SARC: Structured Analytic-Residual Correction for Lightweight Multichannel Speech Enhancement
Authors: Gurui Jin, Xiao-Ping Zhang, Le Xiao, Yue Wang, Zhenyu Liu
SARC combines analytic spatial filtering with neural correction for lightweight multichannel speech enhancement. A learned channel-weighting matrix forms a unit-response estimate from RTF-aligned microphone signals. Target-canceling spatial residuals guide a compact correction network in suppressing the remaining interference.
Status: Submitted to ICASSP 2027.