METHOD FOR EXTRACTING VOICE CHARACTERISTIC SUITABLE FOR CORE WORD DETECTION IN NOISE CIRCUMSTANCE

摘要

Purpose: a kind of method for extracting voice characteristics is arranged to the core word being applicable in noise circumstance detection. Construction: in the method for extracting voice characteristics, if sound is entered (ST1), the spectrum slope for inputting sound is flattened, and after the working range for inhibiting signal, executes the preemphasis that (ST2) is used to increase a signal-to-noise ratio. One window function is multiplied by the signal (ST3) of pre-emphasis, and then a synthetic video is obtained by making increased signal pass through a filter. The analysis of lid flow point is to cover the Local Square Integral Martingale compensation for flowing (ST7) since the distortion of a noise passes through to obtain a Local Square Integral Martingale from synthetic video (ST5).