Reproduce “FlashAudio: Rectified Flows for Fast and High-Fidelity Text-to-Audio Generation” 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
Huadai Liu, Jialei Wang, Rongjie Huang, Yang Liu, Heng Lu, Zhou Zhao, Wei Xue
FlashAudio generates text-conditioned audio in one step by learning a straighter route from noise to audio, then applying reflow and one-step distillation. On the AudioCaps test set, it reports FAD 1.49 and CLAP 0.648, versus 1.80 and 0.645 for Make-An-Audio 2 at 100 steps, with RTF 0.0025. Output is fixed-length; Tango 2 scores higher on CLAP.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.