Can you tell the difference between a real voice and a fake one?
A nationwide investigation carried out by Ringover has revealed how easily people could be scammed or fooled by AI-generated voices.
This subject gained attention recently after Tom Hanks recently had to defend himself when a dental company used an AI-generated voice and image of him without his permission. Our survey results revealed that fewer than 2% of respondents were able to successfully identify the “real” voices in our experiment, with 98% being fooled by an AI-generated voice.
The experiment tasked respondents with identifying well-known real and fake celebrity voices. Prior to beginning the experiment, respondents were asked whether they felt like they would be able to identify an artificially generated voice or a real one. 78.3% respondents believed they would.
By the end of the experiment nearly half (47.6%) said they felt more concerned about artificial intelligence apps mimicking human voices, potentially leading to more scams and security concerns.
- Across five celebrity voices, fewer than 2% of people were able to identify all the “real” voices, with the vast majority being tricked at least once in the test.
- Nearly half (47.6%) of respondents felt more concerned about the potential threat AI generated voices could pose on their safety and security after the experiment.
- The majority of people (68.3%) felt their voices were secure before the experiment, and that banks should continue to use them as a security feature
- Nearly four out of five respondents (78.3%) believed they would be able to tell the difference between AI and real voices, prior to the experiment.
- Two out of every five people surveyed (39.6%) said they believed their friends and/or family would be able to tell if someone attempted to mimic their voice, prior to the experiment.
Would you fall for an AI voice scam?
Prior to our experiment, more than three in four respondents (78.3%) believed that they would be able to tell the difference between an AI generated voice and a real person's voice, with just under two out of every five respondents (39.6%) believing that their friends and family would be able to tell if they contacted by a mimicked voice of their own.
When asked about the past 12 months, 80% of respondents said they had noticed an increase in the number of customer service phone lines or services offering 'voice recognition' as a security measure. This is a security feature already being used by banking giants such as Wells Fargo and Citigroup.
While 68.3% of people consider their voice to be secure, to the point banking companies should be using it to access their financial accounts, our experiment shows how easily humans can be fooled by them.
To show how AI voice mimicking apps and tools have developed, Ringover created and sourced voice clips from five well known and notable celebrities for every age group and gender. The voices were then used to generate AI versions of the celebrities, where ‘fake' voice clips of these celebrities were created–using just 30 seconds of them talking.
Real or AI voices?
Of the over 1,000 respondents that took part in the Ringover experiment, just 18 (1.8%) were able to identify the real voices and correct answers over all five celebrities.
|Celebrity||Percentage that spotted the real life voice|
Results showed that women were 19.5% more likely to be able to identify all the correct voices, with 1.6% of all male respondents and 2% of all female respondents being able to identify all five real voices.
Millennials among worst for identifying AI voices
As only 2% of the respondents were able to correctly identify all AI generated voices correctly, we've extrapolated our results to find out which age groups were "the least wrong."
To do that, we've pivoted our data to include answers from people who correctly identified at least three real voices and then pivoted the data by age groups.
The results show that 0% of those aged above 65 weren't able to correctly identify at least three correct voices. However, it is Millennials that were second to bottom across the age groups - with those between 18 and 34 (the vast majority of millennial years) averaging 12.86%.
The best age group of correctly identifying the most correct answers were in fact those in Generation X, with the majority in this age range averaring 22.94% correct answers.
|Age Groups||Percentage With Least Three Correct Answers|
Can you tell the difference?
Want to see if you can tell the difference between an AI-generated voice and a real one?
Below are two Youtube links with audio content. Please listen to each and at the bottom of the page we'll reveal if both are real, one is real and the other AI-generated, or if they are both AI-generated.
If you want to test against all the 5 voices used in our experiment, please do so here. Like above, all answers are revealed at the bottom of this page.
Are you more concerned?
Given that fewer than 2% of respondents were able to correctly identify the real voice in the Ringover experiment, it's not that surprising that 88.2% of respondents said they found the test harder than they imagined before it began.
Results showed that as a result of doing the experiment, nearly half of the respondents (47.6%) said they were more concerned about the use of AI, while 84.2% of respondents said they were now concerned with how easily people may be fooled by voice generated AI in the future.
Between 10/11/2023 and 10/13/2023, a total of 1,012 people were surveyed using the Amazon survey programme. A series of questions was posed to respondents, which included their current experiences with AI and how they felt they and their friends might be able to identify AI voice mimicking.
Respondents were also given five different celebrity voices, with two voice options (one real and one AI generated) for each celebrity. Respondents were asked to identify the “real life” voice between option A and B, with available answers being Option A, Option B, Both, Neither.
At the end of the experiment respondents were then asked about how they now felt about AI voice generation and tools, and whether they felt more or less concerned about their use and development.
|Prefer not to say||0.20%|
Voice B is the real Miley Cyrus voice and voice A was generated by AI.
Voice B is the real Barack Obama voice, and voice A was generated by AI.
Voice A is the real Oprah Winfrey voice, and voice B was generated by AI.
Voice A is the real Prince Harry voice, and voice B was generated by AI.
Voice A is the real Kevin Hart voice, and voice B was generated by AI.