Would you be ready to listen to a song made by an entity outside the living? You may soon have to ask yourself the question. The number of content-generating AIs continues to grow, in all fields. Today, with software like DALL-E or Stable Diffusion on the image side, the results obtained from a simple sentence are already very impressive. Since then, we have even seen artificial intelligence attacking other areas, such as video with Meta. And now even the music is concerned.
The AI that sings for you
This creation is signed Holly Herndon, an American composer who co-developed a neural network dedicated to song called Spawn with her partner Mat Dryhurst and who notably launched Holly+ last year (in partnership with Never Before Heard Sounds). This software now allows anyone to use a model of the singer’s voice to interpret any title. She also used it herself to release a single that only uses her voice from the software.
The result, which is a cover of the title “Jolene” by Dolly Parton, seems almost indistinguishable from a song performed by a real artist and yet was generated by the software. There are still some digital artifacts when you listen closely, but other than that, the final song is impressively realistic.
A first impulse towards a self-generated art?
Musical creations involving AI have so far been more concerned with generating melodies, but this one now returns directly to the front of the stage to interpret the songs itself. This novelty could one day open the way to new types of works with artists who would use automatically generated voices to create their songs.
The artist notably presented his title at the Sonar Festival in Barcelona and performed real-time demonstrations of Holly+ there. Anyone can now record their performance and have it sung by the software via its web application. However, the web version of the software will not provide you with the same fidelity as on the song, but it is already a good start.