- The voices generated by AI now imitate humans so convincing that detection is almost impossible
- Creating a convincing vocal clone now takes minutes and minimal expertise
- Some synthetic voices have in fact been evaluated more confidence than real human recordings
For years, many people supposed that the discourse generated by AI could always be identified by its slightly “false” qualities.
New research from Queen Mary University of London question this hypothesis, showing the current technology of the vocal AI reached a level where the “vocal clones” and the deep buttocks are almost indistinguishable from real recordings.
In the study, participants compared human voices with two synthetic audio forms: cloned voices designed to imitate real speakers and voices generated from an LLM system without specific homologists.
Beyond realism and domination
Auditors have often had trouble distinguishing between the two, suggesting that technology has entered a phase where human realism is no longer an aspiration, but a reality.
The research team studied not only if the participants could distinguish between synthetic and real voices, but also how they perceived them.
Surprisingly, the two types of voice generated by AI were assessed as more dominant than the most dominant than the most dominant, and in some cases, they were deemed more worthy of confidence.
Dr. Nadine Lavan, lecturer in psychology at Queen Mary University in London, underlined how much her team highlighted and at a lower cost these vocal clones.
“The voices generated by AI are all around us now, it was just a matter of time until AI technology begins to produce a naturalist and human consonance discourse, the process required minimal expertise, only a few minutes of vocal recordings and almost no money,” she said.
She said that ease of use shows how far technology has progressed in a short time.
Such accessibility creates opportunities in fields such as education, communication and accessibility, where tailor -made synthetic voices could improve commitment and realization.
Just as AI writers raise questions about originality, copyright and abusive use, the generation of votes of AI arouses debates on the property of identity and consent.
If realistic audio can be created from a short sample, the risk of unauthorized cloning become difficult to ignore.
While AI tools continue to develop capacities and accessibility, the challenge will be to ensure that the advantages are carried out without opening new paths for deception.
Understanding how people react to these voices is only the first step to approach the ethical, legal and social implications of a technology that is no longer futuristic, but firmly present.
