Endpoint Detection Of Speech Signal Based On Empirical Mode Decomposition

Speech endpoint detection is the front of speech recognition system, its refers to accurately detect the effective speech signal start and end point from the noise environment, for follow-up processes improve accuracy and save time. Although in a quiet laboratory environment, the accuracy of speech endpoint detection can be satisfactory.But the performance of the detection sharply degrades in the practical different kinds of noise environments. Therefore, in actual different background noise and low SNR conditions of speech endpoint detection technology research has the extremely vital significance.Therefore, speech endpoint detection technology research has the extremely vital significance in actual different background noise and low SNR conditions.The speech signal is non-stationary signals. The empirical mode decomposition (EMD) is a new method for adaptive analysis of non-linear and non-stationary signals. The empirical mode decomposition (EMD) is suited to be used in the nonlinear and non-stationary signal processing and it offers an approach to describe the signal instantaneous characters. This paper applies this method to speech endpoint detection.Firstly, with the analysis of some speech features existing typical algorithms of endpoint detection. Then, two new algorithms of the speech endpoint detection are proposed. The first one:endpoint detection of speech signal based on empirical mode decomposition and energy. Simulation results have shown the performance of the typical algorithm is better than others. The second one: endpoint detection of speech signal based on empirical mode decomposition and teager kurtosis. Simulation results have shown the performance of the typical algorithm is better than others, too.