Исследование алгоритма сегментации речевого сигнала методом динамического программирования

Abstract

Speech signal segmentation algorithm which uses previously known information about signal content was explored. Speech synthesizer was used to produce segmented signal that contains same speech as the signal we need to process, and compare this synthesized signal with the natural one using Dynamic Time Warping. This allows us to move segment markers from synthesized signal to natural signal. To compare signals we use feature vectors consisting of spectrum we got by using Bark scale based set of filters, pitch frequency and their finite differences.