Similarity-Guided Diffusion for Long-Gap Music Inpainting

Sean Turland, Eloi Moliner, Vesa Välimäki · 2025

Read Similarity-Guided Diffusion for Long-Gap Music Inpainting by Sean Turland, Eloi Moliner, Vesa Välimäki (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-16342.txt.

Description

Filling a long gap in a music recording is hard for generative models, which tend to wander off style over several seconds. Sean Turland, Eloi Moliner, and Vesa Välimäki guide a diffusion model with look-alike segments pulled from a corpus of music. The retrieved parts steer the filling process toward music that fits the surrounding context.

Listeners judged the guided fills for two-second gaps in piano music as more plausible than unguided diffusion fills. The hybrid also often beat plain look-alike search when fair matches were on hand. Blending search with generation offers a route to longer, more faithful repairs.

Catalog record

CREATORSean Turland, Eloi Moliner, Vesa Välimäki
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE27 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-16342

Browse the full library · About the Tower of Babel