A bayesian adaptive approach for voiced-unvoiced classification of segments of a speech signal