Abstract: It is an important step of speech recognition process to identify accurately the speech endpoint.Under the environment with low SNR,in order to enhance the discrimination of noise better and improve the accuracy of speech endpoint detection system,this paper proposesd a new type of speech endpoint detection algorithm based on sub-band energy-entropy-ratio.The proposed algorithm takes the ratio of short-time sub-band energy and sub-band spectral entropy as an important parameter of endpoint detection,and sets the threshold to speech endpoint detection.Experiments show that the algorithm is fast and efficient,and it also has strong robustness and can detect the voice endpoint under lower SNR accurately.

Key words: Endpoint detection,Sub-band energy,Sub-band spectral entropy,Sub-band energy-entropy-ratio,SNR,Robustness