Method and apparatus for equalizing filtering a speech signal using a statistical signal model