musiclm

Definition

A Google model that generates music from a text description, treating the task as conditional generation over a hierarchy of semantic and acoustic tokens rather than raw waveform prediction. Conditioning goes through a joint text–music embedding, which is what lets free-form prompts like a description of mood and instrumentation steer the result. It is also remembered for the MusicCaps evaluation set released alongside it, and is a standard reference point in text-to-music work.