SPEECH SYNTHESIS DEVICE, METHOD, AND PROGRAM

摘要

PROBLEM TO BE SOLVED: To enable speech fragments to be correctly selected from a speech corpus even when, pertaining to a technique for synthesizing a speech by selecting speech fragments from the speech corpus, rhythm features in a rhythm dictionary for extracting a rhythm are different between input text data and fragment data in the speech corpus.SOLUTION: Owing to that each of a target rhythm pitch feature point extraction unit 203 and a fragment rhythm pitch feature point extraction unit 204 calculates the histogram of, for example, a target rhythm and an intra-speech corpus fragment rhythm pitch frequency, each of these units extracts a prescribed number, e.g., four, of pitch feature points. A rhythm correction unit 205 corrects target rhythm data 201 on the basis of correlation between pitch feature points of the target rhythm (feature points 1-4) and pitch feature points of the fragment rhythm (feature points from 1 to 4), and outputs corrected target rhythm data 206.