MusicGen is a model designed for converting text into music, and it comes in various sizes and versions. The following are the models available for MusicGen:
-
MusicGen - Large - 3.3B: This variant of MusicGen is capable of creating high-quality music samples based on text descriptions or audio prompts. It’s a single-stage auto-regressive Transformer model that has been trained using a 32kHz EnCodec tokenizer with four codebooks sampled at 50. There’s no need for a self-supervised semantic representation as it generates all four codebooks in one go.
-
MusicGen - Melody: This version of MusicGen is specifically designed to concentrate on the generation of music guided by melody.
These models have been created by Meta AI’s FAIR team and are aimed at research into AI-based music creation. They cater to researchers in fields such as audio, machine learning, artificial intelligence, as well as hobbyists who wish to gain a deeper understanding of these models.
