Voice activity detector in the input signal with a noise component consisting of stored noise estimate (113) of input signal, devices (114, 115) to detect spectral similarity of the input signal with the stored noise model to create output decision signal, from devices (112) for actualisation of the stored noise model and of the additive detector (200) to control devices (112) for actualisation to actualise only in the time when the additive detector (200) indicates that the input signal does not contain a speech. Voice activity detector contains also devices (303) to count the parameter of predetermined acquisition of the input signal and modificatory devices (304) to repress actualisation in the case that predetermined acquisition exceeds threshold value.