摘要
This study presents evidence from analyses of the acoustic parameters of fluent continuous speech to show that within-paragraph prosodic phrase boundaries are related more to contrasts of neighborhood prosodic states rather than between-phrase pause durations; prosodic states receive more constraints from higher level discourse information. By revising a modular acoustic model by Tseng＇s hierarchical prosodic phrase grouping framework and examining the much varied prosodic phrase （PPh） boundary B3 within speech paragraph, we show that statistical accounts of layered contributions reveal distinct contrasts between boundary immediate duration and intensity patterns irrespective of pause duration. Contrasts of FO contour patterns were also observed in these locations. Evidence was also obtained to illustrate how PPh boundary states are specified more by higher level discourse information than by lower level prosodic word construction. These combined results suggest that contrastive neighboring prosodic states are more significant cues to PPh boundaries than boundary pause duration. The results also help explain why in fluent speech between-phrase pause durations vary greatly, and can be applied to automatic speech segmentation.
This study presents evidence from analyses of the acoustic parameters of fluent continuous speech to show that within-paragraph prosodic phrase boundaries are related more to contrasts of neighborhood prosodic states rather than between-phrase pause durations; prosodic states receive more constraints from higher level discourse information. By revising a modular acoustic model by Tseng＇s hierarchical prosodic phrase grouping framework and examining the much varied prosodic phrase （PPh） boundary B3 within speech paragraph, we show that statistical accounts of layered contributions reveal distinct contrasts between boundary immediate duration and intensity patterns irrespective of pause duration. Contrasts of FO contour patterns were also observed in these locations. Evidence was also obtained to illustrate how PPh boundary states are specified more by higher level discourse information than by lower level prosodic word construction. These combined results suggest that contrastive neighboring prosodic states are more significant cues to PPh boundaries than boundary pause duration. The results also help explain why in fluent speech between-phrase pause durations vary greatly, and can be applied to automatic speech segmentation.
