An efficient and robust speech dereverberation method using spherical microphone array

In this paper, we present a robust adaptive reverberation cancellation method based on multi-channel linear prediction in the spherical harmonic domain. In practice, the spherical harmonic coefficients with different orders may have significantly different signal-to-noise ratios due to the white noise gain differences at different orders, which can influence the performance of dereverberation especially when using high-order spherical harmonic coefficients. To solve this problem, we propose a dereverberation method that processes spherical harmonic coefficients with the same order and compare the performance to the traditional method. Simulation results demonstrate that the proposed method can reduce up to 2/3 computational complexity, as well as significantly improve the PESQ scores and the SRMR scores.