Kronecker Product Based Linear Prediction Kalman Filter for Dereverberation and Noise Reduction

Integrated sidelobe cancellation and linear prediction (ISCLP) method can effectively reduce reverberation, noise and interference, but the amount of calculation is huge, and speech distortion is often introduced as well. In order to solve these problems, a low-complexity dereverberation algorithm is proposed in this paper. By introducing Kronecker product, the multichannel linear prediction coefficients are expressed as two low-dimensional sub-filter coefficients vectors, which are from spatial filter and inter-frame filter. In the case of introducing spatial filter, inter-frame filter is optimized by the Kalman filter. Compared with the ISCLP, the proposed algorithm can effectively increase calculation efficiency, because the dimension numbers of covariance matrix and vector required for calculation are lower, meanwhile in most conditions, the speech quality is also improved.