AI-generated music is a peculiar thing to think about for two reasons: First, it's a fascinating new field. Second, we don't really know where it's going and it's curious to see if it will have commercial outlets. For me yes, and today I have further proof of this.
OpenAI has revealed new horizons by creating a "jukebox of the future" capable of producing songs of all genres, (including singing) that mimic the styles of artists. AI has already produced “new releases” by Pink Floyd, Ella Fitzgerald, 2Pac, Frank Sinatra, Katy Perry, Elvis Presley and others.
The AI now sings seriously
The results are quite convincing: of course, in some cases the songs seem to be versions sung by our favorite drunk karaoke artists.
Jukebox, the organization's new music generation system, was announced and detailed in an official blog post. According to the site, Jukebox is “a neural network that generates music, including rudimentary singing, as raw audio across a variety of genres and artist styles”. There is also a tool on the site to explore the generated samples. There are totally new songs and others "revisited" by artificial intelligence, like this Rick Astley classic that has been distorted quite a bit:
The jukebox of the future compresses music into codes
The science behind the AI Jukebox is really interesting. To make this possible, the development team used raw audio to train the model, which returned more raw audio. Next, they aimed to generate new audio from compressed code that was then transformed back into raw audio. In this way, the neural network generated songs that did not exist before.
By giving Jukebox genre, artist, and lyrics as input, the AI produces a new music sample from scratch. From Jukebox Sample Explorer, you can already explore nearly 8000 tracks generated by numerous artists, complete with instrumental voices and voices with lyrics.
The next steps
The team is continuing to work on generating audio samples from different types of information, such as MIDI files. In this way, they think they will be able to give musicians more control over the songs they generate. And when everything is missing, they can still play these songs live, with a Robot conductor.
An artist could "patent" his style, and challenge time and inspiration by ensuring that artificial intelligence always produces new songs. A bit like digital versions of actors, an AI inspired by an artist could continue to launch new songs well beyond the physical death of its "inspirer".