BROAD PHONETIC CLASSIFICATION AND SEGMENTATION OF CONTINUOUS SPEECH BY MEANS OF NEURAL NETWORKS AND DYNAMIC-PROGRAMMING

We describe the broad phonetic classification and segmentation of continuous speech. We have trained a multilayered neural network to produce broad phonetic class hypotheses which were used as inputs to a multi-level segmentation algorithm based on dynamic programming. Evaluation of our algorithm on a multi-speaker database of continuously spoken number strings has demonstrated excellent segmentation and classification performances. © 1991.