Speech Enhancement Based on Drifting Models
This work introduces a new paradigm for speech enhancement by replacing iterative sampling with a single-step inference, addressing the efficiency bottleneck for real-time applications.
DriftSE proposes a generative framework for speech enhancement that achieves one-step denoising by evolving a pushforward distribution to match clean speech, outperforming multi-step diffusion baselines on VoiceBank-DEMAND.
We propose Speech Enhancement based on Drifting Models (DriftSE), a novel generative framework that formulates denoising as an equilibrium problem. Rather than relying on iterative sampling, DriftSE natively achieves one-step inference by evolving the pushforward distribution of a mapping function to directly match the clean speech distribution. This evolution is driven by a Drifting Field, a learned correction vector that guides samples toward the high-density regions of the clean distribution, which naturally facilitates training on unpaired data by matching distributions rather than paired samples. We investigate the framework under two formulations: a direct mapping from the noisy observation, and a stochastic conditional generative model from a Gaussian prior. Experiments on the VoiceBank-DEMAND benchmark demonstrate that DriftSE achieves high-fidelity enhancement in a single step, outperforming multi-step diffusion baselines and establishing a new paradigm for speech enhancement.