The research on speech endpoint detection algorithm based on spectrogram row self-correlation
Jiawei Fu, Sh. W. Wang, Xiaolin Cao, Mingzhe Jiang, Sinuo Zhang, Xin Tong Zhao · 2012
In this paper, a novel endpoint detection algorithm based on spectrogram row self-correlation is proposed. Initially, the original speech signals are changed into speech spectrogram. In every spectrogram image the self-correlation of each row data is calculated. Therefore, in the coordinate of self-correlation curve the distances of adjacent extremes are chosen as characters used in speech endpoint detection. Compared to 20–30ms set as a basic process unit in the traditional algorithm, 100ms even much longer speech spectrogram image is adopted in the new method. It contributes to extract characters of signals integrally and prompt the speed of speech endpoints detection. The new algorithm was tried and tested in the research of a vehicle speech recognition system. Some practical results gotten in experiments are given in this paper.