Savitzky-Golay Filtering and Improved Energy Entropy for Speech Endpoint Detection under Low SNR

Abstract

Under the condition of low signal-to-noise ratio (SNR) or non-stationary noise, the performance of endpoint detection is poor. Therefore, this paper proposes a speech endpoint detection algorithm for low SNR. In this paper, Savitzky-Golay filtering, improved sub-band energy entropy and constant Q-transform (CQT) are used to extract features, and single parameter double threshold method is used to realize endpoint detection. In this paper, clean speech and noise-92 speech fragments are used to evaluate the algorithm. Experimental results show that the algorithm in this paper can distinguish speech endpoints well. For Gaussian white noise, Factory noise and Volvo noise, the detection accuracy of the improved algorithm is improved by 5.5%, 5.3% and 4.6% respectively. Therefore, the algorithm in this paper can separate the mute and voice in complex environment.