Reproduce “MFlowAudio: Efficient Text-to-Audio Synthesis via Mamba-based Stateful 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

Hao Dai, Panyu Chen, Jagmohan Chauhan

MFlowAudio generates long-form text-to-audio in chunks using temporal and channel state-space scans, carrying temporal state between chunks to support streaming and prompt changes. For 30-second AudioCaps outputs, it takes 3.10 ± 0.018 seconds versus TangoFlux’s 3.55 ± 0.005 seconds, a reported 14.5% speedup. In dynamic-prompt evaluation, SDCG improves four reported metrics over vanilla, but does not establish universally seamless transitions.

Summary by alphaXiv.

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