Reproduce “SubRoformer: Subband Time-Domain Representation for Music Source Separation” 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
Chunyat Wu, Chuang Ding, Paris Smaragdis, Qiuqiang Kong
SubRoFormer separates vocals, bass, drums, and other instruments from stereo mixes by filtering waveforms into overlapping frequency bands, then combining fine-rate features with a temporally patched RoFormer. On MUSDB18-HQ, four separately trained target-specific models average 10.63 dB SDR, versus 10.02 dB for BS-RoFormer. This reported-system comparison is not architecture-matched; BS-RoFormer scores 1.38 dB higher on bass.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.