Speech Endpoint Detection Algorithm Based on EMD in Highly Noisy Background
SHEN Tao;FENG Gang;Computer Science College,South China Normal University;
The speech endpoint detection is a vital process in speech recognition systems,and plays a crucial role to the accuracy of the system. It is discussed that the deficiencies of the existing speech endpoint detection algorithms in highly noisy background. And through the introduction of HHT,a method based on Hilbert Huang Transform endpoint detection is presented,which uses EMD to decompose finite IMF,does the IMF Hilbert transform and gets the IMF energy spectrum as the characteristic parameters of the speech signal endpoint detection. Finally the emulation experiment proves the effectiveness and robustness of the algorithm in highly noisy background.