AI Voice Cloning Technology Raises Persuasion and Fraud Concerns

Artificial intelligence (AI) can duplicate human voices from brief audio samples. This technology enables scammers to mimic voices with just seconds of sound. Impostor scams are a common form of fraud, according to the Federal Trade Commission.

Kimberly Hyun, an assistant professor of marketing at the University of Cincinnati's Carl H. Lindner College of Business, studies the role of voice in persuasion. Her research uses machine learning to analyze voices. She can analyze voices with less than ten seconds of audio. Hyun's study, "Vocal similarity, timbre and persuasion in consumer-spokesperson interactions," was published in the *Journal of Marketing Research*.

The study found that individuals are more susceptible to persuasion when speaking with a voice that sounds familiar. A spokesperson's persuasiveness increases with vocal quality similar to the consumer's own. Voices resembling those of trusted individuals or even one's own voice are more likely to elicit compliance.

Hyun's team focused on timbre, the unique quality of a voice. Timbre distinguishes voices even with identical pitch, tone, and volume. Every voice possesses a distinct timbre, much like a fingerprint. This allows for individual identification through voice.

Researchers analyzed over 7,000 pitches from the reality television series "Shark Tank." They measured the timbre of spokespeople and their similarity to the investors. The study also examined over 2,000 Kickstarter crowdfunding campaigns. Campaigns with audio that featured voices similar to the target population's average voice saw increased success. Four experiments further demonstrated that participants showed greater trust in familiar-sounding voices in advertisements. Machine learning techniques quantified vocal timbre similarity in these experiments.

This research highlights the power of persuasion beyond business applications, extending to politics and education. Voice technology and cloning are becoming more advanced and accessible. This makes it easier to create AI-generated voices that sound similar to a target individual. This similarity can lower a person's guard, making them more likely to comply with requests, including fraudulent ones.