Initial conditions in the problem of voice recognition
V. N. Sorokin, A. A. Tananykin · Journal of Communications Technology and Electronics · 2010
In the problem of the voice identification, the residual signal that is calculated using the linear prediction is approximated by a mathematical model of the source using the least squares procedure. The approximated results depend on the initial conditions, which must be additionally determined. The problem is solved with the aid of clusterization with respect to multiple data obtained for arbitrary initial conditions. The Euclidian metric appears to be the best one. The centers of clusters for men and women are significantly different. We fail to determine the explicit dependence of the centers of clusters on the fundamental tone frequency.