FIELD: speech recognition.
SUBSTANCE: in accordance to the invention, digitization is performed with signal step Δt, a flow of counts of signal of a certain length is recorded in memorizing device from microphone output in narrator silence mode, divided on a row of R long sections, range (1/Δt) of Fourier energy spectrum of each one of these sections is divided onto a row of frequency intervals (i=1,...,N), values Pi pause are computed, corresponding to each one of the intervals, computed average values Pipause av are recorded into memorizing device, in case of incoming speech signal, values Pi are computed, ratios Pi to Pi pause av are computed in each one of N frequency intervals of all R long sections and max(Pi/Pi pause av) are determined, which are compared to threshold value h, determined in threshold computation circuit and at max(Pi/Pi pause av)≤h decision is made about presence of pause, then value Pi pause av is renewed with consideration of current value of Pi pause.
EFFECT: reduced volume of memory for storing speech and reduced traffic during transmission thereof.
2 cl, 5 dwg
