In her latest video posted on YouTube, the music Holly Herndon does not sing. However, netizens can still listen to a version of the title. Jolene by Dolly Parton. In reality, it is Holly +, the artificial intelligence designed after her voice, who performs the song.
Although the generation of illustrations from artificial intelligence is democratized, as Dall-E is now available in the Office suite, music has not yet been affected by the phenomenon. And Holly Herndon is one of the hosts on this topic, she explains to Engadget.
It is thanks to Spawn, a neural network capable of singing, that Jolene’s cover was conceived from scratch. The result is perfectly recognizable: the rhythm is slower, the tone calmer, but the arrangements remain similar. The technology even manages to reproduce the inspirations reciting the lyrics.
Until now, the use of artificial intelligence in the field of music was limited to creating soundscapes or melodies. But this time, it is possible to make the algorithms sing.
What copyright?
If the video is impressive, you should not expect the same result using Holly+ at home. One site offers to transform audio files using Holly Herndon’s voice model. After a few tries, we were unable to decipher the lyrics sung by the artificial intelligence. And this despite sending songs in English or French, with or without instrumental tape.
The Holly+ site is meant to be a distraction. But for artists who really want to use artificial intelligence, they will have to turn to Spawning. As its name suggests, this tool uses the neural network developed jointly by Holly Herndon, with Mat Dryhurst.
And if data management seems to be an important parameter for Spawning, it is not so important for copyright. On the site, they are presented as an outdated system, which is not suitable for the age of artificial intelligence.
Source: BFM TV
