An Endpoint Detection of Complex Environment Base on Embedded System

For speech recognition systems in embedded environment,endpoint detection is extremely important since location of speech endpoint works on recognition rate directly.However,great differences of input speech signal are shown while low accuracy of endpoint detection in complexity environment.In this paper,an area probability statistical method based on the energy-zero-product is presented.So the conceptual distribution of the speech energy-zero-product is obtained through training,with which first order linear difference method and spectral subtraction are used in endpoint detection.Simulation experiment results illustrate that this method raises SNR further as well as arrives at the solution of low accuracy of endpoint detection in complexity environment.