FIELD: physics.
SUBSTANCE: speech processing method includes the steps of: determining the presence/absence parameter of the vocalization in the current frame of the speech signal, which is a combined parameter reflecting the product of the periodicity parameter and the spectral slope parameter; determining the smoothed presence/absence parameter of the vocalization to include information of the presence/absence parameter of the vocalization in the preceding frame; calculating the difference between the presence/absence parameter of the vocalization in the current frame and the smoothed parameter, and determining, whether the current frame contains unvoiced speech or voiced speech, using the calculated difference as the decision parameter.
EFFECT: improved and more reliable detection of unvoiced, voiced speech.
20 cl, 15 dwg
