Speech endpoint detection in real noise environments
-
Graphical Abstract
-
Abstract
A method of speech endpoint detection in environments of complicated additive noise is presented. Based on the analysis of noise, an adaptive model of stable noise is proposed to detect the section where the signal is unstable. Then the voice is detected in this section by its harmonic structure, and the accurate endpoint is searched using energy. Compared with the typical algorithms, this algorithm operates reliably in most real noise environments.
-
-