0
Skip to content

AI Voice Technology Reaches Human-Level Realism

ai-voice-computer-clone-webnews-100New research shows AI-generated voices fool listeners 58% of the time, creating authentication and fraud challenges for the ARM industry.

10/10/2025 9:30 A.M.

2.5 minute read

The line between human and artificial intelligence is becoming increasingly difficult to distinguish. A new study from Queen Mary University of London confirms what many in the financial services world have anticipated: AI-generated voices have achieved a level of realism that makes them virtually indistinguishable from actual human speech.

This development carries significant implications for how credit and collection professionals authenticate consumers, prevent fraud, and maintain trust.

The Research Findings

The study tested 80 voice samples on listeners tasked with identifying which were real. The results found that cloned voices were labeled as “human” 58% of the time, while generic AI voices fooled listeners 41% of the time.

Perhaps most striking, participants rated AI voices as more dominant and, in some cases, more trustworthy than actual human voices.

Researchers concluded that “under certain conditions, it is not possible for human listeners to accurately discriminate between AI-generated voices and genuine recordings.”

What makes this especially alarming is how easy it is to access the technology. The research team created convincing clones using commercially available software, requiring only a few minutes of voice recordings and minimal cost or expertise.

This ease of access is compounded, in part, by inadequate industry safeguards, a March 2025 Consumer Reports report argued. Its assessment of six major voice cloning companies found that four products “did not employ any technical mechanisms to ensure researchers had the speaker’s consent to generate a clone,” requiring only that users check a box confirming legal rights. Consumer Reports concluded that “AI voice cloning tools have the potential to supercharge impersonation scams.”

The real-world impact is already evident. A McAfee study found that out of 7,000 people surveyed, one in four reported experiencing an AI voice cloning scam or knew someone who had.

Establish Protocols for Potential AI Voice Clones

“This is happening now, and if you haven’t had it happen yet, you should immediately think about how you want it to be handled at your agency,” Kelly Knepper-Stephens, chief legal officer at TrueML Technologies, said in a presentation at ACA International’s 2025 Convention & Expo.

If you suspect consumers are using AI assistants to communicate with your agency, obtaining proper authorization is critical. Develop specific protocols for AI-assisted communications and make sure you have documented authorization and/or consent agreements before engaging in substantive discussions about the debt.

“AI-generated voices are all around us now,” said Dr. Nadine Lavan, senior lecturer in psychology at Queen Mary University and the study’s corresponding author. “Our study shows that this time has come, and we urgently need to understand how people perceive these realistic voices.”

For collection professionals, that understanding should translate into updated policies and procedures, compliance frameworks, and training programs.

To dive deeper into these topics and stay current on the latest trends, consider attending ACA’s 2025 Fall Forum Nov. 5-7 in Chicago, where AI in collections is a featured topic.

Sessions during Fall Forum focus on adopting AI voice agents, using AI to optimize operations, and more. Register here.

If you have executive leadership updates or other member news to share with ACA, contact our communications department at comm@acainternational.org. View our publications page for more information and our news submission guidelines here.

One moment please...

Share Profile

This site utilizes cookies as included in our Privacy Policy and User Agreement.