A Lightweight Speech Enhancement Model with Parallel Processing of Magnitude and Phase Spectra
Wenyi Chen · 2025
Recent advancements in deep learning have led to substantial improvements in speech enhancement over traditional methods. However, these models typically involve a large number of parameters and require considerable computational resources, which limits their applicability on edge devices for real-world use. In this paper, we propose MPCRN, a novel model that simultaneously addresses both magnitude and phase denoising by incorporating GTCRN. MPCRN is optimized for computational efficiency, containing only 1.31 million parameters. Experimental evaluations show that MPCRN exhibits comparable performance to the current leading model, demonstrating strong competitiveness. A comparison of MPCRN with MP-SENet reveals that the former has an advantage in terms of memory utilisation efficiency and the effective reduction of training time.