August 3, 2021

Google AI creates new Nirvana and Jimi Hendrix songs

Thanks to Magenta, a Google algorithm, four members of the 27 club have just released a new album despite being dead for several years. Artificial intelligence has dived into the discography of Kurt Cobain, Jimi Hendrix, Amy Winehouse y Jim Morrison to create four new songs that keep all the essence of the music of their creators.

Scientists create a “new way of life”

Jesus Diaz

‘Lost Tapes of the 27 Group’ is an album that is part of the Canadian NGO Over the Bridge and that seeks to raise awareness about the high incidence of mental health problems among musicians. To compose the new songs they have used Magenta, an open source artificial intelligence (AI) tool created by Google.

Drowned in the Sun, Nirvana

“The musicians and the people who are part of their team, have had mental health problems in a much higher proportion than the adult population in general,” they comment from the NGO. “This problem has not only been ignored. It has been idealized with things like the Club de los 27, a group of musicians whose lives were lost when they were only 27 years old. ”

Man, I Know, Amy Winehouse

They have taken a year and a half to create the album. They first made the AI ​​listen to musicians’ rhythms, melodies and lyrics in isolation. The algorithm then created its own versions based on what it had learned, and finally, a sound engineer put it all together to create the final version of the album.

You’re Gonna Kill Me, Jimmi Hendrix

The voices weren’t computer-generated – although there are already algorithms that could recreate them with astonishing fidelity – instead they used singers like Eric Hogan, ‘frontman’ of a Nirvana tribute band. Hogan told Rolling Stone magazine that for him the lyrics are very successful. They carry the typical spirit of Cobain although they are more complete thoughts and not “a kind of hodgepodge”, as was usual in the Seattle composer.

The Roads Are Alive, The Doors

Sean O’Connor, a member of the board of directors of Over the Bridge, has commented in an interview in the same outlet as Drowned in the Sun, the subject of Nirvana, has been one of the most difficult. Unlike Hendrix songs, which have totally defined sounds, Cobain played the guitar in a punk style that confused the algorithm.

The Goldberg Variations, Glenn Glould

This project is reminiscent of another that used an AI to play Bach’s Goldberg Variations in imitation of the very personal style of American pianist Glenn Gould. In this case, the algorithm did not learn only from listening to Gould’s recordings, but relied on the collaboration of other pianists who helped fine-tune the technology so that the result was as close to the real thing as possible.

Imitation or plagiarism?

One of the fundamental concerns of Over the Bridge has been not to fall into possible infringements for plagiarism. And is that there’s a fine line between what these AIs produce and the music of actual creators. A little over a year ago, rapper Jay-Z denounced the creator of a YouTube video that used a technology like ‘Deep Fake’ to imitate his voice and recite the monologue “To Be, Or Not To Be” from Hamlet.

A few days later, Google, owner of YouTube and which had removed the video from its platform at the singer’s request, uploaded it again claiming that its complaint for copyright infringement was incomplete. The creator of these videos calls himself Vocal Synthesis and in an interview he admits surprised by the controversy.

Jay-Z recitando “To Be, Or Not To Be”

“I have been publishing this type of videos for months and they have not withdrawn any other for this reason” says Vocal Synthesis. “There are also other channels that make voice synthesis videos similar to mine, and I do not know that any of them have videos have been removed for this reason. ”

And it is that, as the same Vocal Synthesis comments, on Youtube there are a lot of imitators singing in the way of their idols and they have never had a problem. “I don’t really understand why mimicking a celebrity’s voice using an AI model should be treated differently than someone naturally doing an (extremely accurate) imitation of that celebrity’s voice.”

Imitation of Frank Sinatra by

This legal problem, which as we see is already beginning to arise, will become even more complicated as this type of technology advances. There will come a time when the voices and songs composed by the AI ​​will be indistinguishable from those of the original musicians. And the creations of these machines, although based on those of other authors, could be considered as new and, to some extent, original.

Creation is a complex phenomenon because no one believes in a vacuum, as this documentary called ‘Everything is a Remix’ explains very well. Los Brincos imitated the Beatles. The Beatles imitated Little Richard, although they later evolved and incorporated new influences. Little Richard imitated the ‘gospel’ sounds he heard at evangelistic masses. And so we could continue until we found the first one who thought that the repetitive noise it made when two stones, two sticks or whatever it was colliding, was music.