Abstract: In view of the problem that conventional spectral entropy speech endpoint detection algorithm’s detection effect is poor under the non-stationary noise,a new feature parameter－sub-band amplitude spectrum entropy was proposed.The new parameter detection of speech endpoint uses non-stationary signal processing technology to combine the signal of time domain and frequency domain characteristics.Firstly,the conventional spectral entropy speech endpoint detection algorithm is improved and the multi-band spectral entropy is calculated,then the endpoint is detected with the combination of short time average magnitude.The simulation results show that this method has better robustness and precision than conventional spectral entropy algorithm and average magnitude algorithm,which proves the effectiveness of the proposed method.

Key words: Endpoint detection,Spectral entropy,Short-time average magnitude,Robustness