Adaptive probabilistic process for speech segmentation

A classical and as yet unsolved problem in speech research is that of segmenting running speech, either as an end in itself, or as a part of a segmentation and labeling task. This paper reports an adaptive process using a modification of the Viterbi algorithm, in which a simple Markov model is applied to running speech. This method is flexible as well as being computationally tractible. The results of the automatic process are compared with segmentation derived from careful manual marking.