Additive and convolutional noise canceling in speaker verification using a stochastic weighted viterbi algorithm