Prosodic Fillers and Discourse Markers–Discourse Prosody and Text Prediction

Mandarin Chinese fluent speech prosody is characterized by a hierarchical multiple-phrase structure that specifies how speech paragraphs are constituted via Prosodic Phrase Grouping. Hence we view spoken discourse prosody as yet another higher node treats PGs (Prosodic Phrase Groups) as sister constituents. The goals of present study are two fold: one is to study how speech paragraphs are connected in speech flow; another is to derive prosody prediction from text analysis. Investigating cross-phrase F0 range narrowing and F0 reset with boundary information, we further conducted corresponding text analysis for prosody prediction. Results revealed two types of PG connectors, one is redundant Prosodic Fillers (PF) that are mostly duration triggered and manifested through narrowed F0 ranges; another is obligatory Discourse Markers (DM) that are lexically and/or syntactically triggered and manifested through widened F0 ranges and resets. Both could be predicted from text analysis. We believe this is a significant step forward towards understanding the organization of discourse prosody. It could also be applied to speech synthesis and/or unlimited TTS for prosody enhancement.