The world's songwriters shouldn't worry too much just yet, but artificial intelligence is taking aim at their talents. This week, research laboratory OpenAI unveiled Jukebox, a neural network that generates songs complete with lyrics and singing.
OpenAI trained Jukebox by feeding it 1.2 million songs along with lyrics and metadata. "The metadata includes artist, album genre and year of the songs, along with common moods or playlist keywords associated with each song," said OpenAI in a blog post on Thursday. The post goes into deep detail on the technical approach to the neural net.
Jukebox wanders into an audio version of the uncanny valley by creating songs in the style of specific artists. OpenAI shared a selection of song samples worth a listen. Jukebox's Alan Jackson song has a distinctive uptempo country flavor. Its Katy Perry pop song is a torchy ballad with reverb-soaked drums.
You can almost hear the sneer in the bouncy Elvis Presley-style song. OpenAI researchers stepped in to co-write some lyrics. The Presley tune starts with "From dust we came with humble start/from dirt to lipid to cell to heart." That's pretty trippy for an Elvis song.
The good news for human songwriters is that the AI still has some big limitations.
"While the generated songs show local musical coherence, follow traditional chord patterns and can even feature impressive solos, we do not hear familiar larger musical structures such as choruses that repeat," said OpenAI. Jukebox hasn't mastered the art of the hook.
OpenAI is riding on a. The research group in San Francisco previously created an , developed a and trained a . It's no surprise OpenAI is now coming for our Top 40.