Can the once familiar voice still be taken seriously? AI voice cloning technology has caused controversy

【Today's Viewpoint】

◎ Zhang Jiaxin, reporter of this newspaper

In April 1970, the Beatles disbanded, leaving millions of fans heartbroken. Now, more than 4 years later, artificial intelligence (AI) is resurrecting "lost memories", recreating and reimagining the Beatles' singing.

The fact that AI voices are simulated so realistically has also raised a series of ethical and legal questions, and even led to the emergence of telecom fraud. According to the Australian website Insider Guides, the latest report shows that Australians lost a record $2022.31 billion to scams in <>.

Sound clones are easy to exploit by scammers

In telecom fraud, scammers use AI software to scan voice recordings and copy them with up to 99% accuracy. They can extract clips from videos and other uploaded audio from social media, feed them into AI software, reproduce the voice and say different phrases or sentences.

Dipper Nguyen, an associate professor at the School of Electrical and Data Engineering at the University of Technology Sydney in Australia, said AI models can reconstruct relatively short speech fragments and string fragments into coherent sentences. Some models and algorithms can synthesize a fairly high-quality cloned sound in just a minute or less.

Nguyen said advanced AI models and algorithms synthesize sounds so well that "it's hard for ordinary people to distinguish between cloned sounds and real ones."

Time magazine reported that a few months ago, a family in Arizona thought they had received a kidnapping call, and the voice on the phone sounded the same as the voice of their loved ones, but it turned out that it was a scam created entirely by AI. The growing number of scams has led to fears that AI could become a technology that threatens people, and that it is readily available.

AI turns ordinary people into singers

AI sound cloning has also spread to music, with people using the technology to create songs that are exactly the same as the voices of stars. Recently, a creator named "Dae Lims" posted several songs written with AI on social media.

"It's a cry! It's beautiful! One listener commented under the song "New", a 2013 single by Paul McCartney, which was remastered with the help of AI and "sung" parts of it by McCartney's friend, the late John Lennon in 1980.

A similar example is that in April this year, foreign netizen "Ghostwriter4" trained the AI with the voices of well-known rapper Drake and R&B singer The Weeknd, imitating the musical style of the two to generate the "chorus" song "Heart on My Sleeve". As soon as the song was released, it went viral on social media.

"We're really entering a new era." One listener responded in the comments, "It's not even possible to tell what's legal and what's fake." ”

Patricia Alexandrini, a composer and assistant professor at the Stanford University Computer Research Center for Music and Acoustics, said the recent large number of AI tracks represents the maturity of a technology that has been developing exponentially but has been largely invisible to the public for the past decade. This means that people can now train AI in any aspect, but "we can't expect it to replace the rich history of human creation of art and culture."

A threat to the music industry

For the music industry, the impact of AI cloning sound to generate music is enormous. With the advancement of technology, in the near future, people can easily convert their singing voice into the voice of their favorite singer using a certain software.

AI has proven to have a huge impact on the copyright community.

In the case of "Heart on My Sleeve", Universal Music Group (UMG), the record label signed by the two singers, quickly made a claim and asked that the song be removed from the streaming service.

Mark Ostrow, a music rights lawyer in New York, said AI-generated music is a "gray area."

How is copyright defined? What level of copying is "fair use"? And how to stop the expansion of its irrational use?

David Isrett, president and CEO of the National Music Publishers Association (NMPA), said songwriters and music publishers should be allowed to better protect their work from unauthorized use, which will be challenged like never before in the era of AI.