AI and the Rise of Accent Neutralization

It all started, as is often the case, with an Instagram ad. “No one warns you if you’re an immigrant, but accent discrimination is very real,” a woman in the video stated. Her accent hints at Eastern European roots—so subtle that it took me multiple listens to catch on.

The ad was for BoldVoice, an AI-driven “accent training” app. A few clicks led me to its “Accent Oracle,” which claimed it could identify my native language. After reading a lengthy sentence, the algorithm confidently announced: “Your accent is Korean, my friend.” Quite smug. But impressive. I am, indeed, of Korean descent.

I’ve resided in the US for over a decade, and my English is not just fluent. You might call it hyperfluent—my diction alone likely surpasses the national average by two standard deviations. Yet that still doesn’t equate to “native.” I picked up English just late enough to miss the crucial window for a native accent. This distinction can lead to various complications over time. In the Book of Judges, the Gileadites reportedly employed the word “shibboleth” to identify and eliminate fleeing Ephraimites, who mispronounced it as “sibboleth.” In 1937, Dominican dictator Rafael Trujillo ordered the execution of any Haitian unable to pronounce the Spanish word perejil (parsley), leading to what is known as the Parsley Massacre.

With such high stakes, I felt anxious as the Accent Oracle continued to evaluate my speech, at one point rating me 89 percent (“Lightly Accented”), another time 92 percent (“Native or Near-native”). The variability was disconcerting. On a bad day, I could have faced dire consequences. To improve my chances, I opted for a free one-week trial.

There’s a profound significance to accents. How you express something frequently reveals more about your background, social class, education, and interests than what you actually say. In many cultures, mastering pronunciation becomes a form of social currency.

As it has with everything else, AI is now reshaping accents. Companies like Krisp and Sanas offer real-time accent “neutralization” for call center employees, transforming a Filipino agent’s voice into something more acceptable for a client in Ohio. The immediate response from critics of AI is that this represents “digital whitewashing,” succumbing to a monolithic, imperial English. This is often framed as a racial issue, likely because advertisements for these services feature people of color and the call centers are located in countries like India and the Philippines.

However, that view may be too simplistic. Adjusting speech for social gain is an age-old practice. Consider George Bernard Shaw’s Pygmalion—and its musical adaptation, My Fair Lady—which revolves around Henry Higgins transforming Eliza Doolittle’s Cockney accent. Even the esteemed German philosopher Johann Gottlieb Fichte altered his Saxon accent upon moving to Jena, fearing he would not be taken seriously if he sounded provincial.

This isn’t merely a remnant of history. A 2022 British study identified that a “hierarchy of accent prestige” persists, unchanged since 1969, with a quarter of working adults reporting experiences of accent discrimination at work, and nearly half saying they’ve been ridiculed or singled out in social situations.

In a Hacker News thread announcing BoldVoice’s launch, one commenter remarked, “I’d prefer to work towards a world where accents matter less than correcting them.” Well, tell that to countless Koreans navigating the tricky phonetic divide between beach and bitch, or coke and cock. That online comment typified the usual self-righteous rhetoric, reflecting the casual moral high ground enjoyed only by native English speakers or those willfully unaware of the daily challenges faced by non-native speakers.