MusiChat Enables Iterative Conversational Music Composition

MusiChat employs a hierarchical controllable generation framework that separates lyric-aligned structure from surface realization. It uses a memory-augmented architecture to allow users to perform iterative, natural-language refinements on existing musical compositions.