Jukebox

🎶 Unleash your inner composer with Jukebox! 🎵 This neural net generates music in various genres and artist styles, complete with rudimentary singing. With model weights, code, and tools to explore samples, the possibilities are endless! 🎼🤖 #AI #MusicGeneration #Jukebox

Jukebox is a neural net that generates music in various genres and styles as raw audio, including singing.
It releases model weights, code, and a tool to explore generated music samples.
Jukebox can output new music samples based on genre, artist, and lyrics provided as input.
It uses VQ-VAE to compress raw audio to a discrete space and generate music codes.
The model learns different music levels and generates codes using transformers.
Various priors model the distribution of music codes and generate music in a compressed discrete space.
Jukebox is trained on a dataset of 1.2 million songs paired with lyrics and metadata.
It can be conditioned on artist, genre, and lyrics to influence the style of generated music.
Future directions involve conditioning on MIDI files and stem files for improved musicality.