent. It’s a distinction that, depending on the era, could lead to certain complications. In the Book of Judges, the Gileadites are said to have used the word “shibboleth” to identify and slaughter fleeing Ephraimites, who couldn’t pronounce the sh sound and said “sibboleth” instead. In 1937, the Dominican dictator Rafael Trujillo ordered the death of any Haitian who couldn’t pronounce the Spanish word perejil (parsley) in what became known as the Parsley Massacre.

So the stakes felt high as the Accent Oracle kept listening to me talk, at one point scoring me 89 percent (“Lightly Accented”), another time 92 percent (“Native or Near-native”). The spread was unsettling. On a bad day, I could have been slaughtered. To improve my odds of survival, I signed up for a free, one-week trial.

There is a medium-is-the-message quality to accents. How you say something often reveals more—about your origin, class, education, interests—than what you say. In most societies, phonetic mastery becomes a form of social capital.

As it has for everything else, AI has now come for the accent. Companies like Krisp and Sanas sell real-time accent “neutralization” for call center workers, smoothing a Filipino agent’s voice into something more palatable for a customer in Ohio. The immediate reaction from the anti-AI camp is that this is “digital whitewashing,” a capitulation to an imperial, monolithic English. This is often framed as a racial issue, perhaps because ads for these services feature people of color and the call centers are in places like India and the Philippines.

But that’d be too hasty. Modulating speech for social advantage is an old story. Remember that George Bernard Shaw’s Pygmalion—and its musical adaptation, My Fair Lady—hinges on Henry Higgins reshaping Eliza Doolittle’s Cockney accent. Even the eminent German philosopher Johann Gottlieb Fichte shed his Saxon accent when he moved to Jena, fearing people would not take him seriously if he sounded rural.

This is no relic of the past. A 2022 British study found that a “hierarchy of accent prestige” persists and has changed little since 1969, with a quarter of working adults reporting some form of accent discrimination on the job, and nearly half of respondents saying they were mocked or singled out in social contexts.

In a Hacker News thread announcing BoldVoice’s launch, one commenter wrote, “I’d rather strive toward a world where accents matter less than fixing accents.” Well, tell that to countless Koreans in this country navigating the treacherous phonetic gulf between beach and bitch or coke and cock. That online comment was characteristic of the usual sanctimonious pablum, the kind of casual moral high ground afforded only to a native English speaker or to someone willfully ignorant of the daily indignities non-native speakers face.