Scholarlink

<p>In public discourse in the modern U.S., complaints about “foreign accents” of call-center workers and the outsourcing of service labor are a common trope (Cowie 2007). Recently, several tech companies have begun to market AI-powered “accent translation” systems that claim to eliminate the “foreign accents” of call-center workers, thereby increasing revenue, improving customer satisfaction, and providing more positive worker experiences (Krisp, 2025, Sanas, 2025). However, these companies provide no details about the technical functions of their programs, and no empirical evidence to show that AM speech is more intelligible or more appealing to listeners. The current study tests U.S. listener perception of unmodified and AM speech of four call-center employees from India and the Philippines, taken from publicly available demos produced by an AM company, in order to address the following research questions: 1. Does the use of Accent Modification software improve intelligibility of these demo call-center workers? 2. Do listeners demonstrate aesthetic preferences for accent-modified voices over original voices? 3. What type of feedback do listeners provide about their experiences with both voice types? Our results indicate that listeners prefer original, unmodified voices over AM versions, rating the originals as less accented, easier to understand and preferable in a customer service context. Evaluation of performance on transcription tasks reveals that listeners perform at a high level for both Indian and Filipino call-center workers on the original speech, with a small improvement in accuracy in AM for Indian speakers only. We argue that despite marketing claims that AM technology improves both understanding and customer satisfaction, these results indicate that AM technologies neither improve across the board listener comprehension nor customer experience. We also argue that AM technologies have the ability to do harm to customers and employees by reinforcing negative ideologies about non-U.S. English, dehumanizing workers and clients, and obscuring the use of international service labor.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Accent modification AI fails to improve listener experience

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月2日

查看全文
  • Nicole R. Holliday,
  • Melissa M. Baese-Berk

摘要

In public discourse in the modern U.S., complaints about “foreign accents” of call-center workers and the outsourcing of service labor are a common trope (Cowie 2007). Recently, several tech companies have begun to market AI-powered “accent translation” systems that claim to eliminate the “foreign accents” of call-center workers, thereby increasing revenue, improving customer satisfaction, and providing more positive worker experiences (Krisp, 2025, Sanas, 2025). However, these companies provide no details about the technical functions of their programs, and no empirical evidence to show that AM speech is more intelligible or more appealing to listeners. The current study tests U.S. listener perception of unmodified and AM speech of four call-center employees from India and the Philippines, taken from publicly available demos produced by an AM company, in order to address the following research questions: 1. Does the use of Accent Modification software improve intelligibility of these demo call-center workers? 2. Do listeners demonstrate aesthetic preferences for accent-modified voices over original voices? 3. What type of feedback do listeners provide about their experiences with both voice types? Our results indicate that listeners prefer original, unmodified voices over AM versions, rating the originals as less accented, easier to understand and preferable in a customer service context. Evaluation of performance on transcription tasks reveals that listeners perform at a high level for both Indian and Filipino call-center workers on the original speech, with a small improvement in accuracy in AM for Indian speakers only. We argue that despite marketing claims that AM technology improves both understanding and customer satisfaction, these results indicate that AM technologies neither improve across the board listener comprehension nor customer experience. We also argue that AM technologies have the ability to do harm to customers and employees by reinforcing negative ideologies about non-U.S. English, dehumanizing workers and clients, and obscuring the use of international service labor.