Reproduce “Smorph: Playable Sound Morphing with Diffusion Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Annie Chu, Hugo Flores García, Johannes Imort, Oriol Nieto, Bryan Pardo, Jordan Rudess, Prem Seetharaman, Justin Salamon
Smorph guides diffusion morphs to change sonic identity while retaining a temporal reference. It uses RMS amplitude envelopes or pitch contours as structural cues and shifts guidance between endpoint prompts relative to a structure-conditioned prediction. In Clotho–Clotho morphing with RMS control, explicit anchoring raises adherence from 0.428 to 0.733, but reduces net target transformation versus CoA in this setting.
Key insights
- Explicit structure anchoring can improve adherence to temporal features, but may constrain how far the output moves toward the target identity.
- Audio-based initialization improves source preservation; in audio-to-audio tests, combining it with FLAM target conditioning also improved reported audio-quality scores over FLAM conditioning alone.
- Musicians in a small exploratory study found the morph space useful for exploring intermediate timbres, while describing pitch and RMS controls as supporting different kinds of musical use.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.