Pause carousel
Play carousel
Study shows that AI can convincingly mimic regional dialects such as Dundonian Scots
Researchers at Abertay University have raised concerns that advances in AI voice technology could be exploited by scammers to target unsuspecting victims by using regional accents.
A new study, led by Dr Neil Kirk of the Department of Sociological and Psychological Sciences, highlights how realistic AI-generated voices have become, with listeners assuming the voices must be real - particularly when mimicking regional dialects like Dundonian Scots.
The research raises worries that scammers could target specific areas of the country and adopt regional dialects to appear more human and deceive people.
Dr Kirk said:
As AI voices become more advanced, it’s getting harder and harder for most people to tell them apart from real human speech. This is a serious concern because AI-augmented voices could be used in scams like phishing or fraud, especially when they convincingly imitate regional accents and dialects.
Published in the journal Computers in Human Behaviour: Artificial Humans, the research involved the creation of six common phrases in both Scottish Standard English and Dundonian dialect—for example, "The dog had five puppies" and "The doag had fev puppies."
The recordings were processed using AI software and when played to 100 Scottish adults and 100 people from southern England, most listeners struggled to distinguish both versions from real voices. On average, Scottish participants correctly identified genuine voices 57% of the time, while English participants scored 52%.
Although Scots were better at recognising the fake voices, many AI versions still succeeded in deceiving them.
Dr Kirk added:
Scottish people are not often accustomed to technology understanding their way of speaking, let alone sounding like us. Therefore, it’s not surprising that Scottish participants in this study were more likely to perceive voices speaking in Dundonian Scots as human, even when those voices were actually AI-generated. This highlights how scammers could exploit regional dialects to make fake voices sound more convincing and trustworthy. Additionally, my research shows that the public often underestimates just how sophisticated this technology has become, while simultaneously overestimating their ability to detect a fake voice.
In the paper, Dr Kirk proposes that one way to help reduce these risks is to increase public awareness that AI-generated voices can sound remarkably realistic, especially when regional accents are involved. This could be achieved through public awareness campaigns or security prompts from services like online banking.
He added:
While such public safety campaigns might lead to more false alarms—such as mistaking a genuine call from a person’s bank for an AI voice—the overall benefit is a lower risk of falling victim to scams or manipulation, which could have serious real-world consequences.
Dr Kirk’s primary research focuses on understanding the cognitive mechanisms underlying bidialectalism—the use of closely related varieties like Scottish English and Dundonian Scots. Recognised by the European Charter for Regional and Minority Languages, Scots is an indigenous Germanic language spoken by approximately 1.5 million people in Scotland.
While Scots and English share a common origin, their statuses have evolved over time. Many individuals regard Scots as a distinct language, but a Scottish Government report found that 64% of respondents see it as “a way of speaking”.
In 2020, Dr Kirk established MIND Your Languages, an online research platform dedicated to studying Minority, Indigenous, Non-standard and Dialect (MIND) varieties.
