Improved Spectral Subtraction Based on Second-Order Differential Array and Phase Spectrum Compensation

Traditional spectral subtraction speech enhancement algorithm is inaccurate in noise estimation in complex environments, and often employ noisy phase while reconstructing the enhanced signal. In this paper, an improved spectral subtraction algorithm based on second-order differential microphone array and phase spectrum compensation is proposed. The method uses second-order differential microphone array beamforming to extract the speech signal and noise signal from the collected array speech and process them accordingly. If the signal-to-noise ratio of the current frame is greater than the preset threshold, the power spectrum subtraction is performed to reduce the impact of noise; conversely, the noise is attenuated directly. Then, phase spectrum compensation is used to obtain more accurate phase information for speech reconstruction. Experimental results demonstrate that the proposed algorithm effectively suppresses various types of noise and outperforms the comparison algorithms in terms of denoising performance and speech intelligibility.