Reproduce “CPR: Combining global composing, local performing and full-sequence refining in piano rendering with continuous autoregressive modelling” 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
Chong Jing, Junan Zhang, Zhizheng Wu
CPR renders target piano MIDI with a reference recording’s sound. Composer plans, Performer generates local patches with flow matching, and Refiner upsamples audio from 24 to 48 kHz. On 100 examples, CPR reaches 78.4% onset F1 and 0.129 FAD_CLAP versus P-MUSE’s 75.7% and 0.161, but lower timbre similarity, 89.3% versus 90.6%. The comparison uses different rates and does not establish faster generation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.