Segmentation of Words into Phones

Abstract

Word segmentation into phones is studied in this paper. The method of change point detection in random sequences is used for phone boundaries detection in a word. It is assumed that phones are stationary signal segments and changes of signal parameters are the boundaries of these phones. Change moments are detected maximizing change points likelihood function or minimizing prediction least mean squared error. On the ground of formulated methodology word segmentation software was developed. Dynamic programming was used for object function optimization and for unknown parameters estimation generalized expectation maximization algorithm was used. Developed software allows to control speech signal processing parameters, to choose segmentation optimality criterion. Detected segments can be labeled, and these segments labels and boundaries can be saved into files for later segmentation results preview. Ill. 2, bibl. 13 (in English; summaries in Lithuanian, English and Russian).