The paper presents a system which reduces reverberative and stationary noise in a real room. A corrupted speech signal is analyzed by an audio transform, i.e. the spectral transformation and their characteristic parameters are adapted to the human auditory system. The concept of this transformation is described in detail. The transformed speech signals are filtered with the inverse modulation transfer function of the room causing the reverberation. This function has to be determined only once and is constant for all positions in the room. After filtering a sinusoidal representation of the input signal is extracted. The dereverberated speech is resynthesized on the basis of this part-tone time pattern.