On-line dereverberation algorithm based on weighted prediction error for noisy time-varying environment

摘要

A system and method for processing a multi-channel audio signal comprises receiving a multi-channel time-domain audio input, converting the input signal to a plurality of multi-channel frequency-domain k-space undersampled subband signals; Buffering and delaying, storing a subset of the spectral frames for prediction filter estimation in each of the spectral frames, estimating the variance of the frequency domain signal in each of the spectral frames, recursive least squares ( (RLS) algorithm to adaptively estimate the prediction filter in an online manner, linearly filter each channel using the estimated prediction filter, and nonlinearly filter the linear filtered output signal. Reverberation and estimated variance Includes reducing, and generating a non-linear filtering output signals, to reconstruct the dereverberation time domain multi-channel audio signal by combining a nonlinear filtering output signal. [Selection] Figure 2