This study addresses a novel compressive coding for multichannel audio signals with independent component analysis (ICA). In recent years, multichannel audio contents are widely spreading, and thus the transmission quantity is increasing. Consequently, audio‐signal compression is a problem demanding prompt attention. Conventional compression methods utilizing sparseness assumption among audio‐source signals have a problem that the method is likely to distort original sources owing to the lack of sparseness in the mixed signals consisting of two or more sound sources. To solve this problem, we propose an audio‐signal compression strategy using ICA; ICA assumes that the source signals are mutually independent, and can estimate the source signals from the mixed signals without a priori information by applying an unmixing matrix. The decomposition of mixed signals leads to an enhancement of sparseness among signals. Furthermore, we propose a mechanism to detect time‐frequency grids where sparseness is not satisfied, and for the detected grids we directly transmit multichannel signals to mitigate the distortion. The reconstruction process utilizes an inverse projection of the unmixing matrix to make correct lateralization of each sound source. Experimental result shows that the proposed method can compress mixed audio signals more efficiently and achieve a higher quality than the conventional method.