Speech endpoint detection with low SNR based on HHTSM

Speech endpoint detection is difficult in low SNR environment. The article presents a detection method based on Hilbert-Huang Transform Spectral Matrix (HHTSM) for improving speech endpoint detection accuracy in low signal-to-noise ratios (SNR). The method analyses signal time-frequency-energy distribution with Hilbert-Huang Transform (HHT), and constructs HHTSM by frames. The frame size is chosen by energy focus character. The speech and noise energy distribution can be distinguished by HHTSM in low SNR. By estimating noise energy distribution, it sets threshold automatically. The result indicates that the method can detect speech endpoint effectively in the negative SNR. So the method based on HHTSM is effective.