Search papers, labs, and topics across Lattice.
This paper introduces BeatEdit, a novel framework for symbolic music generation that leverages explicit editing operations rather than generating sequences from scratch. By utilizing the BEAT encoding, which is structured for editing, BeatEdit enables more precise and perceptually appealing music creation through mechanisms like per-token tagging, iterative refinement, and segment completion. The results demonstrate that BeatEdit outperforms traditional autoregressive and diffusion methods in both quality and efficiency, completing edits in under 100 ms while revealing significant interactions between encoding design and editing effectiveness.
BeatEdit transforms music generation from a linear creation process into a dynamic editing experience, achieving superior quality and speed in under 100 ms.
Music creation is fundamentally a process of revision. Yet symbolic music generation remains dominated by paradigms that produce complete sequences from scratch, with limited support for selective modification. Edit-based methods have proven effective for text transformation tasks, but remain largely unexplored for symbolic music. We trace this absence to the representational level: conventional event-based music encodings lack the structural properties required by explicit music editing. In contrast, the BEAT encoding, a beat-grid-anchored representation originally designed for autoregressive generation, possesses structural properties amenable to editing. We propose BeatEdit, the first framework for symbolic music generation based on explicit edit operations, recasting generation as producing new content by editing a draft rather than synthesizing from scratch. BeatEdit comprises three complementary mechanisms along an axis of increasing edit density: per-token sequence tagging for error correction, iterative refinement for accompaniment editing, and tag-then-fill for segment completion. All these mechanisms share a single encoding and pre-trained backbone, achieving higher precision and perceptual quality than autoregressive and diffusion methods across all three tasks, while remaining efficient, with single-pass inference completing in under 100 ms. Cross-encoding evaluation further reveals that encoding design substantially influences editing effectiveness, with notable encoding-method interaction effects. Code is available at https://github.com/Haoyu-Gu/BeatEdit-code
