SB-RF: Schrödinger Bridge Rectified Flow for speech enhancement

Arxiv CS.SD (Sound/Audio ML)

SB-RF integrates Rectified Flow and Schrödinger Bridge into a one-step generative framework for speech enhancement, enabling robust performance with single-step inference. The approach reduces computational overhead compared to traditional multi-step diffusion models.

Arxiv CS.SD (Sound/Audio ML)·Caixia LuXueyang LvPenglong HuJiaming Xu

6 days ago