De-reverberation of reverberant speech or reproducing dry speech in a reverberant space is an important spatial-audio technolory as well as rendering 3D reverberation from a dry audio signal.The key issue is inverse filtering of the transfer function(TF)in a reverberant space.Inverse filtering,however,is formidable in a reverberant space,since the TF can often be of the non-minimum-phase.This paper investingates the inverse filtering of the TF including non-minimum-phase components using both the least square error(LSE)solution and the multiple-input/output inverse-filtering theorem(MINT)for reproducing intelligible speech in a reverberant space.The results demonstrate that the original sound waveform from the observed signal in the reverberant space can be completely reproduced by using MINT.However,the good de-reverberation of reverberant speech can't be gotten from the LSE method.This deference can be explained by the zeros and poles distribution in z-plane of the recovery TF.