VOICE DURATION DETECTION METHOD FOR VOICE RECOGNIZING SYSTEM

摘要

PURPOSE: A method of detecting a voice section of a voice recognition system is provided to adaptively set a threshold value according to the size of a peripheral noise energy, so as to regularly detect a voice regardless of a peripheral noise and to remove the peripheral noise in the voice section by using energy distribution characteristics. CONSTITUTION: A voice recognition system puts a window on inputted voice signals to divide the signals by frame, and calculates energy of each frame(S1-S3). When setting a threshold value of a voice energy on the basis of the first frames, the voice recognition system sets the threshold value with an 'average value + changed value according to the average value', and decides a section of a presently-inputted signal as a voice signal section based on the set threshold value, or decides the section as a section of a peripheral noise signal(S4-S7).