Description:
<jats:p>This paper presents a prosodic analysis of storytelling speech in Japanese fairy tale. Recent advances in TTS technologies bring us high quality synthetic speech especially on acoustic aspects. On prosodic aspects, however, there is still room for improvement in the expressiveness since most systems use one-sentence speech synthesis scheme. For instance, storytelling applications expect speech synthesis to be capable of having a control mechanism beyond one sentence. In this paper, a prosodic database is built for real storytelling speech of professional narrators on Japanese fairy tale aiming at investigating the actual storytelling strategies and ultimately reflecting them on the expressiveness of speech synthesis. After conducting a baseline speech segmentation for phones, words, and accentual phrases in a semi-automatic way, a multi-layered prosodic tagging is manually performed to extract information on various changes of “story states” relevant to impersonation, emotional involvement, and scene flow control. Storytelling speech materials of six professional narrators for Japanese fairy tale are prepared to analyze their storytelling strategies. In particular, the dynamics of pitch contours is investigated in terms of expressiveness control beyond sentence considering narrator dependency as well. Based on the findings obtained, pitch control schemes are discussed for storytelling speech synthesis.</jats:p>