In the paper a new approach for estimating of the spoken language sound multivariate probability density is suggested. It is based on the use of a projection of a random process to the set of random variables, with the probability density defined as a product of two-dimensional densities. The estimates of two-dimensional probability densities are obtained with the help of filtering of the two-dimensional empirical characteristic function. Therefore, we are suggesting a nonparametric estimate of the characteristic function. On the basis of these estimates nonparametric algorithms of sound classification are constructed.
展开▼
