HMN 2025: How Regional dialect rip-off warning as analysis uncovers AI capabilities

Researchers at Abertay University have raised considerations that advances in AI voice expertise could possibly be exploited by scammers to focus on unsuspecting victims by utilizing regional accents.

A brand new study, led by Dr. Neil Kirk of the Department of Sociological and Psychological Sciences, highlights how life like AI-generated voices have develop into, with listeners assuming the voices should be actual—notably when mimicking regional dialects like Dundonian Scots.

The analysis raises worries that scammers may goal particular areas of the nation and undertake regional dialects to seem extra human and deceive folks.

Dr. Kirk stated, “As AI voices develop into extra superior, it is getting tougher and tougher for most individuals to inform them other than actual human speech. This is a critical concern as a result of AI-augmented voices could possibly be utilized in scams like phishing or fraud, particularly once they convincingly imitate regional accents and dialects.”

Published within the journal Computers in Human Behavior: Artificial Humans, the research concerned the creation of six frequent phrases in each Scottish Standard English and Dundonian dialect—for instance, “The canine had 5 puppies” and “The doag had fev puppies.”

The recordings have been processed utilizing AI software program and when performed to 100 Scottish adults and 100 folks from southern England, most listeners struggled to differentiate each variations from actual voices. On common, Scottish members appropriately recognized real voices 57% of the time, whereas English members scored 52%.

Although Scots have been higher at recognizing the pretend voices, many AI variations nonetheless succeeded in deceiving them.

Dr. Kirk added, “Scottish individuals are not usually accustomed to expertise understanding their method of talking, not to mention sounding like us. Therefore, it is not shocking that Scottish members on this study have been extra prone to understand voices talking in Dundonian Scots as human, even when these voices have been really AI-generated.

“This highlights how scammers may exploit regional dialects to make pretend voices sound extra convincing and reliable. Additionally, my analysis exhibits that the general public usually underestimates simply how refined this expertise has develop into, whereas concurrently overestimating their skill to detect a pretend voice.”

In the paper, Dr. Kirk proposes that a method to assist scale back these dangers is to extend public consciousness that AI-generated voices can sound remarkably life like, particularly when regional accents are concerned. This could possibly be achieved by way of public consciousness campaigns or safety prompts from companies like on-line banking.

He added, “While such public security campaigns would possibly result in extra false alarms—resembling mistaking a real name from an individual’s financial institution for an AI voice—the general profit is a decrease danger of falling sufferer to scams or manipulation, which may have critical real-world penalties.”

Dr. Kirk’s main analysis focuses on understanding the cognitive mechanisms underlying bidialectalism—the usage of carefully associated varieties like Scottish English and Dundonian Scots. Recognized by the European Charter for Regional and Minority Languages, Scots is an indigenous Germanic language spoken by roughly 1.5 million folks in Scotland.

While Scots and English share a typical origin, their statuses have advanced over time. Many people regard Scots as a definite language, however a Scottish Government report discovered that 64% of respondents see it as “a method of talking.”