Preprocessing method and device for speech recognition device

A preprocessing method for a speech recognition device having input means for receiving a speech signal which has been transmitted through a transmission system having a certain transmission band, preprocessing means including an inverse filter for preprocessing the input speech signal and recognition means for recognizing the input speech signal from said preprocessing means;

• comprising the first step of Fourier-transforming the speech signal from said input means to obtain a frequency spectrum, the second step of extracting only a frequency spectrum relating to the frequency spectrum of said transmission band from said frequency spectrum obtained at said first step and continuously rearranging the extracted frequency spectrum from 0 Hz, and the third step of subjecting to inverse Fourier transform the signal corresponding to the rearranged frequency spectrum extracted at said second step, the fourth step of analyzing a signal obtained at said third step to obtain a specified characteristic parameter and the fifth step of setting a characteristic inverse to said characteristic parameter obtained by said fourth step in said inverse filter and passing the input speech signal through said inverse filter, thereby obtaining the input speech signal for said recognition means.