Speech endpoint detection in low-SNRs environment based on perception spectrogram structure boundary parameter

A Perception Spectrogram Structure Boundary (PSSB) parameter is proposed for speech endpoint detection as a preprocess of speech signal. A hearing perception speech enhancement is made as a first step, then a two-dimensional enhancement is performed upon the speech spectrogram according to the difference between the continuous distribution characteristic of pure speech and the random distribution characteristic of noise, in order to emphasize the continuous spectrogram structure of pure speech. PSSB parameter is proposed based on the two-dimensional boundary detection of the enhanced speech spectrogram structure. Experimental results show that, in a variety of SNR environments from -10 dB to 10 dB, the algorithm proposed in this paper can achieve higher accuracy in comparison to the extant endpoint detection algorithms. With our algorithm, accuracy of 75.2% can be reached even in the extreme low SNR at -10 dB. The endpoint detection algorithm using PSSB, is suitable for speech endpoint detection in low-SNRs environment with white noise.