- AI fake voice detection applications fail by 27%

AI fake voice detection applications fail by 27%


A study by the British University College London showed that more than a quarter of the imitated voices using what is known as "deep faking" succeeded in deceiving even the most acute and focused listeners.

According to researchers at the university, more than 500 people who were trained to distinguish real sounds from imitations were able to detect only 73% of the imitations during the study.

Deepfake is an application of artificial intelligence that is used to generate sounds, images and videos that appear to have been created by real individuals, similar to the famous fake video of former US 

President Barack Obama.

The study was conducted using English and Chinese (Mandarin), where the percentages were similar in the results of the two languages, although the research team found that the English speakers were able to distinguish the human voice from the imitated voice generated using artificial intelligence through the rhythm of breathing, while the Chinese speakers indicated that Rhythm, speed and fluency of speech as criteria for discrimination.

It is reported that fake votes are strictly used to defraud people and get money from them by pretending to be a friend or business partner and asking the victim to transfer money to him.

Recent developments in artificial intelligence technology have raised concern about the possibility of spreading fraud by imitating voices, especially with the ability of artificial intelligence applications to imitate voices with high accuracy.

The team of researchers at the British University warned that "as a result of technological developments, it has become possible to generate an almost real copy of the voice of any person using a recording of a few words of his voice."

The researchers said that the results of the study may not reflect the actual reality that the participants in the study - even those who were not trained - knew that they were part of an experiment, and therefore were more focused on discovering the imitated voices.

Most of the efforts to combat counterfeiting of votes are currently focused on using artificial intelligence systems to detect them, which performed at the same level as the performance of people participating in the university experiment, but they excelled compared to people who did not know that they were participating in an experiment to discover fake voices.

The researchers said that in light of the expectation of significant development and progress in the techniques of imitating sounds, what is required is the development of more advanced tools for their discovery.
Source : German news agency

Post a Comment


Post a Comment (0)