Personalization of News Speech Delivery Service Based on Transformation from Written Language to Spoken Language

Abstract

This paper proposes a method of automatic transformation of newspaper articles for generating spontaneous news speech in a news speech delivery service. Among several differences between written and spoken languages in Japanese, this study takes particular note of the importance of the differences in sentence-style, and also works on taigen-dome sentence, which is a Japanese-specific rhetoric style to cut off the words following a certain noun. The sentence-style transformation from written language to spoken language is a kind of personalization of a textto- speech application. We took an example-based approach to complement taigendome sentences. An experiment using Japanese newspaper articles has shown the achievement of the acceptable performance on the selection of the complementary phrase for taigen-dome sentences by using the type of the last noun of the sentence, the tense, and so on.