Jukebox
Generated music and singing in diverse genres.
About Jukebox
Jukebox is a cutting-edge open-source neural network tool designed to create music and rudimentary singing in various genres and artist styles. This innovative tool not only releases the code and model weights but also includes an exploration tool for the generated samples. Users can interact with Jukebox by providing input on genre, artist, and lyrics, leading to the production of unique music samples. Jukebox offers a diverse range of music and singing styles, adapting to lyrics it has not encountered during training. One fascinating feature of Jukebox is its ability to generate music that diverges significantly from its training set when given familiar lyrics. By conditioning on just 12 seconds of audio, users can prompt the tool to complete the rest in a specified style. To address the challenge of handling long raw audio sequences, Jukebox employs an autoencoder to compress the audio into a lower-dimensional space. This approach enables the tool to generate audio within this compressed space and then up-sample it back to its original raw audio form. Breaking new ground in the realm of generative models, Jukebox surpasses traditional tools that generate music symbolically through a piano roll format. It is ideal for individuals eager to experiment with AI-generated music and explore the possibilities of creative expression.