Dual-channel speech enhancement combining blind source separation and lightweight model

The practical deployment of deep learning-based speech enhancement methods is usually constrained by limited computational resources, and the performance significantly deteriorates under low signal-to-noise ratio conditions. This paper proposes lightweight hybrid dual-channel speech enhancement approaches that integrate blind source separation (BSS) algorithms with lightweight speech enhancement models. The outputs of BSS algorithms are used as auxiliary information for the speech enhancement model, while the model further refines the speech quality. Experimental results demonstrate that the proposed approaches can achieve high-quality speech enhancement with minimal model parameters and low computational complexity.