Noise Reduction through Sound Source Reconstruction by Combining Blind Signal Separation and Audio-Language Model

Abstract

In this paper, we propose a novel noise reduction method for acoustic inspections. Existing acoustic noise reduction techniques require prior collection of target sounds or assumptions about the direction of arrival, which limits their applicability. To eliminate these constraints, we introduce a noise reduction approach based on source reconstruction that integrates blind signal separation with audio-language models. Experiments conducted for acoustic inspections demonstrated effective noise reduction across a wide range of SNR conditions.