Prosody control of utterance sequence for information delivering

We propose a conversational speech synthesis system in which the prosodic features of each utterance are controlled throughout the entire input text. We have developed a "news-telling system," which delivered news articles through spoken language. The speech synthesis system for the news-telling should be able to highlight utterances containing noteworthy information in the article with a particular way of speaking so as to impress them on the users. To achieve this, we introduced role and position features of the individual utterances in the article into the control parameters for prosody generation throughout the text. We defined three categories for the role feature: a nucleus (which is assigned to the utterance including the noteworthy information), a front satellite (which precedes the nucleus) and a rear satellite (which follows the nucleus). We investigated how the prosodic features differed depending on the role and position features through an analysis of news-telling speech data uttered by a voice actress. We designed the speech synthesis system on the basis of a deep neural network having the role and position features added to its input layer. Objective and subjective evaluation results showed that introducing those features was effective in the speech synthesis for the information delivering.