The effect of inverse filter length on the robustness of acoustic multichannel equalization

In acoustic multichannel equalization techniques, generally the length of the inverse filters is chosen such that exact inverse filters can be designed for perfectly estimated room impulse responses (RIRs). However, since in practice the estimated RIRs typically differ from the real RIRs, this choice of the inverse filter length may not be optimal. This paper investigates the effect of the inverse filter length on the robustness of acoustic multichannel equalization. A mathematical link between the inverse filter length and robustness to channel estimation errors is provided. Experimental results for speech dereverberation show that using shorter inverse filters yields similar or higher performance for all considered multichannel equalization techniques, while reducing the computational complexity of the inverse filter design.