Endpoint Detection Based on EMD in Noisy Environment

This paper proposed a novel Hilbert-Huang transform (HHT) based endpoint detection method. Firstly, IMF1 and IMF2 are discarded after EMD, and then some IMFs components that can be used in endpoint detection are selected to reconstruct the speech signal by comparing means and variances. The reconstructed signal is used for detection via two-threshold method. The experiences show that the proposed method remains high detection rate of 84.2% compared with two-threshold method. When there are noises existing in speech signal, the proposed method still can implement detecting with detection rate being 85.9% while traditional two-threshold method is invalid. All these demonstrate that the new approach has robustness in some degree.