METHOD OF SELECTING SPEECH PROCESSING SEGMENTS Russian patent published in 2025 - IPC G10L15/05 G10L17/02 

FIELD: computer engineering.

SUBSTANCE: invention relates to computer engineering for processing audio data. Result is achieved due to speech processing segments extraction method, based on input speech signal reception, analogue-to-digital conversion and speech signal fragment extraction, wherein the value of dispersion of the noise of the speech signal fragment of the given duration is measured, using which the value of the threshold and the optimum accumulation interval are calculated, on which the dispersion of the input signal is determined, comparing the dispersion with the calculated threshold, if in the interval equal to half the accumulation interval, the dispersion value is below the threshold, then a decision is made on the classification of this interval as a pause, otherwise, as speech, discrete samples containing pauses are deleted, and the remaining samples are combined and a speech section without pauses is obtained.

EFFECT: high probability of correct segmentation of a speech signal in conditions of noise.

1 cl, 1 dwg