METHOD FOR RECOGNIZING VOICE USING FILTERING OF FEATURE VECTOR

摘要

PURPOSE: A method for recognizing a voice using filtering of a feature vector is provided to improve the capability of voice recognition by technologically realizing masking phenomena observed in the auditory organ of a human being. CONSTITUTION: Inputted aural signals are divided into predetermined time sections to be divided into frames(S110). The aural signals are multiplied by a hamming window(S120). The aural signals are transformed to obtain frequency information of the signals of the frames through a fourier transform(S130). Mutual inhibition between signals is modeled by using the obtained frequency information of the aural signals(S135). The frequency information of the aural signals are put together into a few of frequency band information(S140). Unilateral inhibition is modeled to output of a filter bank(S145). A log is taken to the output of the filter bank and a feature vector is transformed into cepstrum information(S150).