Analysis of Communicative Phrase Prosody based on Linguistic Modalities of Constituent Words

Abstract

In this paper, phrase prosody is analyzed based on linguistic modalities of constituent words for communicative speech synthesis. Since Japanese final particles and auxiliaries play crucial roles to indicate speaker's intention and attitudes as modality differences, Japanese phrase sets showing different degree of the speaker's judgment were employed. Communicative/reading speech data were compared over 5 kinds of modality of epistemic judgment (uncertainty of what the speaker said) and 8 kinds of modality of evaluative judgment (what the speaker wishes listeners to be). These modality differences were quantified in 6-point Semantic Differential (SD) scales. The corresponding phrase communicative/reading prosody differences were measured by the F0 rising in the phrase final mora. Statistical analysis showed negative correlation value between F0 rising in the phrase final mora and SD about judgment only in communicative prosody but not in reading prosody. These results support the communicative prosody control possibilities from the modality information embedded in constituent words.