Speech endpoint detection in noisy environment using Spectrogram Boundary Factor
Di Wu, Zhi Tao, Yuanbo Wu, Cheng Shen, Zhongzhe Xiao, Xiaojun Zhang, Di Wu, Heming Zhao · 2016
An endpoint detection in low signal-to-noise ratio (SNR) environment plays an important role in speech processing. In this paper, we propose an endpoint detection algorithm applying a novel feature parameter, called Spectrogram Boundary Factor (SBF), to improve the endpoint detection performance in noisy environments. In the first step, the time-frequency spectrogram is obtained from the noisy speech. Then we use the erosion algorithm and the dilation algorithm to attenuate the noise and enhance the speech banded structure on the spectrogram respectively. Finally, we calculate the SBF feature parameter after edge detection using the improved spectrogram first-order derivative. The experiments prove that using the proposed algorithm, the speech endpoint can be effectively detected even in a case of low SNR of 0 dB.