ADMM-based audio reconstruction for low-cost-sound-monitoring

For low-cost sound monitoring of machineries, we propose a novel audio reconstruction method superior in terms of accuracy and processing time. A conventional method based on the Orthogonal Matching Pursuit (OMP) has been proposed for audio recovery. However, the conventional method has a low performance for sounds of machineries because, sounds of machineries tend to be not highly sparse, and the reconstruction performance of OMP decreases extremely if the signal is not sufficiently sparse. To solve the problem of the conventional method, the proposed method is based on the Alternating Direction Method of Multipliers (ADMM) for Group Lasso combined with the Gabor dictionary. While OMP's performance decreases with the number of nonzero elements, the proposed method shows a better robustness to variations in sparsity and outputs a reasonable result in a few tens of iterations. Those features among others make the algorithm a reliable solution which offers a better trade-off between accuracy and processing time compared to the conventional method.