Reproduce “BARS: Beat-Adaptive Rap Synthesis via 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
Taemin Cho, Hector Martel, Enyan Koh
BARS generates rap vocals from lyrics, a beat, and a short voice sample in one non-autoregressive pass. It uses beat-length audio latents and scaled rotary positions for lyric keys to encourage coverage across the beat. On mismatched test pairs at 17.9 characters per second, it achieves 12.3% WER versus 22.0% without A-RoPE; WER is ASR-derived, not a human intelligibility rating.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.