Wind Noise Reduction with Orthogonal Polynomial Expansion

This paper is dedicated to mitigating unwanted ambient wind noise in outdoor multi-channel recordings. Different from conventional methodologies such as the multi-channel Wiener filter (MWF) and its extensions, we propose a wind noise reduction method from a beamforming perspective based on the spatial characteristics of wind noise captured by closely spaced microphones. Specifically, the noisy observation is decomposed into a parameterization space concerning frequency through orthogonal polynomial expansion. Subsequently, the estimated target source signals are reconstructed by selecting coefficients corresponding to desired components within the polynomial basis. Moreover, we introduce the Alternating Direction Method of Multipliers (ADMM) framework to attain a sparse solution for the coefficients during the decomposition phase. In comparison to existing multi-channel wind noise reduction methods, simulation results demonstrate the superior performance of the proposed method, particularly in low signal-to-noise ratio (SNR) scenarios.