Reproduce “MuseFlow-TSE: Music-Aware Target Speech Extraction via Cross-Source Band-Splitting and Data-to-Data Flow Matching” 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

Lu Ma, Yongliang Wang, Zhewen Su, Xin Tian

MuseFlow-TSE extracts an enrolled speaker from music-bearing mixtures by estimating target and interference separately, then refining the target while conditioning on interference estimates. In the Complex synthetic scenario, refinement raises SI-SDRi from 13.0 to 13.2 dB and PESQ from 2.71 to 2.88, while MusicLeak rises from 2.9% to 3.4%. Evaluation uses 16 kHz synthetic mixtures, not natural recordings.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.