Effective jointly pdf-based voice activity detector for real-time applications
J. M. Górriz, Javier Ramı́rez, Carlos G. Puntonet · Electronics Letters · 2007
Presented is an accurate speech detection algorithm based on a jointly Gaussian probability distribution function (jGpdf) and a likelihood ratio test evaluated on a new observation model for the input speech. A more realistic description of the dependence between observations in a classical statistical binary framework (H0&H1) provides higher discrimination in speech endpoint detection and reports benefits over a number of recently published methods.