AI RESEARCH
SoundWeaver: Semantic Warm-Starting for Text-to-Audio Diffusion Serving
arXiv CS.CV
•
ArXi:2603.07865v1 Announce Type: cross Text-to-audio diffusion models produce high-fidelity audio but require tens of function evaluations (NFEs), incurring multi-second latency and limited throughput. We present SoundWeaver, the first