To discriminate the type of tones, such as voices, music and noise by an acoustic signal discriminating method capable of efficiently discriminating the type of the acoustic signals to be discriminated with good accuracy, an acoustic signal discriminating system, and an acoustic signal discriminating program.

The method comprises the steps of: obtaining the frequency spectra of a signal to be discriminated; obtaining the correlation coefficient of the spectra adjacent in the time base direction of the frequency spectra; extracting the first amount of feature for discriminating stationary and non-stationary characteristics by the correlation value thereof; determining a correlation coefficient by shifting the frequency spectra obtained at different points of time in the axial direction of the frequency; extracting a pitch component as the second amount of feature from the correlation coefficient; further extracting the third amount of feature to serve as a measure of noise characteristics by extracting the average value within a certain frequency range of the correlation coefficient; extracting the power of the spectra as the fifth amount of feature in addition thereto; determining the likelihood by each kind of tones by these amounts of feature and previously recorded learning data; and determining the type of the tone exhibiting the maximum of the likelihood as the type of the tone of the acoustic signal to be discriminated.